Negotiation scenario generation method, negotiation strategy generation method, negotiation scenario generation apparatus, negotiation strategy generation apparatus, negotiation scenario generation program, and negotiation strategy generation program

Abstract

A negotiation scenario generation method comprising: obtaining one or more performance features for a set of negotiation strategies; projecting the obtained performance features into one or more intrinsic features; and generating one or more first negotiation scenarios with reference to the result of the projection.

Description

NEGOTIATION SCENARIO GENERATION METHOD, NEGOTIATION STRATEGY GENERATION METHOD, NEGOTIATION SCENARIO GENERATION APPARATUS, NEGOTIATION STRATEGY GENERATION APPARATUS, NEGOTIATION SCENARIO GENERATION PROGRAM, AND NEGOTIATION STRATEGY GENERATION PROGRAM

[0001] The present application relates to a negotiation scenario generation method, a negotiation strategy generation method, a negotiation scenario generation apparatus, a negotiation strategy generation apparatus, a negotiation scenario generation program, and a negotiation strategy generation program.

[0002] Automated negotiation is becoming more important in various applications, for example, in business operations. Various techniques for automated negotiation have been developed. For example, Patent Literature 1 (PL1) discloses a method and system for performing negotiation task using reinforcement learning agent.

[0003] PL 1

[0004] Japanese Patent Application Publication Tokukai No. 2020-013568

[0005] In automated negotiation or other machine learning techniques, it is important to prepare an appropriate set of negotiation contexts (scenarios) and / or negotiation strategies for training. However, in the existing methods such as the technique disclosed in PL1, it is difficult to prepare an appropriate set of negotiation scenarios and / or negotiation strategies for training.

[0006] An example aspect of the present invention is attained in view of the problem, and an example object is to provide a technique for generating an appropriate set of negotiation scenarios and / or negotiation strategies.

[0007] Solution to Problem In order to attain the object described above, a negotiation scenario generation method comprising: obtaining one or more performance features for a set of negotiation strategies; projecting the obtained performance features into one or more intrinsic features; and generating one or more first negotiation scenarios with reference to the result of the projection.

[0008] In order to attain the object described above, a negotiation strategy generation method comprising: obtaining one or more intrinsic features for a set of negotiation scenarios, obtaining one or more performance features for the set of negotiation scenarios; clustering the negotiation scenarios with respect to the intrinsic features and the performance features; and generating one or more negotiation strategies with reference to the result of the clustering.

[0009] In order to attain the object described above, a negotiation strategy generation method comprising: obtaining one or more performance features for a set of negotiation scenarios; obtaining one or more negotiation strategies; generating one or more negotiation scenarios with low performance with respect to all the negotiation strategies; and generating one or more negotiation strategies with reference to the one or more generated negotiation scenarios.

[0010] In order to attain the object described above, a negotiation scenario generation apparatus comprising: an obtaining means to obtain one or more performance features for a set of negotiation strategies; a projection means to project the obtained performance features into one or more intrinsic features; and a generation means to generate one or more negotiation scenarios with reference to the result of the projection.

[0011] In order to attain the object described above, a negotiation strategy generation apparatus comprising: a first obtaining means to obtain one or more intrinsic features for a set of negotiation scenarios; a second obtaining means to obtain one or more performance features for the set of negotiation scenarios; a clustering means to cluster the negotiation scenarios with respect to the intrinsic features and the performance features; and a generation means to generate one or more negotiation strategies with reference to the result of the clustering.

[0012] In order to attain the object described above, a negotiation strategy generation apparatus comprising: a first obtaining means to obtain one or more performance features for a set of negotiation scenarios; a second obtaining means to obtain one or more negotiation strategies; a first generation means to generate one or more negotiation scenarios with low performance with respect to all the negotiation strategies; and a second generation means to generate one or more negotiation strategies with reference to the one or more generated negotiation scenarios.

[0013] In order to attain the object described above, a negotiation scenario generation program causing a computer to function as: an obtaining means to obtain one or more performance features for a set of negotiation strategies; a projection means to project the obtained performance features into one or more intrinsic features; and a generation means to generate one or more negotiation scenarios with reference to the result of the projection.

[0014] In order to attain the object described above, a negotiation strategy generation program causing a computer to function as: a first obtaining means to obtain one or more intrinsic features for a set of negotiation scenarios; a second obtaining means to obtain one or more performance features for the set of negotiation scenarios; a clustering means to cluster the negotiation scenarios with respect to the intrinsic features and the performance features; and a generation means to generate one or more negotiation strategies with reference to the result of the clustering.

[0015] In order to attain the object described above, a negotiation strategy generation program causing a computer to function as: a first obtaining means to obtain one or more performance features for a set of negotiation scenarios; a second obtaining means to obtain one or more negotiation strategies; a first generation means to generate one or more negotiation scenarios with low performance with respect to all the negotiation strategies; and a second generation means to generate one or more negotiation strategies with reference to the one or more generated negotiation scenarios.

[0016] According to an example aspect of the present invention, it is possible to generate an appropriate set of negotiation scenarios and / or negotiation strategies.

[0017] Fig. 1 is a flowchart showing a flow of a negotiation scenario generation method according to an example embodiment.Fig. 2 is a block diagram illustrating a configuration of a negotiation scenario generation apparatus according to an example embodiment.Fig. 3 is a flowchart showing a flow of a negotiation strategy generation method according to an example embodiment.Fig. 4 is a block diagram illustrating a configuration of a negotiation strategy generation apparatus according to an example embodiment.Fig. 5 is a flowchart showing a flow of a negotiation strategy generation method according to an example embodiment.Fig. 6 is a block diagram illustrating a configuration of a negotiation strategy generation apparatus according to an example embodiment.Fig. 7 is a block diagram illustrating a configuration of an information processing system according to an example embodiment.Fig. 8 is a schematic illustration of data flow according to an example embodiment.Fig. 9 is a schematic illustration of data flow in an initial dataset generation according to an example embodiment.Fig. 10 is a schematic illustration showing exemplary results of the initial dataset generation according to an example embodiment.Fig. 11 is a schematic illustration of data flow in a scenario set enhancement according to an example embodiment.Fig. 12 is a schematic illustration showing exemplary results of the scenario set enhancement according to an example embodiment.Fig. 13 is a schematic illustration of data flow in a strategy set enhancement according to an example embodiment.Fig. 14 is a schematic illustration showing exemplary results of the strategy set enhancement according to an example embodiment.Fig. 15 is a schematic illustration of data flow in a guided scenario and strategy generation according to an example embodiment.Fig. 16 is a schematic illustration showing exemplary results of the guided scenario and strategy generation according to an example embodiment.Fig. 17 is a schematic illustration of data flow in a final context generation according to an example embodiment.Fig. 18 is a schematic illustration showing exemplary results of the final context generation according to an example embodiment.Fig. 19 is a block diagram illustrating a configuration of an information processing system according to an example embodiment.Fig. 20 is a schematic illustration of data flow in training phase, testing phase, and negotiation phase according to an example embodiment.Fig. 21 is a diagram illustrating a specific example of the initial dataset generation according to an example embodiment.Fig. 22 is a diagram illustrating a specific example of the scenario set enhancement according to an example embodiment.Fig. 23 is a diagram illustrating a specific example of the strategy set enhancement according to an example embodiment.Fig. 24 is a diagram illustrating a specific example of the guided strategy and scenario enhancement according to an example embodiment.Fig. 25 is a diagram illustrating a specific example of the guided strategy and scenario enhancement according to an example embodiment.Fig. 26 is a diagram illustrating a specific example of the final context generation according to an example embodiment.Fig. 27 is a diagram illustrating a specific example of a deployment according to an example embodiment.Fig. 28 is a block diagram illustrating a hardware configuration according to an example embodiments.

[0018] The following description will discuss example embodiments of the present invention. However, the present invention is not limited to the example embodiments described below, but can be altered by a skilled person in the art within the scope of the claims. For example, any embodiment derived by appropriately combining technical means adopted in differing example embodiments described below can be within the scope of the present invention. Further, any embodiment derived by appropriately omitting one or more of the technical means adopted in differing example embodiments described below can be within the scope of the present invention. Furthermore, the advantage mentioned in each of the example embodiments described below is an example advantage expected in that example embodiment, and does not define the extension of the present invention. That is, any embodiment which does not provide the example advantages mentioned in the example embodiments described below can also be within the scope of the present invention.

[0019] <First example embodiment> The following description will discuss a first example embodiment, which is an example embodiment of the present invention, in detail with reference to the drawings. The present example embodiment is basic to each of the example embodiments which will be described later. It should be noted that the applicability of each of the technical means adopted in the present example embodiment is not limited to the present example embodiment. That is, each technical means adopted in the present example embodiment can be adopted in another example embodiment included in the present disclosure, to the extent of constituting no specific technical obstacle. Further, each technical means illustrated in the drawings referred to for the description of the present example embodiment can be adopted in another example embodiment included in the present disclosure, to the extent of constituting no specific technical obstacle.

[0020] (Negotiation Scenario Generation Method S1) The following description will discuss a negotiation scenario generation method S1 according to the first example embodiment with reference to Fig. 1. Fig. 1 is a flowchart showing a flow of the negotiation scenario generation method S1. The negotiation scenario generation method S1 can be carried out, for example, by a negotiation scenario generation apparatus 1 explained below. However, this does not limit the first example embodiment.

[0021] (Step S11) In step S11, one or more performance features for a set of negotiation strategies are obtained. For example, in step S11, an obtaining section 11 of the negotiation scenario generation apparatus 1 may obtain the one or more performance features for a set of negotiation strategies. Here, the performance feature is, for example, a profits (a score) of each strategy in a given negotiation scenario. For example, if we have three negotiation strategies STA, STB, and STC in a given negotiation scenario, corresponding performance features PFA, PFB, and PFC are obtained in step S11. Note that the performance features may also be referred to as performance feature values.

[0022] (Step S12) In step S12, the performance features obtained in Step S11 are projected into one or more intrinsic features. In other words, one or more intrinsic features of one or more negotiation scenarios are generated with reference to the performance features obtained in Step S11. For example, in step S12, a projection section 12 of the negotiation scenario generation apparatus 1 may project the performance features obtained in Step S11 into one or more intrinsic features of one or more negotiation scenarios. Here, the intrinsic feature is, for example, an observable scenario-specific feature. In an example, the intrinsic feature is an observable feature that may affect a negotiation opponent's behavior.

[0023] In an example, the projection from the performance features to the intrinsic features may be carried out by a machine-learned projection model (also referred to as PI projector). However, this does not limit the present example embodiment.

[0024] (Step S13) In step S13, one or more negotiation scenarios are generated with reference to the result of the projection carried out in step S12. For example, in step S12, a generation section 13 of the negotiation scenario generation apparatus 1 may generate the one or more negotiation scenarios with reference to the result of the projection carried out in step S12.

[0025] (Advantageous effect) According to the first example embodiment, the negotiation scenario generation method S1 includes: obtaining one or more performance features for a set of negotiation strategies (step S11), projecting the obtained performance features into one or more intrinsic features (step S12), and generating one or more negotiation scenarios with reference to the result of the projection (step S13).

[0026] As mentioned above, in the negotiation scenario generation method S1, the performance features for a set of negotiation strategies are projected into one or more intrinsic features, and then one or more negotiation scenarios are generated reference to the result of the projection. Therefore, the negotiation scenario generation method S1 can generate one or more negotiation scenarios appropriate for the given set of negotiation strategies.

[0027] (Negotiation Scenario Generation Apparatus 1) The following description will discuss a negotiation scenario generation apparatus 1 according to the first example embodiment with reference to Fig. 2. Fig. 2 is a block diagram illustrating a configuration of the negotiation scenario generation apparatus 1. As illustrated in Fig. 2, the negotiation scenarios generation apparatus 1 includes an obtaining section 11, a projection section 12, and a generation section 13.

[0028] (Obtaining section 11) The obtaining section 11 obtains one or more performance features for a set of negotiation strategies. Here, as mentioned above, the performance feature is, for example, a profits (a score) of each strategy in a given negotiation scenario. For example, if we have three negotiation strategies STA, STB, and STC in a given negotiation scenario, the obtaining section 11 obtains the corresponding performance features PFA, PFB, and PFC.

[0029] (Projection section 12) The projection section 12 projects the performance features obtained by the obtaining section 11 into one or more intrinsic features. In other words, the projection section 12 generates one or more intrinsic features of one or more negotiation scenarios with reference to the performance features obtained by the obtaining section 11. Here, as mentioned above, the intrinsic feature is, for example, an observable scenario-specific feature. In an example, the intrinsic feature is an observable feature that may affect a negotiation opponent's behavior.

[0030] In an example, the projection from the performance features to the intrinsic features may be carried out by a machine-learned projection model (also referred to as PI projector). However, this does not limit the present example embodiment.

[0031] (Generation section 13) The generation section 13 generate one or more negotiation scenarios with reference to the result of the projection carried out by the projection section 12.

[0032] (Advantageous effect) According to the first example embodiment, the negotiation scenario generation apparatus 1 includes: an obtaining section 11 to obtain one or more performance features for a set of negotiation strategies, a projection section 12 to project the obtained performance features into one or more intrinsic features, and a generation section 13 to generate one or more negotiation scenarios with reference to the result of the projection.

[0033] As mentioned above, according to the negotiation scenario generation apparatus 1, the performance features for a set of negotiation strategies are projected into one or more intrinsic features, and then one or more negotiation scenarios are generated reference to the result of the projection. Therefore, the negotiation scenario generation apparatus 1 can generate one or more negotiation scenarios appropriate for the given set of negotiation strategies.

[0034] (Negotiation Strategy Generation Method S2) The following description will discuss a negotiation strategy generation method S2 according to the first example embodiment with reference to Fig. 3. Fig. 3 is a flowchart showing a flow of the negotiation strategy generation method S2. The negotiation generation strategy method S2 can be carried out, for example, by a negotiation strategy generation apparatus 2 explained below. However, this does not limit the first example embodiment.

[0035] (Step S21) In step S21, one or more intrinsic features for a set of negotiation scenarios are obtained. For example, in step S21, a first obtaining section 21 of the negotiation strategy generation apparatus 2 may obtain the one or more intrinsic features for a set of negotiation scenarios. Here, as mentioned above, the intrinsic feature is, for example, an observable scenario-specific feature. For example, if we have two negotiation scenarios SCA, and SCB, a set of intrinsic features (IFA1, IFA2, IFA3) associated with the negotiation scenarios SCA, and a set of intrinsic features (IFB1, IFB2, IFB3) associated with the negotiation scenarios SCB are obtained.

[0036] (Step S22) In step S22, one or more performance features for the set of negotiation scenarios are obtained. For example, in step S22, a second obtaining section 22 of the negotiation strategy generation apparatus 2 may obtain the one or more performance features for the set of negotiation scenarios. Here, as mentioned above, the performance feature is, for example, a profits (a score) of each strategy in a given negotiation scenario. For example, if we have three negotiation strategies STA, STB, and STC and two negotiation scenarios SC1 and SC2, then performance features PFA1, PFB1, and PFC1 for the negotiation scenarios SC1, and performance features PFA2, PFB2, and PFC2 for the negotiation scenarios SC2 are obtained. Here, the performance features PFA1, PFB1, and PFC1 are the performance features of the negotiation strategies STA, STB, and STC, respectively, in the negotiation scenario SC1. Similarly, the performance features PFA2, PFB2, and PFC2 are the performance features of the negotiation strategies STA, STB, and STC, respectively, in the negotiation scenario SC2.

[0037] (Step S23) In step S23, the negotiation scenarios with respect to the intrinsic features and the performance features are clustered. For example, in step S23, a clustering section 23 of the negotiation strategy generation apparatus 2 may cluster the negotiation scenarios with respect to the intrinsic features and the performance features. In an example, the clustering of the negotiation scenarios with respect to the intrinsic features are carried out, and the clustering of the negotiation scenarios with respect to the performance features are carried out. In another example, the clustering of the negotiation scenarios with respect to both of the intrinsic features and the performance features are carried out.

[0038] (Step S24) In step S24, one or more negotiation strategies are generated with reference to the result of the clustering carried out in step S23. For example, in step S24, a generation section 24 of the negotiation strategy generation apparatus 2 may generate one or more negotiation strategies with reference to the result of the clustering carried out in step S23. In an example, generation of the one or more negotiation strategies in step 24 is carried out so as to reduce a mismatch between one or more intrinsic features and one or more performance features.

[0039] (Advantageous effect) According to the first example embodiment, the negotiation strategy generation method S2 includes: obtaining one or more intrinsic features for a set of negotiation scenarios (step S21), obtaining one or more performance features for the set of negotiation scenarios (step S22), clustering the negotiation scenarios with respect to the intrinsic features and the performance features (step S23), and generating one or more negotiation strategies with reference to the result of the clustering (step S24). As mentioned above, in the negotiation strategy generation method S2, the negotiation scenarios with respect to the intrinsic features and the performance features are clustered, and then one or more negotiation strategies are generated with reference to the result of the clustering. Therefore, the negotiation strategy generation method S2 can generate one or more negotiation strategy appropriate for the given set of negotiation scenarios.

[0040] (Negotiation Strategy Generation Apparatus 2) The following description will discuss a negotiation strategy generation apparatus 2 according to the first example embodiment with reference to Fig. 4. Fig. 4 is a block diagram illustrating a configuration of the negotiation strategy generation apparatus 2. As illustrated in Fig. 4, the negotiation strategy generation apparatus 2 includes a first obtaining section 21, a second obtaining section 22, a clustering section 23, and a generation section 24.

[0041] (First obtaining section 21) The first obtaining section 21 obtains one or more intrinsic features for a set of negotiation scenarios. Here, as mentioned above, the intrinsic feature is, for example, an observable scenario-specific feature. For example, if we have two negotiation scenarios SCA, and SCB, the first obtaining section 21 obtains a set of intrinsic features (IFA1, IFA2, IFA3) associated with the negotiation scenarios SCA, and a set of intrinsic features (IFB1, IFB2, IFB3) associated with the negotiation scenarios SCB.

[0042] (Second obtaining section 22) The second obtaining section 22 obtains one or more performance features for the set of negotiation scenarios. Here, as mentioned above, the performance feature is, for example, a profits (a score) of each strategy in a given negotiation scenario. For example, if we have three negotiation strategies STA, STB, and STC and two negotiation scenarios SC1 and SC2, then the second obtaining section 22 obtains performance features PFA1, PFB1, and PFC1 for the negotiation scenarios SC1, and performance features PFA2, PFB2, and PFC2 for the negotiation scenarios SC2. Here, the performance features PFA1, PFB1, and PFC1 are the performance features of the negotiation strategies STA, STB, and STC, respectively, in the negotiation scenario SC1. Similarly, the performance features PFA2, PFB2, and PFC2 are the performance features of the negotiation strategies STA, STB, and STC, respectively, in the negotiation scenario SC2.

[0043] (Clustering section 23) The clustering section 23 clusters the negotiation scenarios with respect to the intrinsic features and the performance features. In an example, the clustering of the negotiation scenarios with respect to the intrinsic features are carried out, and the clustering of the negotiation scenarios with respect to the performance features are carried out. In another example, the clustering of the negotiation scenarios with respect to both of the intrinsic features and the performance features are carried out.

[0044] (Generation section 24) The generation section 24 generates one or more negotiation strategies with reference to the result of the clustering carried out by the clustering section 23. In an example, generation of the one or more negotiation strategies is carried out so as to reduce a mismatch between one or more intrinsic features and one or more performance features.

[0045] (Advantageous effect) According to the first example embodiment, the negotiation strategy generation apparatus 2 includes: a first obtaining section 21 to obtain one or more intrinsic features for a set of negotiation scenarios, a second obtaining section 22 to obtain one or more performance features for the set of negotiation scenarios, a clustering section 23 to cluster the negotiation scenarios with respect to the intrinsic features and the performance features, and a generation section 24 to generate one or more negotiation strategies with reference to the result of the clustering. As mentioned above, according to the negotiation strategy generation apparatus 2, the negotiation scenarios with respect to the intrinsic features and the performance features are clustered, and then one or more negotiation strategies are generated with reference to the result of the clustering. Therefore, the negotiation strategy generation apparatus 2 can generate one or more negotiation strategy appropriate for the given set of negotiation scenarios.

[0046] (Negotiation Strategy Generation Method S3) The following description will discuss a negotiation strategy generation method S3 according to the first example embodiment with reference to Fig. 5. Fig. 5 is a flowchart showing a flow of the negotiation strategy generation method S3. The negotiation generation strategy method S3 can be carried out, for example, by a negotiation strategy generation apparatus 3 explained below. However, this does not limit the first example embodiment.

[0047] (Step S31) In step S31, one or more performance features for a set of negotiation scenarios are obtained. For example, in step S31, a first obtaining section 31 of the negotiation strategy generation apparatus 3 may obtain the one or more performance features for a set of negotiation scenarios. Here, as mentioned above, the performance feature is, for example, a profits (a score) of each strategy in a given negotiation scenario. For example, if we have three negotiation strategies STA, STB, and STC and two negotiation scenarios SC1 and SC2, then performance features PFA1, PFB1, and PFC1 for the negotiation scenarios SC1, and performance features PFA2, PFB2, and PFC2 for the negotiation scenarios SC2 are obtained. Here, the performance features PFA1, PFB1, and PFC1 are the performance features of the negotiation strategies STA, STB, and STC, respectively, in the negotiation scenario SC1. Similarly, the performance features PFA2, PFB2, and PFC2 are the performance features of the negotiation strategies STA, STB, and STC, respectively, in the negotiation scenario SC2.

[0048] (Step S32) In step S32, one or more negotiation strategies are obtained. For example, in step S32, a second obtaining section 32 of the negotiation strategy generation apparatus 3 may obtain the one or more negotiation strategies. For example, the negotiation strategies STA, STB, and STC, are obtained. Each of the negotiation strategies includes one or more pieces of information to define the strategy. Therefore, in step S32, the one or more pieces of information to define each of the one or more negotiation strategy are obtained. The negotiation strategies obtained in step S32 may include, for example, time-based strategies. However, this does not limit the first example embodiment.

[0049] (Step S33) In step S33, one or more negotiation scenarios with low performance with respect to all the negotiation strategies are generated. For example, in step S33, a first generation section 33 of the negotiation strategy generation apparatus 3 may generate the one or more negotiation scenarios with low performance with respect to all the negotiation strategies. In an example, the first generation section 33 may carry out iterative steps of generating a negotiation scenario, calculating the performance features for each of the negotiation strategies in the generated negotiation scenario, and determining whether all of the calculated performance features are below the predetermined threshold.

[0050] (Step S34) In step S34, one or more negotiation strategies are generated with reference to the one or more negotiation scenarios generated in step S33. For example, in step S34, a second generation section 34 of the negotiation strategy generation apparatus 3 may generate the one or more negotiation strategies with reference to the one or more negotiation scenarios generated in step S33.

[0051] (Advantageous effect) According to the first example embodiment, the negotiation strategy generation method S3 includes: obtaining one or more performance features for a set of negotiation scenarios (step S31), obtaining one or more negotiation strategies (step S32), generating one or more negotiation scenarios with low performance with respect to all the negotiation strategies (step S33), and generating one or more negotiation strategies with reference to the one or more generated negotiation scenarios (step S34).

[0052] As mentioned above, in the negotiation strategy generation method S3, the one or more negotiation scenarios with low performance with respect to all the negotiation strategies are generated, and then the one or more negotiation strategies are generated with reference to the one or more generated negotiation scenarios. Therefore, the negotiation strategy generation method S3 can generate one or more negotiation strategy appropriate for the given set of negotiation scenarios.

[0053] (Negotiation Strategy Generation Apparatus 3) The following description will discuss a negotiation strategy generation apparatus 3 according to the first example embodiment with reference to Fig. 6. Fig. 6 is a block diagram illustrating a configuration of the negotiation strategy generation apparatus 3. As illustrated in Fig. 6, the negotiation strategy generation apparatus 3 includes a first obtaining section 31, a second obtaining section 32, a first generation section 33, and a second generation section 34.

[0054] (First obtaining section 31) The first obtaining section 31 obtains one or more performance features for a set of negotiation scenarios. For example, if we have three negotiation strategies STA, STB, and STC and two negotiation scenarios SC1 and SC2, then the first obtaining section 31 obtains performance features PFA1, PFB1, and PFC1 for the negotiation scenarios SC1, and performance features PFA2, PFB2, and PFC2 for the negotiation scenarios SC2. Here, the performance features PFA1, PFB1, and PFC1 are the performance features of the negotiation strategies STA, STB, and STC, respectively, in the negotiation scenario SC1. Similarly, the performance features PFA2, PFB2, and PFC2 are the performance features of the negotiation strategies STA, STB, and STC, respectively, in the negotiation scenario SC2.

[0055] (Second obtaining section 32) The second obtaining section 32 obtains one or more negotiation strategies. For example, the second obtaining section 32 obtains the negotiation strategies STA, STB, and STC. Each of the negotiation strategies includes one or more pieces of information to define the strategy. Therefore, the second obtaining section 32 obtains the one or more pieces of information to define each of the one or more negotiation strategy. The negotiation strategies obtained by the second obtaining section 32 may include, for example, time-based strategies. However, this does not limit the first example embodiment.

[0056] (First generation section 33) The first generation section 33 generates one or more negotiation scenarios with low performance with respect to all the negotiation strategies. In an example, the first generation section 33 may carry out iterative steps of generating a negotiation scenario, calculating the performance features for each of the negotiation strategies in the generated negotiation scenario, and determining whether all of the calculated performance features are below the predetermined threshold.

[0057] (Second generation section 34) The second generation section 34 generate one or more negotiation strategies with reference to the one or more negotiation scenarios generated by the first generation section 33.

[0058] (Advantageous effect) According to the first example embodiment, the negotiation strategy generation apparatus 3 includes: a first obtaining section 31 to obtain one or more performance features for a set of negotiation scenarios, a second obtaining section 32 to obtain one or more negotiation strategies, a first generation section 33 to generate one or more negotiation scenarios with low performance with respect to all the negotiation strategies, and a second generation section 34 to generate one or more negotiation strategies with reference to the one or more generated negotiation scenarios.

[0059] As mentioned above, according to the negotiation strategy generation apparatus 3, the one or more negotiation scenarios with low performance with respect to all the negotiation strategies are generated, and then the one or more negotiation strategies are generated with reference to the one or more generated negotiation scenarios. Therefore, the negotiation strategy generation apparatus 3 can generate one or more negotiation strategy appropriate for the given set of negotiation scenarios.

[0060] <Second example embodiment> The following description will discuss details of a second example embodiment of the invention with reference to the drawings. Note that the same reference numerals are given to elements having the same functions as those described in the first example embodiment, and descriptions of such elements are omitted as appropriate.

[0061] (Configuration of information processing apparatus 1A) The following description will discuss a configuration of an information processing apparatus 1A according to the second example embodiment with reference to Fig. 7. Fig. 7 is a block diagram illustrating a configuration of the information processing apparatus 1A. As illustrated in Fig. 7, the information processing apparatus 1A includes a control section 10A, a storage section 20A, a communication section 30, and an input-output section 40. The information processing apparatus 1A may be referred to as a negotiation apparatus, negotiation scenario generation apparatus, or a negotiation strategy generation apparatus.

[0062] (Communication section 30) The communication section 30 communicates with an apparatus external to the information processing apparatus 1A. The communication section 30 transmits, to the external apparatus, data supplied from the control section 10A, and / or supplies, to the control section 10A, data received from the external apparatus. (Input-output section 40) The input-output section 40 includes, for example, an input-output device which is at least any one selected from the group consisting of a keyboard, a mouse, a display, a printer, a touch panel, a camera, a speaker, and the like. The input-output section 40 may include, instead of any of these input-output devices, an interface such as universal serial bus (USB), for example. Further, the interface may be connected to at least any one of the input-output devices.

[0063] The input-output section 40 accepts, from at least any one of the input-output devices, various information for the information processing apparatus 1A. Further, under control of the control section 10A, the input-output section 40 outputs various information to at least any one of the input-output devices.

[0064] For example, the input-output section 40 may accept, from a user, one or more parameters which specify processes carried out in the control section 10A. For another example, the section 40 may output visualized information relating to the processes carried out by the information processing apparatus 1A.

[0065] (Storage section 20A) The storage section 20A stores therein various information referred to by the control section 10A and various information derived by the control section 10A. In an example, as illustrated in Fig. 4, the storage section 20A stores therein the followings: - a set of scenarios SC_SET - performance features PF - a set of strategy ST_SET - a PI projector PIP - a IP projector IPP - a context classifier CC.

[0066] (Set of scenarios SC_SET) Here, the set of scenarios SC_SET includes one or more negotiation scenarios SC. Each of the negotiation scenarios SC may be specified by a scenario ID, and one or more pieces of information defining the scenario. The pieces of information defining the scenario may be referred to as intrinsic features IF. If the scenario is related to procurement, such pieces of information (intrinsic features IF) may include, for example, the number of supplier, the number of competitors, and production cost at each of the competitors. In another example, the intrinsic features IF may include a degree of demand and a degree of competition. Alternatively, as shown in Fig. 7, it may be expressed that a set of intrinsic features IF are associated with each of the scenarios SC. It may also be expressed that the intrinsic feature IF is, for example, an observable scenario-specific feature. In an example, the intrinsic feature is an observable feature that may affect a negotiation opponent's behavior. The intrinsic features IF may also be referred as intrinsic feature values.

[0067] (Set of strategies ST_SET) The set of strategy ST_SET includes one or more negotiation strategies ST. Each of the negotiation strategies ST may be specified by a strategy ID, and one or more pieces of information defining the strategy. Each of the strategies ST may be configured such that if the strategy ST is applied to a given scenario SC, a profits (a score) may be calculated. In other words, if pieces of information defining the given scenario SC are input to the strategy ST, the strategy ST provides a profits (a score) associated with the given scenario SC. In an example, such pieces of information defining the strategy may be one or more parameters. In an example, such parameters may be targets of training. Note that the set of strategy ST_SET may include, for example, time-based strategies. However, this does not limit the first example embodiment.

[0068] (Performance features PF) Each of the performance features PF is, for example, a profits (a score) provided by each strategy ST in a given negotiation scenario SC. For example, if we have three negotiation strategies STA, STB, and STC and two negotiation scenarios SC1 and SC2, then the performance features PFA1, PFB1, and PFC1 for the negotiation scenarios SC1, and performance features PFA2, PFB2, and PFC2 for the negotiation scenarios SC2 may be calculated. Here, the performance features PFA1, PFB1, and PFC1 are the performance features of the negotiation strategies STA, STB, and STC, respectively, in the negotiation scenario SC1. Similarly, the performance features PFA2, PFB2, and PFC2 are the performance features of the negotiation strategies STA, STB, and STC, respectively, in the negotiation scenario SC2.

[0069] (PI projector PIP) The PI projector PIP is a projector which project one or more performance features PF into one or more intrinsic features IF. The PI projector PIP may be expressed as a generator to generate one or more intrinsic features IF with reference to one or more performance features PF. The PI projector PIP may be realized as a machine-learned projection model, or more specifically as a machine-learned neural network. However, this does not limit the present example embodiment.

[0070] (IP projector IPP) The IP projector PIP is a projector which project one or more intrinsic features IF into one or more performance features PF. The IP projector IPP may be expressed as a generator to generate one or more performance features PF with reference to one or more intrinsic features IP. The IP projector IPP may be realized as a machine-learned projection model, or more specifically as a machine-learned neural network. However, this does not limit the present example embodiment.

[0071] (Context classifier CC) The context classifier CC is a classifier to classify the contexts. Here, each of contexts may include one or more negotiation scenarios SC. The context classifier CC may be realized as a machine-learned neural network model which maps one or more intrinsic features IF to one or more contexts. The context classifier CC may also be expressed as a feature classifier.

[0072] (Control section 10A) As illustrated in Fig. 7, the control section 10A includes an obtaining section 11, a projection section 12, a clustering section 23, and a generation section 13. The obtaining section 11 may function as the obtaining section 11, the first obtaining section 21, the second obtaining section 22, the first obtaining section 31, and the second obtaining section 32 explained in the first example embodiment, the obtaining section 11 may also be referred to as the obtaining section 11 (22, 22, 31, 32).

[0073] The generation section 13 may function as the generation section 13, the generation section 24, the first generation section 33, and the second generation section 34 explained in the first example embodiment, the generation section 13 may also be referred to as the generation section 13 (24, 33, 34). As shown in Fig. 7, the generation section 13 includes a training section 131.

[0074] (Obtaining section 11) The obtaining section 11 obtains one or more performance features PF for a set of negotiation strategies ST. The obtaining section 11 may also obtain one or more intrinsic features IF for a set of negotiation scenarios SC. The obtaining section 11 may also obtain one or more negotiation strategies ST. In an example, the obtaining section 11 may obtain the above explained information from an apparatus external to the information processing apparatus 1A via the communication section 30. In another example, the obtaining section 11 may obtain the above explained information from the storage section 20A. A specific example of processes carried out by the obtaining section 11 will be described later.

[0075] (Projection section 12) The projection section 12 projects one or more performance features PF into one or more intrinsic features IF by using the PI projector PIP. The projection section 12 may also project one or more intrinsic features IF into one or more performance features PF by using the IP projector IPP. A specific example of processes carried out by the projection section 12 will be described later.

[0076] (Clustering section 23) The clustering section 23 clusters the negotiation scenarios SC with respect to the intrinsic features IF and the performance features PF. In an example, the clustering section 23 clusters the negotiation scenarios SC with respect to the intrinsic features IF. In another example, the clustering section 23 clusters the negotiation scenarios SC with respect to the performance features PF. In yet another example, the clustering section 23 clusters the negotiation scenarios SC with respect to both of the intrinsic features IF and the performance features PF. A specific example of processes carried out by the clustering section 23 will be described later.

[0077] (Generation section 13) The generation section 13 generate one or more negotiation scenarios SC with reference to the result of the projection carried out by the projection section 12. The generation section 13 may also generate one or more negotiation strategies ST with reference to the result of the clustering carried out by the clustering section 23. The generation section 13 may also function as a first generation section to generate one or more negotiation scenarios SC with low performance with respect to all the negotiation strategies ST. The generation section 13 may also function as a second generation section to generate one or more negotiation strategies ST with reference to the one or more negotiation scenarios generated by the first generation section.

[0078] As shown in Fig. 7, the generation section 13 includes the training section 131. The training section 131 may trains the one or more negotiation strategies ST for each of clusters generated by the clustering section 23. In other words, the training section 131 may update the one or more parameters defining the negotiation strategy ST for each of clusters generated by the clustering section 23. It may be expressed that generation of the one or more negotiation strategies by the generation section 13 is carried out so as to reduce a mismatch between one or more intrinsic features IF and one or more performance features PF.

[0079] The training section 131 may also train the IP projector IPP to project one or more intrinsic features IP to one or more performance features PF. The training section 131 may also determine whether an accuracy of the IP projector IPP reaches a predefined threshold. The training section 131 may also train the PI projector PIP to project one or more performance features PF to one or more intrinsic features IF. The training section 131 may also determine whether an accuracy of the PI projector PIP reaches a predefined threshold or the maximum value. A specific example of processes carried out by the generation section 13 and the training section 131 will be described later.

[0080] (First Overview of processes in the information processing apparatus 1A) The following description will discuss an overview of processes carried out in the information processing apparatus 1A with reference to Fig. 8. Fig. 8 is a schematic illustration of data flow in the information processing apparatus 1A according to the second example embodiment.

[0081] (Step S10A) In step S10A, processes of an initial dataset generation is carried out. More specifically, the obtaining section 11 obtains one or more negotiation scenarios SC from the scenario set SC_SET, and one or more negotiation strategies ST from the strategy set ST_SET. Then, the generation section calculates the performance features PF with reference to the negotiation scenarios SC and the negotiation strategies ST. The initial scores shown in Fig. 8 are examples of the performance features PF calculated in step S10A.

[0082] (Step S1A) Then, in step S1A, processes of a scenario set enhancement is carried out. More specifically, the obtaining section 11 obtains the performance features PF (initial scores) calculated in step S10A. Then, the projection section 12 projects the performance features PF obtained by the obtaining section 11 into one or more intrinsic features IF. The generation section 13 generate one or more new negotiation scenarios SC with reference to the result of the projection carried out by the projection section 12.

[0083] (Step S2A) Then, in step S2A, processes of a strategy set enhancement is carried out. More specifically, the obtaining section 11 obtains the new negotiation scenarios SC generated in step S1A. For example, the obtaining section 11 obtains one or more intrinsic features IF for the new negotiation scenarios SC. The obtaining section 11 also obtains one or more performance features PF for the new negotiation scenarios SC. Then, the clustering section 23 clusters the new negotiation scenarios SC with respect to the intrinsic features IF and the performance features PF. Then, the generation section 24 generates one or more new negotiation strategies ST with reference to the result of the clustering carried out by the clustering section 23. In an example, generation of the one or more negotiation strategies is carried out so as to reduce a mismatch between one or more intrinsic features IF and one or more performance features PF.

[0084] (Step S3A) Then, in step S3A, processes of a guided scenario and strategy generation is carried out. More specifically, the obtaining section 11 obtains the new negotiation strategies ST generated in step S2A. For example, the obtaining section 11 obtains one or more performance features PF for the new negotiation scenarios SC generated in step S1A or updated in step S2A. The obtaining section 11 also obtains the new negotiation strategies ST generated in step S2A. Then, the first generation section 13 function as a first generation section to generate one or more negotiation scenarios SC with low performance with respect to all the negotiation strategies ST. Then, the generation section 13 function as a second generation section to generate one or more negotiation strategies ST with reference to the one or more negotiation scenarios SC generated by the first generation section.

[0085] (Step S20A) In step S20A, processes of a final context generation is carried out. More specifically, the obtaining section 11 obtains the new negotiation scenarios SC generated in step S1A, or updated in step S2A or S3A. Then, the clustering section 23 clusters the new negotiation scenarios SC with respect to the performance features PF and the intrinsic features IF. Then, the generation section 13 generates the contexts and the context classifier CC. Here, one context generated by the generation section 13 may include one or more negotiation scenarios SC belonging to the same cluster in the feature space. Each of the contexts generated by the generation section 13 in step S20A is associated with the corresponding negotiation strategy ST also generated by the generation section 13.

[0086] (Second Overview of processes in the information processing apparatus 1A) The processes carried out in the information processing apparatus 1A may be summarized as follows.

[0087] (1) Initial Datasets Generation: Bootstrap Intrinsic and Performance Features (1-1) Run S simulations with existing strategies, (1-2) Record intrinsic scenario features and performance features.

[0088] (2) Scenario Set Enhancement: Better Scenario Coverage (2-1) Project on performance features (p-space) (2-2) Project on intrinsic features (i-space) (2-3) Find unvisited locations in p-space and simulate existing strategies utilizing a weak Performance->Intrinsic Predictor, (2-1) to (2-3) may be iterated.

[0089] (3) Strategy Set Enhancement: Intrinsic-Performance Matching (3-1) Cluster on ALL features (optionally low weight on performance) (3-2) Train a strategy for each cluster and record its performance (3-3) Optionally augment small clusters with new scenarios (3-4) Train an Intrinsic -> Performance predictor (IP Projector) (3-1) to (3-4) may be iterated and stopped on good enough prediction accuracy.

[0090] (4) Guided Scenario and Strategy Generation: Better Strategies and Contexts (4-1) Find scenarios with low performance for all strategies (4-2) Cluster new scenarios on ALL features (4-3) Train a strategy for each cluster, update performance features (4-1) to (4-3) may be iterated. (2) to (4) may be iterated.

[0091] (5) Final Context Generation (5-1) Cluster on performance -> Contexts (5-2) Train per context -> Strategies (5-3) Train Intrinsic -> Context predictor -> Classifier.

[0092] (Initial Dataset Generation) The following description will discuss the initial dataset generation carried out in the information processing apparatus 1A with reference to Fig. 9 and Fig. 10. Fig. 9 is a schematic illustration of data flow in the initial dataset generation according to the second example embodiment. A set of processes in Fig. 9 may correspond to a specific example of step S10A in Fig. 8. It may be expressed that the initial dataset generation is a process to generating data for training the agent, Inputs to the initial dataset generation may be records of negotiation scenarios and opponent strategies. Outputs of the initial dataset generation may be features of each negotiation session and agent scores for each of them.

[0093] (Step S101A, S102A) In step S101A, the generation section 13 generates a scenario s. In other words, the generation section 13 generates intrinsic features IF defining the scenario s. Then, in step S102A, the generation section 13 run a simulation with reference to the strategy set ST_SET and the scenario s generated in step S101A. As a result of the simulation, the generation section 13 obtains performance features PF of the scenario s with respect to each of the strategies included in the strategy set ST_SET. All the features including the intrinsic features IF and the performance features PF are stored in the scenario set SC_SET. Note that a specific example of the scenario s generated by the generation section 13 may be a SCML (Supply Chain Management League) competition scenario, and the strategy set ST_SET may include one or more SCML competition strategies. However, this does not limit the present example embodiment.

[0094] (Step S103A) Then, in step S103A, the clustering section 23 clusters the scenarios SC included in the scenario set SC_SET on a performance space. In other words, the clustering section 23 clusters the performance features PF included in the scenario set SC_SET on a performance space. Here, the performance space is a space spanned by the performance features. For example, if the performance features include a feature A and B, then the corresponding performance space is spanned by the feature A and the feature B.

[0095] (Step S104A)) Then, in step S104A, the clustering section 23 keeps only representative strategies and performance features PF. In an example, with reference to the result of the clustering in step S103A, the clustering section 23 deletes the performance features PF which are not representative, and deletes the corresponding strategies applicable to the unrepresentative performance features PF. In other words, with reference to the result of the clustering in step S103A, the clustering section 23 deletes the strategies ST which are not representative (unneeded strategies ST), and deletes the corresponding performance features PF applicable to the unneeded strategies ST.

[0096] Note that the processes in step S103A and S104 may be regarded as optional processes. Therefore, the present example embodiment includes the Initial Dataset Generation processes where the steps S103A and S104 are not included. Note also that the step S102A may be iterated a plurality of times.

[0097] Fig. 10 is a schematic illustration showing exemplary results of the initial dataset generation. The i-space is an intrinsic space spanned by the intrinsic features IF of the scenarios SC, while the p-space is the performance space spanned by the performance features IF of the scenarios SC. In Fig. 10, each of the numbers 1 to 7 represents the different scenarios. It may be understood that as a result of the processes in step S103A and S104A, the scenarios 1 to 3 form a cluster, while the scenarios 4 to 6 form another cluster in the performance space. As shown in Fig. 10, according to the initial data generation, we have a set of scenarios SC with intrinsic features IF and performance features PF.

[0098] (Scenario Set Enhancement) The following description will discuss the scenario set enhancement carried out in the information processing apparatus 1A with reference to Fig. 11 and Fig. 12. Fig. 11 is a schematic illustration of data flow in the scenario set enhancement according to the second example embodiment. A set of processes in Fig. 11 may correspond to a specific example of step S1A in Fig. 8. It may be expressed that the scenario set enhancement is a process to find new scenarios SC that can distinguish the behavior of different existing strategies ST. It may also be expressed that the scenario set enhancement is a process to find situations that missing data in the initial data-set generation phase. Inputs to the scenario set enhancement may be features of each negotiation session and agent scores for each of them. Outputs of the scenario set enhancement may be features of new negotiation scenarios and agent scores for each of them.

[0099] (Step S12A) In step S12A, the obtaining section 11 obtains the performance features PF of the scenarios SC included in the scenario set SC_SET. Then, the clustering section 23 clusters the scenarios SC included in the scenario set SC_SET on a performance space. In other words, the clustering section 23 clusters the performance features PF included in the scenario set SC_SET on a performance space.

[0100] (Step S13A) Then, in step S13A, the generation section 13 generates one or more scenarios SC with reference to the clustering carried out in step S12A. The processes in step S13A may include: projecting the performance features PF obtained in step S12A into one or more intrinsic features IF; and generating one or more new negotiation scenarios SC with reference to the result of the projection. More specifically, the step S13A may include the following steps S131A to S134A.

[0101] (S131A) In step S131A, the generation section 13 searches (finds) unvisited location(s) in the performance space. Here, the unvisited location is a location where scenario(s) are not present nearby. In an example, the generation section 13 may find such an unvisited location with reference to distributions of the scenarios SC in the performance space. In Fig. 11, the unvisited location found by the generation section 13 is indicated as s in the performance space.

[0102] (Step S132A) Then, in step S132A, the generation section 13 (training section 131) trains the PI projector PIP, for example by a supervised learning, with reference to: the intrinsic features IF included in the scenario set SC_SET; the performance features PF included in the scenario set SC_SET; and the unvisited location s found by the generation section 13.

[0103] (Step S133A) In step S133A, the projection section 12 projects the unvisited location s into the intrinsic space by the trained PI projector PIP. In Fig. 11, the projected location in the intrinsic space is also indicated as s.

[0104] (Step S134A) Then , in step S134A, the generation section 13 run a simulation of the existing strategies ST with reference to the projected location s in the intrinsic space, and generate a new scenario SC corresponding to the projected location s in the intrinsic space. Then, the generation section 13 adds, to the scenario set SC_SET, the new scenario SC with all the features including the intrinsic features IF and the performance features PF. Note that the processes in step S131A to S134A may be iterated a plurality of times.

[0105] Fig. 12 is a schematic illustration showing exemplary results of the scenario set enhancement. As shown in Fig. 12, new scenarios x, y, and z are added by the generation section 13 and the corresponding new intrinsic features x, y, and z are obtained in the intrinsic space (i-space) by the PI projector PIP. As shown in Fig. 12, according to the scenario set enhancement, we have a set of scenarios SC with better coverage of the performance space (p-space).

[0106] (Strategy Set Enhancement) The following description will discuss the strategy set enhancement carried out in the information processing apparatus 1A with reference to Fig. 13 and Fig. 14. Fig. 13 is a schematic illustration of data flow in the strategy set enhancement according to the second example embodiment. A set of processes in Fig. 13 may correspond to a specific example of step S2A in Fig. 8. It may be expressed that the strategy set enhancement is a process to find new strategies ST that provide a better intrinsic-performance mapping (correspondence). It may also be expressed that the strategy set enhancement is a process to match intrinsic and performance scores by optionally training new strategies. Inputs to the strategy set enhancement may be features of each negotiation session and agent scores for each of them. Outputs of the strategy set enhancement may be new strategies and features of new negotiation session and agent scores for each of them.

[0107] (Step S 23A) In step S23A, the clustering section 12 clusters the scenarios SC included in the scenario set SC_SET. The clustering of the scenarios SC is carried out on all the features including the intrinsic features IF and the performance features PF.

[0108] (Step S24A) In step S24A, the generation section 13 generates one or more new strategies with reference to the result of the clustering in step S23A. More particularly, the step S24A may include the following steps S241A to S246A. It may be expressed that the generation of the one or more negotiation strategies ST is carried out so as to reduce a mismatch between one or more intrinsic features IF and one or more performance features PF.

[0109] (Step S241A) In step S241A, the generation section 13 determines whether there is a small cluster in the clusters of the scenarios SC generated in step S23A. In an example, the generation section 13 determines a size of a target cluster by comparing the number of scenarios SC included in the target cluster with a predetermined threshold. If the number of the scenarios SC included in the target cluster is less than the predetermined threshold, the generation section 13 may determine that the target cluster is small. In a case where there is a small cluster, the processes proceed to step S242A, otherwise the processes proceed to step S243A.

[0110] (Step S242A) In a case where there is a small cluster in the clusters of the scenarios SC generated in step S23A, the generation section 13 carries out an augmentation of the small cluster in step S242A. In an example, the generation section 13 generates one or more scenarios SC appropriate for this small cluster, and augments the small cluster with the generated new scenarios SC. The augmented cluster of the scenarios SC is stored in the scenario set SC_SET. Note that steps S241A and S242A may be regarded as optional processes. Therefore, the present example embodiment includes the Strategy Set Enhancement processes where the steps S241A and S242A are not included (Step S243A) In step S243A, the generation section 13 generates a new strategy ST and trains the new strategy ST for each of the clusters generated in step S23A, or augmented in step S242A. For example, the generation section 13 trains a new strategy ST with reference to the scenarios SC11 and SC12 included in the cluster CL1, and to the scenarios SC21 and SC22 included in the cluster CL2. The trained new strategies ST are stored in the strategy set ST_SET.

[0111] (Step S244A) Then, the generation section 13 runs the trained new strategies on all the scenarios SC, and obtains the performance features PF of each of the new strategies for each of these scenarios SC. The performance features PF of the scenarios SC are stored in the scenario set SC_SET.

[0112] (Step S245A) Then, the generation section 13 (training section 131) trains the IP projector IPP with reference to the intrinsic features IF and the performance features PF associated with each of the scenarios SC included in the scenario set SC_SET. One or more strategies ST included in the strategy set ST_SET may also be referred in the training of the IP projector IPP.

[0113] (Step S246A) Then, in step S246A, the generation section 13 determines whether the prediction accuracy of the IP projector IPP is enough or not. In an example, the generation section 13 determines that the prediction accuracy of the IP projector IPP is enough if the prediction accuracy is equal to or more than the predetermined threshold. In a case where the prediction accuracy of the IP projector IPP is enough, the processes end, otherwise the processed proceed to step S23A.

[0114] Fig. 14 is a schematic illustration showing exemplary results of the strategy set enhancement. As shown in Fig. 14, new scenarios x, y, and z are added by the generation section 13 and the corresponding performance features x, y, and z are obtained in the performance space (p-space) by the IP projector IPP. Here, the performance space is spanned by the new performance features X and Y representing the performance of the new scenarios SC generated in S243A. As shown in Fig. 14, according to the strategy set enhancement, we have a set of strategies ST with better matching between intrinsic features IF and performance features PF as confirmed by the IP Projector IPP.

[0115] (Guided Scenario and Strategy Generation) The following description will discuss the guided scenario and strategy generation carried out in the information processing apparatus 1A with reference to Fig. 15 and Fig. 16. Fig. 15 is a schematic illustration of data flow in the guided scenario and strategy generation according to the second example embodiment. A set of processes in Fig. 15 may correspond to a specific example of step S3A in Fig. 8. It may be expressed that the guided scenario and strategy generation is a process to find new strategies ST and scenarios SC that exhibit the high performance in the performance space. It may also be expressed that the guided scenario and strategy generation is a process to improve the set of strategies and contexts for the sake of agent training by improving the matching between intrinsic features and performance features. Inputs to the guided scenario and strategy generation may be features of all negotiation session and agent scores for each of them. Outputs of the guided scenario and strategy generation may be new strategies and scenarios for which intrinsic and performance feature distributions approximately match.

[0116] (Step S33A) In step S33A, the generation section 13 generates one or more negotiation scenarios SC with low performance with respect to one or more negotiation strategies ST. In an example, the generation section 13 generates one or more negotiation scenarios SC with low performance with respect to all negotiation strategies ST included in the strategy set ST_SET. More particularly, the step S33A may include the following steps S331A to S333A.

[0117] (Step S331A, S332A) In step S331A, the generation section 13 generate a new scenario SC. Then, in step S332A, the generation section 13 determines whether the generated scenario SC is hard or not. In an example, the generation section 13 determines that the generated scenario SC is hard if all the performance features PF of a plurality of strategies ST included in the strategy set ST_SET are below the predetermined threshold. In another example, the generation section 13 determines that the generated scenario SC is hard if all the performance features PF of all the strategies ST included in the strategy set ST_SET are below the predetermined threshold.

[0118] In a case where the generation section 13 determines that the generated scenario SC is hard, the generated scenario SC is stored in the scenario set SC_SET, otherwise the generation section 13 generates another new scenario SC in step S331A.

[0119] (Step S333A) In step S333A, the generation section 13 determines whether the number of trials reaches the predetermined number. The generation section 13 determines that the number of trials reaches the predetermined number, if the number of new scenarios generated and stored in the scenario set SC_SET reaches the predetermined number. In a case where the generation section 13 determines that the number of trials reaches the predetermined number, the processes end, otherwise the generation section 13 generates another new scenario SC in step S331A.

[0120] (Step S34A) Then, in step S34A, the generation section 13 generates one or more negotiation strategies ST with reference to the one or more negotiation scenarios SC generated in step S33A. More particularly, the step S34A may include the following steps S341A to S345A.

[0121] (Step S341A) In step S341A, the generation section 13 clusters a plurality of scenarios SC in the scenario set SC_SET. For example, the generation section 13 clusters all the scenarios SC included in the scenario set SC_SET with respect to all the features including the intrinsic feature IF and the performance feature PF. Note that step S341A may be carried out by the clustering section 23.

[0122] (Step S342A) Then, in step S342A, the generation section 13 (training section 131) generates a new strategy ST and trains the new strategy ST for each cluster. The trained strategy ST is stored in the strategy set ST_SET and is referred in step S343A.

[0123] (Step S343A) In step S343A, the generation section 13 runs each of the new strategies ST on all the scenarios SC in the scenario set SC_SET, and obtains the performance features PF of all the scenarios SC with respect to the new strategies ST. The performance features PF of the scenarios SC are stored in the scenario set SC_SET.

[0124] (Step S344A, S345A) In step S344A, the generation section 13 (training section 131) trains (updates) the IP projector IPP with reference to the intrinsic features IF and the performance features PF included in the scenario set SC_SET. Then, in step S345A, the generation section 13 (training section 131) determines whether the accuracy of the IP projector IPP decreases or not. In an example, the generation section 13 determines that the accuracy of the IP projector IPP decreases, if the prediction accuracy of the currently updated IP projector IPP is below the prediction accuracy of the previously updated IP projector IPP. In a case where the accuracy of the IP projector IPP decreases, the generation section 13 ends the processes and stores the previously updated IP projector IPP (which has the maximum prediction accuracy) in the storage section 20A. In a case where the accuracy of the IP projector IPP does not decrease, the processes proceed to step S331A. In this manner, according to step S344A and S345A, the IP projector IPP is updated such that it has the maximum prediction value. In other words, the IP projector IPP is updated until the accuracy of the IP projector IPP cannot be improved.

[0125] Fig. 16 is a schematic illustration showing exemplary results of the guided scenario and strategy generation. As shown in Fig. 16, new scenarios a to e are generated in such a way that the scenarios 1, 7, and a to e form a plurality of clusters in the intrinsic space (i-space). Note that according to the processes included the guided scenario and strategy generation, the scenarios 1, 7, and a to e also form a plurality of clusters in the performance space (p-space) in such a way that the clusters in the performance space (p-space) respectively corresponds to the clusters in the intrinsic space (p-space). As shown in Fig. 16, according to the guided scenario and strategy generation, we have a set of strategies ST with better matching between intrinsic features IF and performance features PF, while realizing high coverage of the intrinsic feature space (i-space).

[0126] (Final Context Generation) The following description will discuss the final context generation carried out in the information processing apparatus 1A with reference to Fig. 17 and Fig. 18. Fig. 17 is a schematic illustration of data flow in the final context generation according to the second example embodiment. A set of processes in Fig. 17 may correspond to a specific example of step S20A in Fig. 8. It may be expressed that the final context generation is a process to find new strategies ST that provide a better intrinsic-performance mapping (a better intrinsic-performance correspondence). Here, a context is a set of scenarios SC. In other words, a context is a cluster of scenarios SC. It may also be expressed that the final context generation is a process to define the final context, strategies and context classifier. Inputs to the final context generation may be all trained and original strategies and all new scenarios with both intrinsic and performance features. Outputs of the final context generation may be: Contexts: defining sets of scenarios that are similar in intrinsic features and lead to similar performance results, Strategies: One strategy for each context trained to achieve high performance, Context classifier: A classifier that checks a negotiation scenario and decides to which context does it belong.

[0127] (Step S201A In step S201A, the clustering section 23 clusters the scenarios SC in the scenario set SC_SET with respect to the performance features. One of more scenarios SC included in a cluster form a context. The contexts are referred in step S202A.

[0128] (Step S202A) In step S202A, the generation section 13 (training section 131) trains, per a context, a strategy ST in the strategy set ST_SET and obtains the trained strategy ST as a final strategy ST. The contexts and the final strategies are stored in the final context-strategy set.

[0129] (Step S203A) In step S203A, the generation section 13 (training section 131) trains the context classifier CC. Here, the context classifier CC is a classifier to classify the contexts. In an example, the context classifier CC is realized as a machine-learned neural network model which maps one or more intrinsic features IF to one or more contexts. The context classifier CC may also be expressed as a feature classifier.

[0130] Fig. 18 is a schematic illustration showing exemplary results of the final context generation. As shown in Fig. 18, each of machine-learned strategies ST which has been trained in step S202A are associated with each of the clusters. Here, as shown in Fig. 18, each of the clusters includes one or more learned contexts which have been trained, for example, in step S342A. As shown in Fig. 18, according to the final context generation, we have the contexts that can be discerned from the intrinsic features IF, and each of the contexts are associated with a high-performing strategy ST.

[0131] (Advantageous effect) According to the second example embodiment, the information processing apparatus 1A includes: an obtaining section 11 to obtain one or more performance features for a set of negotiation strategies, a projection section 12 to project the obtained performance features into one or more intrinsic features, and a generation section 13 to generate one or more negotiation scenarios with reference to the result of the projection.

[0132] As mentioned above, according to the information processing apparatus 1A, the performance features for a set of negotiation strategies are projected into one or more intrinsic features, and then one or more negotiation scenarios are generated reference to the result of the projection. Therefore, the information processing apparatus 1A can generate one or more negotiation scenarios appropriate for the given set of negotiation strategies. The generated negotiation scenarios may be used, for example, in a training process of negotiation apparatus, or in a reinforcement learning, etc..

[0133] According to the first example embodiment, the information processing apparatus 1A includes: a first obtaining section 21 (11) to obtain one or more intrinsic features for a set of negotiation scenarios, a second obtaining section 22 (11) to obtain one or more performance features for the set of negotiation scenarios, a clustering section 23 to cluster the negotiation scenarios with respect to the intrinsic features and the performance features, and a generation section 24 (13) to generate one or more negotiation strategies with reference to the result of the clustering.

[0134] As mentioned above, according to the information processing apparatus 1A, the negotiation scenarios with respect to the intrinsic features and the performance features are clustered, and then one or more negotiation strategies are generated with reference to the result of the clustering. Therefore, the information processing apparatus 1A can generate one or more negotiation strategy appropriate for the given set of negotiation scenarios. The generated negotiation strategies may be used, for example, in a training process of negotiation apparatus, or in a reinforcement learning, etc..

[0135] According to the first example embodiment, the information processing apparatus 1A includes: a first obtaining section 31 (11) to obtain one or more performance features for a set of negotiation scenarios, a second obtaining section 32 (11) to obtain one or more negotiation strategies, a first generation section 33 (13) to generate one or more negotiation scenarios with low performance with respect to all the negotiation strategies, and a second generation section 34 (13) to generate one or more negotiation strategies with reference to the one or more generated negotiation scenarios.

[0136] As mentioned above, according to the information processing apparatus 1A, the one or more negotiation scenarios with low performance with respect to all the negotiation strategies are generated, and then the one or more negotiation strategies are generated with reference to the one or more generated negotiation scenarios. Therefore, the information processing apparatus 1A can generate one or more negotiation strategy appropriate for the given set of negotiation scenarios. The generated negotiation strategies and scenarios may be used, for example, in a training process of negotiation apparatus, or in a reinforcement learning, etc..

[0137] The advantageous effect of the information processing apparatus 1A may be expressed as follows. Negotiation is widely used to reach agreements in procurement, transportation and related fields. With the recent explosion of AI applications, it is expected that automated negotiation within such scenarios will be of paramount importance. Effective negotiation depends on features of the scenario surrounding the negotiation process. Training a single automated negotiation (auto-negotiation) strategy for all possible scenarios is not effective and for each individual scenario is not feasible. Moreover, there is no known method for defining training contexts that allows for effective yet feasible training of auto-negotiation strategies.

[0138] The information processing apparatus 1A enables effective negotiation in a wide variety of scenarios which is expected to increase the market penetration of auto-negotiation and deliver value for customers using it.

[0139] The advantageous effect of the information processing apparatus 1A may also be expressed as follows. Although there have been a wide variety of researches and attempts to realize an auto-negotiation, most researches in auto-negotiation assumes a fixed utility function with no context which is almost never the case in real world scenarios. Most research in training with feedback (e.g. RL) assume a fixed environment or an independent set of environments. Methods for in-context learning in an existing method cannot use existing information about the context (e.g. market), assume full observability of the state, or fail to disentangle context learning from policy training. The need for generating data-driven training contexts is not obvious in all three cases.

[0140] The above difficulties may be expressed as follows. We do not put any limitation on the number of possible scenarios or any semantics on scenario defining characteristics which makes it hard to find usable contexts for training. The boundaries of usable training contexts can be discrete and hard to ascertain without having a set of “good” strategies trained for them but having these strategies “assumes” that we already have good context definition. This is a chicken-and-egg problem that we solve.

[0141] The information processing apparatus 1A appropriately defines the problem of domain-dependent contexts for training auto-negotiation agents (negotiation apparatus 1A,1B) (this can be generalized for other types of agents as well). The proposed solution may be based on iterative utilization of clustering and classification in a unique order.

[0142] <Third example embodiment> The following description will discuss details of a third example embodiment of the invention with reference to the drawings. Note that the same reference numerals are given to elements having the same functions as those described in the first and second example embodiments, and descriptions of such elements are omitted as appropriate.

[0143] (Brief overview of negotiation techniques according to the third example embodiment) Negotiation techniques (method, apparatus, system) according to the third example embodiment deal with negotiation processes between a negotiation agent and one or more opponent (or partner) agents. In particular, the negotiation techniques according to the third example embodiment may include a step (means) of obtaining one or more offers provided by an opponent (or partner) agent, and a step (means) of generating one or more counter-offers to the opponent (or partner) agent with reference to the offers provided by the opponent (or partner) agent. The number of opponent (or partner) agents which may be dealt with the negotiation techniques according to the third example embodiment may not be limited to one, but it may be plural.

[0144] As used herein, the term "outcome(s)" may include one or more offers provided by an opponent (or partner) agent, and one or more counter-offers generated by the negotiation agent. The term "negotiation agent" may also be referred to as "negotiation entity". The "negotiation agent" and "opponent (or partner) agent" may be an apparatus, may be a human, or may be an organization.

[0145] The term "negotiation apparatus" refers to, for example, an apparatus that functions as a negotiation entity. Examples of the negotiation apparatus may include a computer, a robot, a drone, an automated driving vehicle, and the like. The term "negotiation" refers to, for example, proceedings in which negotiation entities alternately provide draft agreement candidates (offers and counter-offers) until the negotiation entities reach an agreement, a time limit expires, or a limited number of sessions is reached. The negotiation may be negotiation between entities in conflict of interest, which may correspond to a negotiation between the negotiation agent and an opponent agent. The negotiation may be negotiation (adjustment) between entities not in conflict of interest, which may correspond to a negotiation between the negotiation agent and a partner agent. The negotiation techniques according to the first example embodiment may be applied to various fields including manufacturing and transportation.

[0146] (Configuration of negotiation system 100B) The following description will discuss a configuration of a negotiation system 100B according to the third example embodiment with reference to Fig. 19. Fig. 19 is a block diagram illustrating a configuration of the negotiation system 100B. As illustrated in Fig. 19, the negotiation system 100B includes a negotiation apparatus 1B, and a plurality of negotiation agents 50-0, 50-1, 50-2, etc.. Here, the negotiation apparatus 1B and negotiation agents 50-0, 50-1, 50-2, etc. are connected each other via a network N. A specific configuration of the network N does not limit the present example embodiment, and the network N is, for example, a wireless local area network (LAN), a wired LAN, a wide area network (WAN), a public network, a mobile data communication network, or a combination of these networks. Each of the negotiation agents 50-0, 50-1, 50-2 are, for example, an opponent agent or a partner agent.

[0147] (Negotiation apparatus 1B) As illustrated in Fig. 19, the negotiation apparatus 1B includes a control section 10B, a storage section 20A, a communication section 30, and an input-output section 40. As shown in Fig. 19, the negotiation apparatus 1B includes all the configurations included in the information processing apparatus (negotiation apparatus) 1A according to the second example embodiment. Furthermore, the negotiation apparatus 1B includes a negotiation section 15 which carry out a negotiation based on the trained strategy ST trained by the training section 131.

[0148] (Overview of processes in negotiation system 100B) The following description will discuss a brief overview of processes carried out in negotiation system 100B with reference to Fig. 20. Fig. 20 is a schematic illustration of data flow in training phase, testing phase, and negotiation phase according to the second example embodiment.

[0149] (Training Phase) The left of Fig. 20 indicates data flows in the training phase. Here, the reference numeral pnindicates the learning agent (agent which is target for training), while the reference numerals p1to pn-1indicate the background agents (agents which are not target for training). As shown in left of Fig. 20, the background agents p1to pn-1are controlled in the negotiation environment by the training section 131 so as to generate their offers with reference to their own utility functions u1to un-1, respectively.

[0150] In the training phase, the state of the learning agent pnis observed by the observation manager. Here, the state of the learning agent pnincludes offer(s) which have been provided by any of the background agents p1to pn-1and then received by the learning agent pn. Note that the states observed by the observation manager may include an extra state other than the state of the learning agent pn. Note also that the training section 131 may play a role of the observation manager. The state (offer) of the learning agent pnis also provided, under a control of the training section 131, to a reward function. Note that the utility un, which may be referred by the learning agent pn, is also provided to the reward function. The reward function calculates a reward corresponding to the state (offer) of the learning agent pnand the utility un.

[0151] The observed state (offer) observed by the observation manager and the reward obtained from the state (offer) are provided to a training algorithm. The training algorithm receives the state (offer) and the reward, then generates observation information with reference to the state (offer) and the reward. The observation information is input to a model. The model receives the observation information and generates an action (counter-offer) with reference to the observation information and the policy (strategy). Note here that the observation information corresponds to an observation observed by the observation manager in the testing phase and the negotiation phase (after deployment). Note also that the action is also input to the training algorithm and taken into account in generating the observation information in the next iterative step.

[0152] The action generated by the model is then provided to the action manager. Here, the training section 131 may play a role of the action manager. The action manager provides the action (counter-offer) to the learning agent pn. The learning agent pnprovides the counter-offer to the background agents p1to pn-1and then receives next offers provided by any of the background agents p1to pn-1.

[0153] In the training phase, the training section 131 iterates above processes and trains the model (in other words, the policy or strategy) until the reward satisfies a predetermined criterion.

[0154] (Testing Phase) As shown in the middle of Fig. 20, in the testing phase, the training section 131 does not refer to the training algorithm. The state (offer) observed by the observation manager is provided to the model as the observation information. The training section 131 calculates the reward and check that the reward satisfies the desired condition.

[0155] (Negotiation Phase) As shown in the right of Fig. 20, in the negotiation phase (after deployment), the negotiation section 15 plays a role of the observation manager and obtains the state (received offer). The state (received offer) are negotiation section 15, and the counter-offer is generated by the negotiation section 15 with reference to the trained policy. Note also that in the negotiation phase, the negotiation section 15 may also play a role of the action manager.

[0156] Note that in the above explanations of the training phase and the testing phase, the negotiation environment, the observation manager, the reward function, and the action manager may refer to the contexts and strategies generated and trained by the generation section 13. Note also that the training phase and / or the testing phase may also include: a process of defining contexts (each context is a set of scenarios), a process of training a good strategy (e.g. model or policy) per context, and a process of training a selector that finds an appropriate context.

[0157] Here, the selector may correspond to the context classifier CC generated or trained in step S203A as explained above. Note also that the training phase and / or the testing phase may also include a process of learning the contexts. In an example, the contexts are trained by the training section 131 based on the performance of existing and new strategies ST on existing and new scenarios SC. Then, the training section 131 may train a selector (the context classifier CC) on, for example, observable market features, etc..

[0158] In this manner, according to the third example embodiment, it is possible to train appropriate strategies for unseen scenarios in terms of an unsupervised data-driven method.

[0159] (Application example) The following description will discuss an application example of the negotiation apparatus 1A and 1B. The following application example corresponds to a case where the negotiation apparatus 1A and 1B are applied to procurement in supply chains. In the procurement, how to procure may vary depending on production-needs from suppliers, and the aim is to maximize profit through negotiation. A scenario may corresponds to a complete specification of the market (e.g. n. suppliers, n. competitors, trading prices, etc). We can only observe some information about the scenario (e.g. we do not know production costs of our competitors). We can simulate any specific “scenario”, while in the real world, we can be in an unseen ”scenario”.

[0160] The main challenges in an application to procurement are for example listed as below.

[0161] We need to train multiple strategies (for different “contexts”), How to select the training set for each of strategy (i.e. what is a good context)? We need to select the appropriate strategy in unseen scenarios How to detect an appropriate “context” for the current scenario? According to the negotiation apparatus 1A and 1B, as explained above, the above listed challenges can be solved. In the following, we will discuss a more specific application example of the negotiation apparatus 1A and 1B with reference to Fig. 21 to Fig. 27.

[0162] Fig. 21 is a diagram illustrating a specific example of the initial dataset generation according to an example embodiment. In Fig. 21, a plurality of scenarios S0 to S3 are listed, each of which has a plurality of intrinsic features (Number of Suppliers, Number of Competitors, and Competitor's Production Cost). A scenario set including these scenarios and a strategy set (strategy A, B, C) are input to a process of the initial dataset generation. Then the performance map (performance features) are obtained foe each of the scenarios S0 to S3. In this example, the performance features of the strategy A are 0,12, -0.4, 0.0, and 0.11, respectively for the scenarios S0, S1,S2, and S3. As shown in Fig. 21, the performance features PF are obtained after removing the redundant scenarios. This process may correspond to step S104A in Fig. 9. In this example, the scenario S3 is determined as redundant by the clustering section 23, therefore is dropped.

[0163] Fig. 22 is a diagram illustrating a specific example of the scenario set enhancement according to an example embodiment. As shown is Fig. 22, by the scenario set enhancement, a new scenario S4' is added whose performance features are -0.4, 0.6, and -0.4, respectively for the strategies A, B, and C. This process may correspond to step S131A in Fig. 11. Then, the corresponding intrinsic features are obtained by the PI projector. This process may correspond to step 133A in Fig. 11. Then, the updated performance features are obtained. In this example, there updated performance features for the scenario S4' are -0.3, 0,31, and -0.28, respectively for strategies A, B, and C. This process may correspond to step S134A in Fig. 11.

[0164] Fig. 23 is a diagram illustrating a specific example of the strategy set enhancement according to an example embodiment. As shown is Fig. 23, by the strategy set enhancement, new strategies D and E are added, and the performance features are updated. These processes may correspond to step S243A and S244A in Fig. 13.

[0165] Fig. 24 is a diagram illustrating a specific example of the guided strategy and scenario enhancement according to an example embodiment. As shown is Fig. 24, by the guided strategy and scenario enhancement, new scenarios S5' and S6' are added. This process may correspond to step S33A in Fig. 15.

[0166] Fig. 25 is a diagram illustrating another specific example of the guided strategy and scenario enhancement according to an example embodiment. As shown is Fig. 25, by the guided strategy and scenario enhancement, new strategies F and G are added and the strategies B, C, D are deleted. In other words, the strategies B, C, D are replaced by or updated to the new strategies F and G. These processes may correspond to step S341A to S343A in Fig. 15.

[0167] Fig. 26 is a diagram illustrating a specific example of the final context generation according to an example embodiment. As shown is Fig. 26, by the final context generation, three clusters {S0}, {S1, S5', S6'}, {S2, S4} are formed (generated). Then training of new strategies for each cluster is carried out. In this example, the strategy A is trained for the cluster {S0}, the new strategy G is trained for the cluster {S1, S5', S6'}, and the new strategy G is trained for the cluster {S2, S4}. Then, the context classifier CC is trained based on the performance features. In this example, the context classifier CC is trained such that it carries out a classification as follows: If Number of Suppliers is more than 8, and Number of Competitors is less than 2, then select the strategy A, Else Number of Suppliers + Number of Competitors is more than 12, then select the strategy G, Else select the strategy H.

[0168] Fig. 27 is a diagram illustrating a specific example of the negotiation process after deployment. As shown in Fig. 27, the negotiation section 15 finds intrinsic features of the target scenario (in this example, the scenario S0). Then, by using the context classifier CC, the negotiation section 15 selects the strategy G as a negotiation policy (negotiation strategy) after the deployment. <Example of configuration achieved by software> One or some of or all of the functions of the negotiation apparatus (negotiation scenario generation apparatus, negotiation strategy generation apparatus, information processing apparatus) 1, 2, 3, 1A, 1B can be realized by hardware such as an integrated circuit (IC chip) or can be alternatively realized by software.

[0169] In the latter case, each of the apparatus 1, 2, 3, 1A, 1B is realized by, for example, a computer that executes instructions of a program that is software realizing the foregoing functions. Fig. 28 illustrates an example of such a computer (hereinafter, referred to as "computer C"). The computer C includes at least one processor C1 and at least one memory C2. The memory C2 stores a program P for causing the computer C to function as any of the apparatus 1, 2, 3, 1A, 1B. In the computer C, the processor C1 reads the program P from the memory C2 and executes the program P, so that the functions of any of the apparatus 1, 2, 3, 1A, 1B are realized.

[0170] As the processor C1, for example, it is possible to use a central processing unit (CPU), a graphic processing unit (GPU), a digital signal processor (DSP), a micro processing unit (MPU), a floating point number processing unit (FPU), a physics processing unit (PPU), a microcontroller, or a combination of these. The memory C2 can be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination of these.

[0171] Note that the computer C can further include a random access memory (RAM) in which the program P is loaded when the program P is executed and in which various kinds of data are temporarily stored. The computer C can further include a communication interface for carrying out transmission and reception of data with other devices. The computer C can further include an input-output interface for connecting input-output devices such as a keyboard, a mouse, a display, and a printer.

[0172] The program P can be stored in a non-transitory tangible storage medium M which is readable by the computer C. The storage medium M can be, for example, a tape, a disk, a card, a semiconductor memory, a programmable logic circuit, or the like. The computer C can obtain the program P via the storage medium M. The program P can be transmitted via a transmission medium. The transmission medium can be, for example, a communications network, a broadcast wave, or the like. The computer C can obtain the program P also via such a transmission medium. <Additional Remark 1> The present invention is not limited to the foregoing example embodiments, but may be altered in various ways by a skilled person within the scope of the claims. For example, the present invention also encompasses, in its technical scope, any example embodiment derived by properly combining technical means disclosed in the foregoing example embodiments.

[0173] <Additional Remark 2> The whole or part of the example embodiments disclosed above can be described as follows. Note, however, that the present invention is not limited to the following example aspects.

[0174] (Aspect A1) A negotiation scenario generation method comprising obtaining one or more performance features for a set of negotiation strategies, projecting the obtained performance features into one or more intrinsic features, and generating one or more first negotiation scenarios with reference to the result of the projection.

[0175] (Aspect A2) A negotiation scenario generation method according to Aspect A1, further comprising obtaining one or more intrinsic features for the set of negotiation scenarios, clustering the negotiation scenarios with respect to the intrinsic features and the performance features, and generating one or more first negotiation strategies with reference to the result of the clustering.

[0176] (Aspect A3) A negotiation strategy generation method according to Aspect A2, further comprising generating one or more second negotiation scenarios with low performance with respect to all the first negotiation strategies, and generating one or more second negotiation strategies with reference to the one or more second negotiation scenarios.

[0177] (Aspect A4) A negotiation strategy generation method according to Aspect A3, further comprising clustering the one or more second negotiation scenarios to generate one or more contexts, each of which contexts includes at least one of the second negotiation scenarios, and training a context classifier with reference to the one or more generated contexts.

[0178] (Aspect B1) A negotiation strategy generation method comprising obtaining one or more intrinsic features for a set of negotiation scenarios, obtaining one or more performance features for the set of negotiation scenarios, clustering the negotiation scenarios with respect to the intrinsic features and the performance features, and generating one or more negotiation strategies with reference to the result of the clustering.

[0179] (Aspect B2) A negotiation method according to Aspect B1, wherein generating the one or more negotiation strategies includes training the one or more negotiation strategies for each of clusters generated by the clustering the negotiation scenarios.

[0180] (Aspect B3) A negotiation method according to Aspect B2, wherein generating the one or more negotiation strategies further includes: training a projector to project one or more intrinsic features to one or more performance features; and determining whether an accuracy of the projector reaches a predefined threshold.

[0181] (Aspect B4) A negotiation method according to any one of Aspects B1 to B3, wherein generating the one or more negotiation strategies is carried out so as to reduce a mismatch between one or more intrinsic features and one or more performance features.

[0182] (Aspect C1) A negotiation strategy generation method comprising obtaining one or more performance features for a set of negotiation scenarios, obtaining one or more negotiation strategies, generating one or more negotiation scenarios with low performance with respect to all the negotiation strategies, and generating one or more negotiation strategies with reference to the one or more generated negotiation scenarios.

[0183] (Aspect C2) A negotiation method according to Aspect C1 further includes training a projector to project one or more intrinsic features for the one or more generated negotiation scenarios to one or more performance features.

[0184] (Aspect D1) A negotiation scenario generation apparatus comprising an obtaining means to obtain one or more performance features for a set of negotiation strategies, a projection means to project the obtained performance features into one or more intrinsic features, and a generation means to generate one or more negotiation scenarios with reference to the result of the projection.

[0185] (Aspect E1) A negotiation strategy generation apparatus comprising a first obtaining means to obtain one or more intrinsic features for a set of negotiation scenarios, a second obtaining means to obtain one or more performance features for the set of negotiation scenarios, a clustering means to cluster the negotiation scenarios with respect to the intrinsic features and the performance features, and a generation means to generate one or more negotiation strategies with reference to the result of the clustering.

[0186] (Aspect F1) A negotiation strategy generation apparatus comprising a first obtaining means to obtain one or more performance features for a set of negotiation scenarios, a second obtaining means to obtain one or more negotiation strategies, a first generation means to generate one or more negotiation scenarios with low performance with respect to all the negotiation strategies, and a second generation means to generate one or more negotiation strategies with reference to the one or more generated negotiation scenarios.

[0187] (Aspect G1) A negotiation scenario generation program causing a computer to function as: an obtaining means to obtain one or more performance features for a set of negotiation strategies, a projection means to project the obtained performance features into one or more intrinsic features, and a generation means to generate one or more negotiation scenarios with reference to the result of the projection.

[0188] (Aspect H1) A negotiation strategy generation program causing a computer to function as: a first obtaining means to obtain one or more intrinsic features for a set of negotiation scenarios, a second obtaining means to obtain one or more performance features for the set of negotiation scenarios, a clustering means to cluster the negotiation scenarios with respect to the intrinsic features and the performance features, and a generation means to generate one or more negotiation strategies with reference to the result of the clustering.

[0189] (Aspect I1) A negotiation strategy generation program causing a computer to function as: a first obtaining means to obtain one or more performance features for a set of negotiation scenarios, a second obtaining means to obtain one or more negotiation strategies, a first generation means to generate one or more negotiation scenarios with low performance with respect to all the negotiation strategies, and a second generation means to generate one or more negotiation strategies with reference to the one or more generated negotiation scenarios.

[0190] <Additional Remark 3> The whole or part of the example embodiments disclosed above can be described as follows. Note, however, that the present invention is not limited to the following example aspects.

[0191] (Aspect J1) A method for finding new scenarios that can distinguish the performance of a set of negotiation strategies. In one embodiment, this can be achieved by training a neural network (e.g. PI projector) to project performance features into intrinsic scenario features (both may be observable and unobservable).

[0192] (Aspect J2) A method for training new strategies that reduce the mismatch between the distribution of scenarios in the performance space and intrinsic feature space. In one embodiment, this is done by clustering scenarios, training a strategy for each cluster and training an IP Projector to map intrinsic features to performance stopping the process when the accuracy of the IP Projector reaches a predefined threshold.

[0193] (Aspect J3) A method for finding new strategies and scenarios so that at least one strategy has high performance for each scenario. In one embodiment, this is achieved by generating new scenarios with low performance on all strategies, training new strategies to cover them and updating the IP projector until the accuracy of the IP projector cannot be improved.

[0194] (Aspect J4) A method for training a set of strategies on a set of scenario clusters called contexts and a context classifier CC that can assign any scenario to one context by running the methods in Aspects J1 to J3 in a sequence and finally conducting a clustering of the scenarios based on performance, training a strategy for each cluster and then training a classifier CC to detect the context.

[0195] S1 Negotiation Scenario Generation Method S2, S3 Negotiation Strategy Generation Method 1 Negotiation Scenario Generation Apparatus 2, 3 Negotiation Strategy Generation Apparatus 1A, 1B Negotiation Apparatus (Information Processing Apparatus, Negotiation Scenario Generation Apparatus, Negotiation Strategy Generation Apparatus) 11, 21, 22, 31, 32 Obtaining Section 12 Projection Section 13, 24, 33, 34 Generation Section 23 Clustering Section

Claims

1. A negotiation scenario generation method comprising: obtaining one or more performance features for a set of negotiation strategies; projecting the obtained performance features into one or more intrinsic features,; and generating one or more first negotiation scenarios with reference to the result of the projection.

2. The negotiation scenario generation method according to Claim 1, further comprising: obtaining one or more intrinsic features for the set of negotiation scenarios; clustering the negotiation scenarios with respect to the intrinsic features and the performance features; and generating one or more first negotiation strategies with reference to the result of the clustering.

3. The negotiation strategy generation method according to Claim 2, further comprising: generating one or more second negotiation scenarios with low performance with respect to all the first negotiation strategies; and generating one or more second negotiation strategies with reference to the one or more second negotiation scenarios.

4. The negotiation strategy generation method according to Claim 3, further comprising: clustering the one or more second negotiation scenarios to generate one or more contexts, each of which contexts includes at least one of the second negotiation scenarios; and training a context classifier with reference to the one or more generated contexts.

5. A negotiation strategy generation method comprising: obtaining one or more intrinsic features for a set of negotiation scenarios; obtaining one or more performance features for the set of negotiation scenarios; clustering the negotiation scenarios with respect to the intrinsic features and the performance features; and generating one or more negotiation strategies with reference to the result of the clustering.

6. The negotiation method according to Claim 5, wherein generating the one or more negotiation strategies includes training the one or more negotiation strategies for each of clusters generated by the clustering the negotiation scenarios.

7. The negotiation method according to Claim 6, wherein generating the one or more negotiation strategies further includes: training a projector to project one or more intrinsic features to one or more performance features; and determining whether an accuracy of the projector reaches a predefined threshold.

8. The negotiation method according to any one of Claims 5 to 7, wherein generating the one or more negotiation strategies is carried out so as to reduce a mismatch between one or more intrinsic features and one or more performance features.

9. A negotiation strategy generation method comprising: obtaining one or more performance features for a set of negotiation scenarios; obtaining one or more negotiation strategies; generating one or more negotiation scenarios with low performance with respect to all the negotiation strategies; and generating one or more negotiation strategies with reference to the one or more generated negotiation scenarios.

10. The negotiation method according to Claim 9 further includes training a projector to project one or more intrinsic features for the one or more generated negotiation scenarios to one or more performance features.

11. A negotiation scenario generation apparatus comprising: an obtaining means to obtain one or more performance features for a set of negotiation strategies; a projection means to project the obtained performance features into one or more intrinsic features; and a generation means to generate one or more negotiation scenarios with reference to the result of the projection.

12. A negotiation strategy generation apparatus comprising: a first obtaining means to obtain one or more intrinsic features for a set of negotiation scenarios; a second obtaining means to obtain one or more performance features for the set of negotiation scenarios; a clustering means to cluster the negotiation scenarios with respect to the intrinsic features and the performance features; and a generation means to generate one or more negotiation strategies with reference to the result of the clustering.

13. A negotiation strategy generation apparatus comprising: a first obtaining means to obtain one or more performance features for a set of negotiation scenarios; a second obtaining means to obtain one or more negotiation strategies; a first generation means to generate one or more negotiation scenarios with low performance with respect to all the negotiation strategies; and a second generation means to generate one or more negotiation strategies with reference to the one or more generated negotiation scenarios.

14. A negotiation scenario generation program causing a computer to function as: an obtaining means to obtain one or more performance features for a set of negotiation strategies; a projection means to project the obtained performance features into one or more intrinsic features; and a generation means to generate one or more negotiation scenarios with reference to the result of the projection.

15. A negotiation strategy generation program causing a computer to function as: a first obtaining means to obtain one or more intrinsic features for a set of negotiation scenarios; a second obtaining means to obtain one or more performance features for the set of negotiation scenarios; a clustering means to cluster the negotiation scenarios with respect to the intrinsic features and the performance features; and a generation means to generate one or more negotiation strategies with reference to the result of the clustering.

16. A negotiation strategy generation program causing a computer to function as: a first obtaining means to obtain one or more performance features for a set of negotiation scenarios; a second obtaining means to obtain one or more negotiation strategies; a first generation means to generate one or more negotiation scenarios with low performance with respect to all the negotiation strategies; and a second generation means to generate one or more negotiation strategies with reference to the one or more generated negotiation scenarios.

Images