Proposal generation device, learning device, proposal generation method, learning method, and program

Abstract

To enable the use of the same model in negotiations across different fields. [Solution] The proposal generation device comprises: domain embedding means for each of the one or more items designated as items to be proposed, which converts each of the possible bids, which are combinations of that item and one of the proposal options that are selectable proposal content for that item, into a numerical vector that can identify the bid; bid history embedding means for converting each bid included in the bidding history in negotiations into a numerical vector that can identify the bid; and bid determination means for determining which bid to propose to the other party from among all possible bids, using the numerical vectors obtained by converting each of the possible bids and the numerical vectors obtained by converting each of the bids included in the bidding history.

Description

[Technical Field]

[0001] The present invention relates to a proposal generation device, a learning device, a proposal generation method, a learning method, and a program. [Background technology]

[0002] In some cases, proposals in negotiations are decided using models. For example, the method described in Patent Document 1 describes how a negotiating agent and a counterpart agent are simultaneously trained using reinforcement learning to discourse using bits of an interpretable sequence. In the method described in Patent Document 1, the negotiating agent and the counterpart agent perform several rounds of negotiation levels with each other and learn to cooperate with each other based on the outcome as a reward function. [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2020-013568 [Overview of the project] [Problems that the invention aims to solve]

[0004] It is preferable to be able to use the same model for negotiations in different fields.

[0005] One example of the purpose of this disclosure is to provide a proposal generation device, a learning device, a proposal generation method, a learning method, and a program that can solve the above-mentioned problems. [Means for solving the problem]

[0006] According to a first aspect of this disclosure, the proposal generation device includes: domain embedding means for each of the one or more items designated as items to be proposed, which converts each of the possible bids, which are combinations of that item and one of the proposal options that are selectable proposal content for that item, into a numerical vector that can identify the bid; bid history embedding means for converting each bid included in the bidding history in negotiations into a numerical vector that can identify the bid; and bid determination means for determining which bid to propose to the other party from among all the possible bids, using the numerical vectors obtained by converting each of the possible bids and the numerical vectors obtained by converting each of the bids included in the bidding history.

[0007] According to a second aspect of this disclosure, the learning device includes: domain embedding means for each of the one or more items designated as items to be proposed, which converts each of the possible bids, which are combinations of that item and one of the proposed options that are selectable proposed content for that item, into a numerical vector that can identify the bid; bid history embedding means for converting each bid included in the bidding history in negotiations into a numerical vector that can identify the bid; bid determination means for determining which bid to propose to the other party from among all possible bids, using the numerical vectors obtained by converting each of the possible bids and the numerical vectors obtained by converting each of the bids included in the bidding history; and learning processing means for learning the bid determination method by the bid determination means.

[0008] According to a third aspect of this disclosure, the proposal generation method includes a computer converting each of the possible bids, for each of the one or more items designated as items to be proposed, that are combinations of that item and one of the proposal options which are selectable proposal contents for that item, into a bid-identifiable numerical vector; converting each bid included in the bidding history in negotiations into a bid-identifiable numerical vector; and using the numerical vectors converted from each of the possible bids and the numerical vectors converted from each of the bids included in the bidding history, the computer determines which bid to propose to the other party from among all the possible bids.

[0009] According to a fourth aspect of this disclosure, the learning method includes the computer converting each of the possible bids, for each of the one or more items designated as items to be proposed, that are combinations of that item and one of the proposed options which are selectable proposed content for that item, into a bid-identifiable numerical vector; converting each bid included in the bidding history in negotiations into a bid-identifiable numerical vector; using the numerical vectors converted from each of the possible bids and the numerical vectors converted from each of the bids included in the bidding history, determining which bid to propose to the other party from among all the possible bids; and learning the method for determining the bid.

[0010] According to a fifth aspect of this disclosure, the program causes a computer to perform the following actions: convert each of the bids that can be made by combining one or more items designated as items to be proposed with one of the proposed options which are selectable proposed content for that item into a numerical vector that can identify the bid; convert each bid included in the bidding history in negotiations into a numerical vector that can identify the bid; and determine which bid to propose to the other party from among all the bids that can be made, using the numerical vectors obtained by converting each of the bids that can be made and the numerical vectors obtained by converting each bid included in the bidding history.

[0011] According to the sixth aspect of this disclosure, the program causes a computer to perform the following actions: convert each of the bids that can be made by combining one or more items designated as items to be proposed with one of the proposed options which are selectable proposed content for that item into a numerical vector that can identify the bid; convert each bid included in the history of bids in negotiations into a numerical vector that can identify the bid; determine which bid to propose to the other party from among all the bids that can be made using the numerical vectors obtained by converting each of the bids that can be made and the numerical vectors obtained by converting each bid included in the history of bids; and learn the method for determining the bid. [Effects of the Invention]

[0012] According to the aspects of this disclosure, the same model can be used in common for negotiations in different fields. [Brief explanation of the drawing]

[0013] [Figure 1] This figure shows an example of the configuration of a learning device according to at least one embodiment. [Figure 2]This figure shows an example of the configuration of a bid selection unit according to at least one embodiment. [Figure 3] This figure shows an example of data input and output in a learning device according to at least one embodiment. [Figure 4] This figure shows an example of the configuration of a proposal generation device according to at least one embodiment. [Figure 5] This figure shows an example of data input and output in a proposal generation device according to at least one embodiment. [Figure 6] This figure shows the experimental results of a proposal generation device according to at least one embodiment. [Figure 7] This figure shows an example of the configuration of a proposal generation device according to at least one embodiment. [Figure 8] This figure shows an example of the configuration of a learning device according to at least one embodiment. [Figure 9] This figure shows an example of the processing steps in a proposal generation method according to at least one embodiment. [Figure 10] This figure shows an example of the processing steps in a learning method according to at least one embodiment. [Figure 11] This is a schematic block diagram showing the configuration of a computer according to at least one embodiment. [Modes for carrying out the invention]

[0014] The embodiments will be described below with reference to the drawings.

[0015] <First Embodiment> Figure 1 shows an example of the configuration of a learning device according to at least one embodiment. As shown in Figure 1, the learning device 100 comprises a communication unit 110, a display unit 120, an operation input unit 130, a storage unit 180, and a processing unit 190. The storage unit 180 comprises a domain information storage unit 181 and a bid history storage unit 182. The processing unit 190 comprises a domain embedding unit 191, an encoder unit 192, a bid history embedding unit 193, a decoder unit 194, a bid selection unit 195, a value calculation unit 196, and a learning processing unit 197.

[0016] The learning device 100 learns how to generate proposals in negotiations. Learning, in this context, refers to adjusting the parameter values ​​of a machine learning model. Learning can also be referred to as training. The learning device may be configured using a computer such as a workstation (WS) or a personal computer (PC).

[0017] The learning device 100 shall conduct negotiations within the following framework. A proposal in negotiations is also called a bid. Here, we will assume a two-party negotiation. Of the two negotiating parties, the learning device 100 will be referred to as "Self," and the other negotiating party will be referred to as "Opponent." In the negotiation, Self and Opponent will take turns making bids. Here, we assume the other party is also a device and will refer to it as the "other agent device." However, the other party may also be a person.

[0018] One step is defined as both the player and the opponent making one bid each. The combination of one player's bid and the opponent's bid is also called a one-step bid. Within a one-step bid, both the player's bid and the opponent's bid are referred to as one bid. In the following, time will be represented in time steps, and the bid at step t will also be referred to as time t or step t. Here, t is an integer such that t ≥ 1. It is assumed that proposals are made for n items in one bid. Here, n is an integer where n ≤ 1. The items to be proposed (the items targeted for proposal) are also referred to as Issues. The set I of items to be proposed is shown as in Equation (1).

[0019]

Number

[0020] I1, I2, ···, I n each represent an item to be proposed. For each item to be proposed, any one of k i options of proposed content is to be selected. Here, i is an integer from 1 ≤ i ≤ n that identifies the item to be proposed. The selectable proposed content is also referred to as an Option or a proposal option. The set V of proposal options i is shown as in Equation (2).

[0021]

Number

[0022] V i 1, V i 2, ··· V i ki each represent a proposal option. An individual bid (one bid) ω is shown as in Equation (3).

[0023]

Number

[0024] c i represents the proposal option selected for the i-th item to be proposed, where 1 ≤ C i ≤ k i is an integer. The set of bids (the set of combinations of proposed options that can be selected in a single bid) Ω is given by equation (4).

[0025]

number

[0026] The learning device 100 assumes that the set of bids Ω is known. The learning device 100 will perform reinforcement learning. Reinforcement learning, as used here, is a machine learning method that learns a policy—a set of rules for an agent's actions in a given environment—based on the state of the environment and a reward representing the evaluation of that state or action.

[0027] In learning by the learning device 100, the combination of the set of bids and the opposing agent device can be considered as the environment. Its own bids (bids made by the learning device 100) can be considered as actions, and the rules for generating its own bids can be considered as policies. Furthermore, the opposing agent device is assumed to have a state, and that state can be considered as a state in reinforcement learning.

[0028] Within the framework of the set of bids, the state of the opposing agent device transitions according to the bid made by the learning device 100. The opposing agent device makes a bid according to its state and the bid made by the learning device 100. Furthermore, the negotiation results can be considered as a reward. In learning by the learning device 100, a reward is given at the end of the negotiation according to the negotiation results.

[0029] As described above, one's own bid can be considered as an action, and the set of bids Ω can be considered as the set of actions that the learning device 100 can take. The set of actions A is shown in equation (5).

[0030]

number

[0031] |Ω| represents the number of elements in the set of bids Ω. Therefore, |Ω| represents the number of combinations of proposed options that can be selected in a single bid. Furthermore, the learning device 100 is assumed to be unaware of the state of the opposing agent device. Therefore, the learning device 100 represents the state in reinforcement learning as the history of the bids in the most recent past step. State s at time t t This is expressed as shown in equation (6).

[0032]

number

[0033] ω s t ω represents its own bid at time t. o t This indicates the opponent's bid at time t. Furthermore, as a reward, for example, the reward shown in equation (7) can be used.

[0034]

number

[0035] ω a ω indicates the agreed bid. For example, if an agreement was reached at the bid time t, then ω a =(ω s t ,ω o t ) can be done as follows. Function U s The agreed bid ω a Outputs the evaluation value for function U. s As, bid ω a Various functions can be used that output larger values ​​the better the evaluation of the value. However, the reward used by the learning device 100 is not limited to a specific one.

[0036] Policy function π θ This is shown in equation (8).

[0037]

number

[0038] Here, θ is the policy function π θ The parameters are shown below. s 1:t This shows the state history from time 1 to t. The policy function is π. θ Based on the state history from time 1 to t, its own bid ω at time t is determined. s t The policy function π is determined. θ Based on the bidding history from time 1 to t-1, its own bid ω at time t s t To decide. Policy function π θ This corresponds to a machine learning model. The parameter θ corresponds to the parameter to be learned.

[0039] The state transition function T is given by equation (9).

[0040]

number

[0041] π i,j This indicates the opponent's strategy (their negotiation strategy). As shown in equation (9), the bit ω at time t s t In contrast, probabilistically according to the opponent's negotiation strategy, the opponent's bid ω o t This is selected. However, the learning device 100 is assumed to be unaware of the state transition probability p. The learning device 100 can be understood as learning strategies for dealing with various negotiation opponents (negotiation strategies for determining its own bid) through learning.

[0042] Policy function π θ and state transition function T The expected value of the reward under the conditions E πθ,T This is shown in equation (10).

[0043]

number

[0044] The expected value of a reward is also called the expected reward. Here, if the reward is expressed as in equation (7), the learning device 100 may be unaware of the reward during the negotiation. Therefore, the learning device 100 uses the value function V shown in equation (11). θ The value shall be calculated using the following method.

[0045]

number

[0046] Here, θ represents the parameter of the value function. s 1:t This shows the state history from time 1 to time t. State history s from time 1 to time t 1:t This is shown in equation (12).

[0047]

number

[0048] As described above, the learning device 100 represents the state in reinforcement learning as the history of bids in the most recent past step. Therefore, the history of bids from time 1 to t is used as the state history from time 1 to t. State history s 1:t , bid history1:t It is also called by this name. Value function V θ This represents the value s for the history of the state from time 1 to t. 1:t We calculate the value (value of the value function) V at time t. θ (s 1:t This can be interpreted as an evaluation of the bid history from time 1 to t. The learning device 100 determines the value V at the end of negotiations. θ (s 1:t ) is the expected reward E at the end of negotiations. πθ,T (Σr t The value function V approximates ) θ Learn about this.

[0049] The communication unit 110 communicates with other devices. For example, the communication unit 110 sends its own bid to the other agent device. The communication unit 110 also receives the other agent device's bid.

[0050] The display unit 120 includes a display screen such as a liquid crystal panel or an LED (Light Emitting Diode) panel, and displays various images. For example, the display unit 120 may display various information related to negotiations, such as the bidding history. Alternatively, the display unit 120 may display various information related to learning, such as the number of negotiations performed, as an indicator of the learning progress.

[0051] The operation input unit 130 includes, for example, input devices such as a keyboard and a mouse, and accepts user operations. For example, the operation input unit 130 may be configured to accept user operations for various learning-related settings such as the learning rate and discount rate.

[0052] The memory unit 180 stores various types of data. The memory unit 180 is configured using the memory devices provided by the learning device 100. The domain information storage unit 181 stores information indicating the domain of negotiation. This information indicating the domain of negotiation is also called domain information. The domain of negotiation referred to here is the area in which negotiation is applied.

[0053] The domain information storage unit 181 stores a set of bids Ω as domain information. The number of proposed items, the number of proposed options per proposed item, and the values ​​of the proposed options (contents of the proposed options) may differ depending on the negotiation domain. Furthermore, by using the bid set Ω as domain information, it is expected that it will be possible to determine whether two negotiations are in similar domains (even the same domain) or different domains, without the need to explicitly distinguish the domains to which the negotiations belong.

[0054] The bid history storage unit 182 stores the bid history s at time t. 1:t Remember this.

[0055] The processing unit 190 controls various parts of the learning device 100 to perform various processes. The functions of the processing unit 190 are performed, for example, by the CPU (Central Processing Unit) of the learning device 100 reading a program from the storage unit 180 and executing it.

[0056] The domain embedding unit 191 converts information indicating the negotiation domain into a numerical vector (a vector whose elements are numerical values). The domain embedding unit 191 is an example of a domain embedding means. Specifically, the domain embedding unit 191 converts the set of bids Ω into a numerical vector. For example, the domain embedding unit 191 may convert the set of bids Ω into a numerical vector based on equation (13).

[0057]

number

[0058] Function F is a function that transforms the set of bids Ω into a numerical vector. However, the output of function F may be treated as a vector or as a set. That is, the elements of the output of function F may or may not be ordered. It is sufficient that the similarity between the outputs of function F can be calculated.

[0059] Function f is for proposed item I i and proposed option v i ci This is a function that converts each of these into a numerical value. Function f is for proposed item I i If this is entered, and suggested option v i ci In all cases where a value is input, the function f outputs a numerical value that uniquely identifies the input. In other words, function f maps the input to a numerical value via a one-to-one mapping. Function f is also called an embedding function. As function f, proposed item I i and proposed option v i ci In each case, various functions can be used to map the values ​​to numbers using a one-to-one mapping. In equation (13), the function F is f(I i ) and f(v i ci The linear sum is taken with ). The inventors of this application have found that a linear sum can be used as the function F, as shown in equation (13). Domain embedding section 191 is proposed item I i and proposed option v i ci The combination with f(I i )+f(v i ci By converting them to numerical values ​​using ), each bid included in the set of bids Ω is transformed into a numerical vector that can identify the bid, and in this way, the set of bids Ω can be seen as being transformed into a numerical vector.

[0060] The encoder unit 192 converts the numerical vector output by the domain embedding unit 191 into a numerical vector that takes attention into account. In particular, the encoder unit 192 accepts variable-length data input. This allows the learning device 100 to determine and learn its own bids in response to various cases regarding the number of proposed items and the number of proposed options. As a result, the learning device 100 can handle various domains. The encoder unit 192 may be an encoder from a known base model.

[0061] The bid history embedding unit 193 converts the bid history into a numerical vector using the function f. The bid history embedding unit 193 is an example of a bid embedding means. For example, the bid history embedding unit 193 may convert the bid history into a numerical vector based on equation (14).

[0062]

number

[0063] v c i;js This shows the proposed content for the i-th item in its own bid at time j. c i;js The 'c' in this case indicates the c-th proposed option among the proposed options. v c i;jo This shows the proposal for the i-th item in the opponent's bid at time j. c i;jo The 'c' in this case indicates the c-th proposed option among the proposed options.

[0064] The bid history embedding unit 193 uses the same method as the domain embedding unit 191 uses to convert bids into numerical vectors, to embed the bid history s 1:t Each bid contained in is converted into a numerical vector that can identify the bid, thereby creating the history of the bids. 1:t It can be seen as a process that converts a value into a numerical vector.

[0065] The decoder unit 194 performs a conversion on the numerical vector output by the bit history embedding unit 193 using the numerical vector output by the encoder unit 192. In particular, the decoder unit 194 accepts variable-length data as input. This allows the learning device 100 to use a history of bids of any length to determine its own bid. Specifically, the learning device 100 can use the history of all bids, both its own and the opponent's, from the start of negotiations to the present. In this respect, it is expected that the learning device 100 can learn how to determine its own bid with high accuracy and determine its own bid with high accuracy.

[0066] In this context, making a bid with high accuracy may also mean that the reward obtained is large (the reward indicates a good evaluation). Making a bid decision with high accuracy may also mean learning to make bids that result in a large reward (the reward indicates a good evaluation). The decoder unit 194 may be a decoder from a known substrate model.

[0067] The bid selection unit 195 uses the numerical vector output by the decoder unit 194 to determine its own bid (the bid it will present to the other party). Figure 2 shows an example of the configuration of the bid selection unit 195. In the configuration shown in Figure 2, the bid selection unit 195 comprises a linear processing unit 361 and a selection processing unit 362.

[0068] The linear processing unit 361 calculates an evaluation value for each candidate bid based on the numerical vector output by the decoder unit 194. The selection processing unit 362 selects the candidate with the best evaluation based on the evaluation value for each candidate bid output by the linear processing unit 361. Alternatively, the selection processing unit 362 may select the candidate with the largest evaluation value output by the linear processing unit 361 using the softmax function.

[0069] The process performed by the bid selection unit 195 can be regarded as the same as the process in the output layer of the neural network that performs class classification. The process performed by the linear processing unit 361 can be regarded as the process of calculating the likelihood for each bid candidate by taking the full connection of the element values of the numerical vector output by the decoder unit 194. However, the configuration of the bid selection unit 195 is not limited to a specific one. The combination of the encoder unit 192, the decoder unit 194, and the bid selection unit 195 corresponds to an example of bid determination means.

[0070] The value calculation unit 196 inputs the output of the decoder unit 194 into the value function V θ to calculate the value V θ (s 1:t ). As the value function V θ used by the value calculation unit 196, various value functions in known reinforcement learning can be used.

[0071] The learning processing unit 197 performs the learning of the policy function π θ and the learning of the value function v θ . The learning processing unit 197 corresponds to an example of learning processing means. Regarding the learning of the policy function π θ , the learning processing unit 197 performs the learning of the policy function π θ in reinforcement learning. The learning processing unit 197 may perform the learning of the policy function π θ using a known reinforcement learning method. The bid selection unit 195 can also be regarded as constituting the policy function π θ . Alternatively, the combination of the bid history embedding unit 193, the decoder unit 194, and the bid selection unit 195 can be regarded as constituting the policy function π θ .

[0072] Regarding the learning of the value function v θ , the learning processing unit 197 uses the value V θ (s 1:t ) at the end of negotiation, where the value Vπθ,T (Σr t The value function V approximates ) θ Learn the value function v. θ As the machine learning model and its learning method that constitute it, various types of machine learning models and learning methods can be used that can bring the output of the machine learning model closer to the value shown as the ground truth. For example, the value function v θ The system is constructed using a neural network (NN), and the learning processing unit 197 uses backpropagation to perform a value function v θ Learning may be conducted in this manner, but is not limited to this. The value calculation unit 196 calculates the value function V θ It can also be seen as constituting the value function V. Alternatively, the combination of the bid history embedding unit 193, the decoder unit 194, and the value calculation unit 196 can be seen as the value function V. θ It can also be seen as something that constitutes it.

[0073] Figure 3 shows an example of data input and output in the learning device 100. In the example shown in Figure 3, the domain embedding unit 191 reads the set of bids Ω from the domain information storage unit 181 and converts the read set of bids Ω into a numerical vector. The domain embedding unit 191 outputs the numerical vector obtained by converting the set of bids Ω to the encoder unit 192.

[0074] The encoder unit 192 converts the numerical vector output by the domain embedding unit 191 into a numerical vector that takes attention into account. The encoder unit 192 outputs the numerical vector that takes attention into account to the decoder unit 194. If the set of bids does not change during negotiations, the domain embedding unit 191 and the encoder unit 192 only need to perform processing once at the start of negotiations. If the set of bids may change during negotiations, the domain embedding unit 191 and the encoder unit 192 may perform processing at the start of negotiations and whenever the set of bids changes.

[0075] The bid history embedding unit 193 retrieves the bid history s from the bid history storage unit 182. 1:t Read the history of the read bits s 1:t Converts the bid history into a numerical vector. The bid history embedding unit 193 stores the bid history s 1:t The converted numerical vector is output to the decoder unit 194.

[0076] The decoder unit 194 performs a conversion on the numerical vector output by the bid history embedding unit 193 using the numerical vector output by the encoder unit 192. The decoder unit 194 outputs the converted numerical vector to the bid selection unit 195 and the value calculation unit 196.

[0077] The bid selection unit 195 selects one of the candidate bids based on the numerical vector output by the decoder unit 194. The bid selection unit 195 then transmits the selected bid to the other agent device 910 via the communication unit 110 as its own bid. The opposing agent device 910 receives its own bid, determines the opposing bid, and transmits the determined opposing bid to the learning device 100. In the learning device 100, the communication unit 110 receives the other party's bid. The processing unit 190 updates the bid history stored in the bid history storage unit 182 so that the combination of its own bid and the other party's bid in return is added to the bid history.

[0078] The value calculation unit 196 calculates a value that approximately represents the reward based on the numerical vector output by the decoder unit 194. The value calculated by the value calculation unit 196 is based on the bid history π 1:t This can be interpreted as an evaluation of [the subject].

[0079] Figure 4 shows an example of the configuration of a proposal generation device according to at least one embodiment. As shown in Figure 4, the proposal generation device 200 comprises a communication unit 110, a display unit 120, an operation input unit 130, a storage unit 180, and a processing unit 290. The storage unit 180 comprises a domain information storage unit 181 and a bid history storage unit 182. The processing unit 190 comprises a domain embedding unit 191, an encoder unit 192, a bid history embedding unit 193, a decoder unit 194, and a bid selection unit 195.

[0080] In Figure 4, parts that have the same function as the parts in Figure 1 are denoted by the same reference numerals (110, 120, 130, 180, 181, 182, 191, 192, 193, 194, 195), and detailed explanations are omitted here. In the proposal generation device 200, the processing unit 290 does not include the value calculation unit 196 and the learning processing unit 197, which are among the parts of the processing unit 190 of the learning device 100. In all other respects, the proposal generation device 200 is the same as the learning device 100.

[0081] The trained learning device 100 can be used as the proposal generator 200. The proposal generator 200 generates and outputs its own bids in the same way as the learning device 100. On the other hand, the proposal generator 200 uses the policy function π θ Learning and value function v θ We will not conduct that type of learning. Here, of the two negotiating parties, the proposal generation device 200 is referred to as "Self," and the other negotiating party is referred to as "Opponent."

[0082] Figure 5 shows an example of data input and output in the proposal generation device 200. Comparing the example in Figure 5 with the example in Figure 3, in the example in Figure 5, the proposal generation device 200 does not have a value calculation unit 196, and the decoder unit 194 outputs a numerical vector to the bid selection unit 195 but does not output to the value calculation unit 196. In all other respects, the example in Figure 5 is the same as the example in Figure 3.

[0083] Figure 6 shows the experimental results of the proposal generation device 200. A single learning device 100 was applied to multiple fields and multiple strategies of the opponent to perform training. Then, the trained learning device 100 was used as a proposal generation device 200 to negotiate for each of the five combinations of fields and five strategies of the opponent shown in Figure 3, and the negotiation results were evaluated. In Figure 3, the evaluation value for the negotiation result is shown as a real number within the range of 0 to 1. A higher evaluation value indicates a better evaluation. We obtained positive evaluation results for every combination of field and opponent's strategy.

[0084] As described above, the domain embedding unit 191 converts each of the bids that can be formed by combining one or more items designated as items to be proposed with one of the proposed options, which are the selectable proposed content for that item, into a numerical vector that can identify the bid.

[0085] The bid history embedding unit 193 converts each bid included in the bid history during negotiations into a numerical vector that can identify the bid. The combination of the encoder unit 192, the decoder unit 194, and the bid selection unit 195 determines which bid to propose to the other party from among all possible bids, using the numerical vectors obtained by converting each of the possible bids and the numerical vectors obtained by converting each bid included in the bid history.

[0086] The proposal generator 200 allows the same model to be used for negotiations in different fields. Specifically, the proposal generator 200 can determine a bid without needing to identify the domain (field of negotiation) and the negotiating partner's strategy, and one proposal generator 200 can be used for various domains and various strategies of negotiating partners. In particular, the proposal generator 200 can determine a bid even when the domain and the negotiating partner's strategy are unknown. Furthermore, the proposal generator 200 can determine a bid even for domains and negotiating partner strategies that have not been learned.

[0087] Furthermore, the domain embedding unit 191 inputs each of the proposed item and proposed option into an embedding function, which is a function that converts both the proposed item and the proposed option into numerical values ​​that can identify them, for each combination of proposed item and proposed option included in the bid. By taking a linear sum of the numerical values ​​obtained from the proposed item and the numerical values ​​obtained from the proposed option, the domain embedding unit 191 converts each of all possible bids into a numerical vector. The bid history embedding unit 193 converts each bid included in the bid history into a numerical vector using the same conversion method as the domain embedding unit 191 uses to convert bids into numerical vectors.

[0088] According to the proposal generation device 200, the computational load is expected to be relatively small because it performs a simple calculation of taking a linear sum between the numerical values ​​obtained by converting the proposed items and the numerical values ​​obtained by converting the proposed options.

[0089] Furthermore, the bid history embedding unit 193 converts each bid included in the bid history, which includes both bids from the proposal generation device 200 and bids from the negotiating partner, into a numerical vector.

[0090] According to the proposal generation device 200, bids can be determined based on the history of both bids made by the proposal generation device 200 and bids made by the negotiating partner. In this respect, the proposal generation device 200 is expected to be able to determine bids with relatively high accuracy. In this context, making a highly accurate bid decision may also mean receiving a large reward (i.e., a good evaluation indicated by the reward).

[0091] Furthermore, the domain embedding section 191 converts each of the one or more items designated as items to be proposed, and each of the possible bids that combine that item with one of the proposed options which are the selectable proposed content for that item, into a numerical vector that can identify the bid.

[0092] The bid history embedding unit 193 converts each bid included in the bid history during negotiations into a numerical vector that can identify the bid. The combination of the encoder unit 192, the decoder unit 194, and the bid selection unit 195 determines which bid to propose to the other party from among all possible bids, using the numerical vectors obtained by converting each of the possible bids and the numerical vectors obtained by converting each bid included in the bid history. The learning processing unit 197 learns the method for determining bids by combining the encoder unit 192, the decoder unit 194, and the bid selection unit 195.

[0093] The learning device 100 allows for learning using the same model across different negotiation fields, and enables the use of the same model across different negotiation fields. Specifically, the learning device 100 can learn how to make bids (proposals) without needing to identify the domain (area of ​​negotiation) and the strategy of the negotiating partner, and one learning device 100 can learn about various domains and various strategies of negotiating partners. Using a model trained by the learning device 100, bids can be determined even when the domain and the negotiating partner's strategy are unknown. Furthermore, using a model trained by the learning device 100, bids can be determined even for domains and negotiating partner strategies that have not been learned.

[0094] Furthermore, the domain embedding unit 191 inputs each of the proposed item and proposed option into an embedding function, which is a function that converts both the proposed item and the proposed option into numerical values ​​that can identify them, for each combination of proposed item and proposed option included in the bid. By taking a linear sum of the numerical values ​​obtained from the proposed item and the numerical values ​​obtained from the proposed option, the domain embedding unit 191 converts each of all possible bids into a numerical vector. The bid history embedding unit 193 converts each bid included in the bid history into a numerical vector using the same conversion method as the domain embedding unit 191 uses to convert bids into numerical vectors.

[0095] According to the learning device 100, the computational load is expected to be relatively small because it performs a simple calculation of taking a linear sum between the numerical values ​​obtained by converting the proposed target item and the numerical values ​​obtained by converting the proposed options.

[0096] Furthermore, the bid history embedding unit 193 converts each bid included in the bid history, which includes both bids made by the learning device 100 and bids made by the negotiating partner, into a numerical vector.

[0097] By using a model trained by the learning device 100, bids can be determined based on the history of both bids made by the learning device 100 and bids made by the negotiating partner. In this respect, it is expected that bid determinations can be made with relatively high accuracy by using a model trained by the learning device 100.

[0098] Both the learning device 100 and the proposal generation device 200 can be used to negotiate routes in the autonomous driving of mobile vehicles, such as autonomous driving of automobiles. The learning device 100 or the proposal generation device 200 may negotiate routes with the other mobile vehicle and yield routes to each other. The mobile vehicle may then automatically proceed along the route determined by the negotiation.

[0099] Both the learning device 100 and the suggestion generation device 200 can be used to control robots in warehouses and other similar environments. The learning device 100 or the suggestion generation device 200 may negotiate inventory adjustments in the manufacturing process and control the robots according to the determined inventory plan. Alternatively, the learning device 100 or the suggestion generation device 200 may negotiate shipping plans and control the robots according to the determined shipping plans.

[0100] The learning device 100 or the suggestion generation device 200 may interact with a person to adjust the schedule. For example, the learning device 100 or the suggestion generation device 200 may plan a flight or coordinate the date and time of a visit to a customer. Furthermore, the learning device 100 or the suggestion generation device 200 may coordinate the delivery date and time with the recipient of the package. In addition, the learning device 100 or the suggestion generation device 200 may determine the delivery plan for the package, including the delivery route and delivery time, based on the determined delivery date and time.

[0101] In both the learning device 100 and the proposal generation device 200, the negotiating partner may be a person, or it may be a system or device configured using a model such as a Large Language Model (LLM). By equipping the learning device 100 or the proposal generation device 200 with an interface (which may be a user interface or a communication interface) that is appropriate for the negotiating partner, it becomes possible to perform various adjustments with various negotiating partners such as people, robots, mobile devices, or artificial intelligence.

[0102] <Second Embodiment> Figure 7 shows an example of the configuration of a proposal generation device according to at least one embodiment. As shown in Figure 7, the proposal generation device 610 comprises a domain embedding unit 611, a bid history embedding unit 612, and a bid determination unit 613.

[0103] In this configuration, the domain embedding unit 611 converts each of the bids that can be formed by combining one or more items designated as items to be proposed with one of the proposed options, which are the selectable proposed content for that item, into a numerical vector that can identify the bid. The bid history embedding unit 612 converts each bid included in the bid history during negotiations into a numerical vector that can identify the bid.

[0104] The bid determination unit 613 uses the numerical vectors obtained by transforming each of the possible bids and the numerical vectors obtained by transforming each bid included in the bid history to determine which bid to propose to the other party from among all possible bids. The domain embedding unit 611 is an example of a domain embedding means. The bid history embedding unit 612 is an example of a bid history embedding means. The bid determination unit 613 is an example of a bid determination means.

[0105] The proposal generator 610 allows the same model to be used for negotiations in different fields. Specifically, the proposal generator 610 can determine a bid without needing to identify the domain (field of negotiation) and the negotiating partner's strategy, and one proposal generator 610 can be used for various domains and various strategies of negotiating partners. In particular, the proposal generator 610 can determine a bid even when the domain and the negotiating partner's strategy are unknown. Furthermore, the proposal generator 610 can determine a bid even for domains and negotiating partner strategies that have not been learned.

[0106] The domain embedding unit 611 can be implemented using, for example, the functions of the domain embedding unit 191 in Figure 4. The bid history embedding unit 612 can be implemented using, for example, the functions of the bid history embedding unit 193 in Figure 4. The bid determination unit 613 can be implemented using, for example, the functions of the encoder unit 192, decoder unit 194, and bid selection unit 195 in Figure 4.

[0107] <Third Embodiment> Figure 8 shows an example of the configuration of a learning device according to at least one embodiment. As shown in Figure 8, the learning device 620 comprises a domain embedding unit 621, a bid history embedding unit 622, a bid determination unit 623, and a learning processing unit 624.

[0108] In this configuration, the domain embedding unit 621 converts each of the bids that can be formed by combining one or more items designated as items to be proposed with one of the proposed options, which are the selectable proposed content for that item, into a numerical vector that can identify the bid. The bid history embedding unit 622 converts each bid included in the bid history during negotiations into a numerical vector that can identify the bid.

[0109] The bid determination unit 623 uses the numerical vectors obtained by transforming each of the possible bids and the numerical vectors obtained by transforming each bid included in the bid history to determine which bid to propose to the other party from among all possible bids. The learning processing unit 624 learns the bid determination method used by the bid determination unit 623. The domain embedding unit 621 is an example of a domain embedding means. The bid history embedding unit 622 is an example of a bid history embedding means. The bid determination unit 623 is an example of a bid determination means. The learning processing unit 624 is an example of a learning processing means.

[0110] The learning device 620 allows for learning using the same model across different negotiation domains, and enables the use of the same model across different negotiation domains. Specifically, the learning device 620 can learn how to make bids (proposals) without needing to identify the domain (area of ​​negotiation) and the strategy of the negotiating partner, and a single learning device 620 can learn about various domains and various strategies of negotiating partners. Using a model trained by the learning device 620, bids can be determined even when the domain and the negotiating partner's strategy are unknown. Furthermore, using a model trained by the learning device 620, bids can be determined even for domains and negotiating partner strategies that have not been learned.

[0111] The domain embedding unit 621 can be implemented using, for example, the domain embedding unit 191 in Figure 1. The bid history embedding unit 622 can be implemented using, for example, the bid history embedding unit 193 in Figure 1. The bid determination unit 623 can be implemented using, for example, the encoder unit 192, decoder unit 194, and bid selection unit 195 in Figure 1. The learning processing unit 624 can be implemented using, for example, the learning processing unit 197 in Figure 1.

[0112] <Fourth Embodiment> Figure 9 shows an example of the processing steps in a proposal generation method according to at least one embodiment. The proposal generation method shown in Figure 9 includes embedding a domain (step S611), embedding a history of bids (step S612), and determining a bid (step S613).

[0113] In embedding the domain (step S611), the computer converts each of the possible bids, for each of the one or more items designated as items to be proposed, that combine that item with one of the proposed options which are selectable proposed contents for that item, into a numerical vector that can identify the bid.

[0114] In embedding the bidding history (step S612), the computer converts each bid included in the bidding history of the negotiation into a numerical vector that can identify the bid. In determining the bid (step S613), the computer uses the numerical vectors obtained by converting each of the possible bids and the numerical vectors obtained by converting each bid included in the bid history to determine which bid to propose to the other party from among all possible bids.

[0115] The proposal generation method shown in Figure 9 allows the same model to be used for negotiations in different fields. Specifically, the proposal generation method shown in Figure 9 can determine bids without needing to identify the domain (area of ​​negotiation) and the negotiating partner's strategy, and one computer can be used for various domains and various strategies of negotiating partners. In particular, the proposal generation method shown in Figure 9 can determine bids even when the domain and the negotiating partner's strategy are unknown. Furthermore, the proposal generation method shown in Figure 9 can determine bids even for domains and negotiating partner strategies that have not been trained.

[0116] <Fifth Embodiment> Figure 10 shows an example of the processing steps in a learning method according to at least one embodiment. The learning method shown in Figure 10 includes embedding a domain (step S621), embedding a history of bids (step S622), determining a bid (step S623), and performing learning (step S624).

[0117] In embedding the domain (step S621), the computer converts each of the possible bids, for each of the one or more items designated as items to be proposed, that combine that item with one of the proposed options which are the selectable proposed contents for that item, into a numerical vector that can identify the bid. In embedding the bidding history (step S622), the computer converts each bid included in the bidding history of the negotiation into a numerical vector that can identify the bid.

[0118] In determining the bid (step S623), the computer uses the numerical vectors obtained by converting each of the possible bids and the numerical vectors obtained by converting each bid included in the bid history to determine which bid to propose to the opponent from among all possible bids. In the learning step (S624), the computer learns how to determine bids.

[0119] The learning method shown in Figure 10 allows for the use of the same model for negotiations in different fields. Specifically, the learning method shown in Figure 10 allows for learning bid decisions without needing to identify the domain (area of ​​negotiation) and the negotiating partner's strategy, enabling a single computer to learn about various domains and various strategies of negotiating partners. Using a model trained with the learning method shown in Figure 10, bids can be determined even when the domain and negotiating partner's strategy are unknown. Furthermore, using a model trained with the learning method shown in Figure 10, bids can be determined even for domains and negotiating partner strategies that have not been previously learned.

[0120] Figure 11 is a schematic block diagram showing the configuration of a computer according to at least one embodiment. In the configuration shown in Figure 11, the computer 700 comprises a CPU 710, a main memory 720, an auxiliary memory 730, an interface 740, and a non-volatile recording medium 750.

[0121] One or more of the above-mentioned learning device 100, proposal generation device 200, proposal generation device 610, and learning device 620, or a part thereof, may be implemented in the computer 700. In that case, the operation of each processing unit described above is stored in the auxiliary storage device 730 in the form of a program. The CPU 710 reads the program from the auxiliary storage device 730, expands it in the main memory 720, and executes the above processing according to the program. The CPU 710 also reserves memory areas in the main memory 720 corresponding to each of the above-mentioned storage units according to the program. Communication between each device and other devices is performed by the interface 740 having a communication function and performing communication according to the control of the CPU 710. The interface 740 also has a port for the non-volatile recording medium 750 and reads information from the non-volatile recording medium 750 and writes information to the non-volatile recording medium 750.

[0122] When the learning device 100 is implemented in the computer 700, the operation of the processing unit 190 and each of its parts is stored in the auxiliary storage device 730 in the form of a program. The CPU 710 reads the program from the auxiliary storage device 730, loads it into the main memory 720, and executes the above processing according to the program.

[0123] Furthermore, the CPU 710 reserves a memory area for the memory unit 180 in the main memory 720 according to the program. Communication with other devices by the communication unit 110 is performed by the interface 740 having a communication function and operating under the control of the CPU 710. Display of images by the display unit 120 is performed by the interface 740 having a display device and displaying various images under the control of the CPU 710. Acceptance of user operations by the operation input unit 130 is performed by the interface 740 having an input device and accepting user operations under the control of the CPU 710.

[0124] When the proposal generation device 200 is implemented in the computer 700, the operation of the processing unit 290 and each of its parts is stored in auxiliary storage device 730 in the form of a program. The CPU 710 reads the program from the auxiliary storage device 730, expands it into the main memory 720, and executes the above processing according to the program.

[0125] Furthermore, the CPU 710 reserves a memory area for the memory unit 180 in the main memory 720 according to the program. Communication with other devices by the communication unit 110 is performed by the interface 740 having a communication function and operating under the control of the CPU 710. Display of images by the display unit 120 is performed by the interface 740 having a display device and displaying various images under the control of the CPU 710. Acceptance of user operations by the operation input unit 130 is performed by the interface 740 having an input device and accepting user operations under the control of the CPU 710.

[0126] When the proposal generation device 610 is implemented in the computer 700, the operations of the domain embedding unit 611, the bid history embedding unit 612, and the bid determination unit 613 are stored in auxiliary storage device 730 in the form of a program. The CPU 710 reads the program from the auxiliary storage device 730, expands it into the main memory 720, and executes the above processes according to the program.

[0127] Furthermore, the CPU 710 reserves memory in the main memory 720 for the proposal generation device 610 to process according to the program. Communication between the proposal generation device 610 and other devices is performed by the interface 740 having a communication function and operating under the control of the CPU 710. Interaction between the proposal generation device 610 and the user is performed by the interface 740 having input and output devices, presenting information to the user via the output device and accepting user operations via the input device under the control of the CPU 710.

[0128] When the learning device 620 is implemented in the computer 700, the operations of the domain embedding unit 621, the bid history embedding unit 622, the bid determination unit 623, and the learning processing unit 624 are stored in auxiliary storage device 730 in the form of a program. The CPU 710 reads the program from auxiliary storage device 730, expands it into main memory device 720, and executes the above processes according to the program.

[0129] Furthermore, the CPU 710 allocates memory in the main memory 720 for the learning device 620 to process according to the program. Communication between the learning device 620 and other devices is performed by the interface 740 having a communication function and operating under the control of the CPU 710. Interaction between the learning device 620 and the user is performed by the interface 740 having input and output devices, presenting information to the user via the output device and accepting user input via the input device under the control of the CPU 710.

[0130] One or more of the above-mentioned programs may be recorded on the non-volatile recording medium 750. In this case, the interface 740 may read the program from the non-volatile recording medium 750. The CPU 710 may then either directly execute the program read by the interface 740, or temporarily save it in the main memory 720 or auxiliary memory 730 before executing it.

[0131] Alternatively, programs for executing all or part of the processing performed by the learning device 100, the proposal generation device 200, the proposal generation device 610, and the learning device 620 may be recorded on a computer-readable recording medium, and the programs recorded on this recording medium may be loaded into a computer system and executed to perform the processing of each part. The term "computer system" here includes hardware such as an operating system (OS) and peripheral devices. Furthermore, "computer-readable recording media" refers to portable media such as flexible disks, magneto-optical disks, ROMs (Read Only Memory), CD-ROMs (Compact Disc Read Only Memory), and storage devices such as hard disks built into computer systems. The above-mentioned program may be intended to implement only a part of the functions described above, and may also be able to implement the above-mentioned functions in combination with programs already recorded in the computer system.

[0132] Although the present disclosure has been described above with reference to embodiments, the present disclosure is not limited to the embodiments described above. Various modifications to the structure and details of the present disclosure are possible, as can be understood by those skilled in the art within the scope of the present disclosure. Furthermore, the embodiments described above may be combined with other embodiments as appropriate.

[0133] Some or all of the above embodiments may also be described as follows, but are not limited to the following:

[0134] (Note 1) For each of the one or more items designated as items subject to proposal, a domain embedding means converts each of all possible bids, which are combinations of that item and one of the proposal options that are selectable for that item, into a numerical vector that can identify the bid, A bid history embedding means that converts each bid included in the bid history in negotiations into a numerical vector that can identify the bid, A bid determination means that determines which bid to propose to the other party from among all possible bids, using the numerical vector obtained by transforming each of the possible bids and the numerical vector obtained by transforming each bid included in the history of the bids, A proposal generation device equipped with the following features.

[0135] (Note 2) The domain embedding means inputs each of the proposed item and proposed option included in the bid into a function that converts both the proposed item and the proposed option into numerical values ​​that can identify them, and then takes a linear sum of the numerical values ​​obtained from the proposed item and the numerical values ​​obtained from the proposed option to convert each of the possible bids into a numerical vector. The bid history embedding means converts each bid included in the bid history into a numerical vector using the same conversion method as the domain embedding means for converting bids into numerical vectors. The proposal generation device described in Appendix 1.

[0136] (Note 3) The bid history embedding means converts each bid included in the bid history, which includes both bids from the proposal generation device and bids from the negotiating partner, into a numerical vector. The proposed generation device as described in Appendix 1 or Appendix 2.

[0137] (Note 4) For each of the one or more items designated as items subject to proposal, a domain embedding means converts each of all possible bids, which are combinations of that item and one of the proposal options that are selectable for that item, into a numerical vector that can identify the bid, A bid history embedding means that converts each bid included in the bid history in negotiations into a numerical vector that can identify the bid, A bid determination means that determines which bid to propose to the other party from among all possible bids, using the numerical vector obtained by transforming each of the possible bids and the numerical vector obtained by transforming each bid included in the history of the bids, A learning processing means for learning the method of determining bids by the bid determination means, A learning device equipped with the following features.

[0138] (Note 5) The domain embedding means inputs each of the proposed item and proposed option included in the bid into a function that converts both the proposed item and the proposed option into numerical values ​​that can identify them, and then takes a linear sum of the numerical values ​​obtained from the proposed item and the numerical values ​​obtained from the proposed option to convert each of the possible bids into a numerical vector. The bid history embedding means converts each bid included in the bid history into a numerical vector using the same conversion method as the domain embedding means for converting bids into numerical vectors. The learning device described in Appendix 4.

[0139] (Note 6) The bid history embedding means converts each bid included in the bid history, which includes both bids made by the learning device and bids made by the negotiating partner, into a numerical vector. A learning device as described in Appendix 4 or Appendix 5.

[0140] (Note 7) Computers For each of the one or more items designated as items to be proposed, all possible bids that combine that item with one of the proposed options (which are the selectable proposed contents for that item) are converted into numerical vectors that allow for identifiable bids. Each bid included in the bidding history during negotiations is converted into a numerical vector that allows for bid identification. Using the numerical vectors obtained by transforming each of the possible bids, and the numerical vectors obtained by transforming each bid included in the history of the bids, the bid to propose to the other party is determined from among all possible bids. A proposal generation method that includes the following.

[0141] (Note 8) Converting each of the possible bids into a numerical vector is done by the computer, for each combination of proposed item and proposed option included in the bid, inputting each of the proposed item and proposed option into a function that converts both the proposed item and the proposed option into numerical values ​​that can identify them, and then taking a linear sum of the numerical values ​​obtained from the converted proposed item and the numerical values ​​obtained from the converted proposed option, thereby converting each of the possible bids into a numerical vector. Converting each bid included in the bidding history in the negotiation into a numerical vector means that the computer converts each bid included in the bidding history into a numerical vector using the same conversion method as when converting each of the bids that could be considered as bids into a numerical vector. The proposal generation method described in Appendix 7.

[0142] (Note 9) Converting each bid included in the bidding history in the aforementioned negotiation into a numerical vector means that the computer converts each bid included in the bidding history, which includes both bids made by the computer itself and bids made by the negotiating party, into a numerical vector. The proposal generation method described in Appendix 7 or Appendix 8.

[0143] (Note 10) Computers For each of the one or more items designated as items to be proposed, all possible bids that combine that item with one of the proposed options (which are the selectable proposed contents for that item) are converted into numerical vectors that allow for identifiable bids. Each bid included in the bidding history during negotiations is converted into a numerical vector that allows for bid identification. Using the numerical vectors obtained by transforming each of the possible bids, and the numerical vectors obtained by transforming each bid included in the history of the bid, the bid to propose to the other party is determined from among all possible bids. Learn the method for determining the bid. A learning method that includes this.

[0144] (Note 11) Converting each of the possible bids into a numerical vector is done by the computer, for each combination of proposed item and proposed option included in the bid, inputting each of the proposed item and proposed option into a function that converts both the proposed item and the proposed option into numerical values ​​that can identify them, and then taking a linear sum of the numerical values ​​obtained from the converted proposed item and the numerical values ​​obtained from the converted proposed option, thereby converting each of the possible bids into a numerical vector. Converting each bid included in the bidding history in the negotiation into a numerical vector means that the computer converts each bid included in the bidding history into a numerical vector using the same conversion method as when converting each of the bids that could be considered as bids into a numerical vector. The learning method is described in Appendix 10.

[0145] (Note 12) Converting each bid included in the bidding history in the aforementioned negotiation into a numerical vector means that the computer converts each bid included in the bidding history, which includes both bids made by the computer itself and bids made by the negotiating party, into a numerical vector. The learning method described in Appendix 10 or Appendix 11.

[0146] (Note 13) On the computer, For each of the one or more items designated as items subject to proposal, all possible bids that combine that item with one of the proposal options (which are selectable proposal options for that item) are converted into numerical vectors that allow for identifiable bids. Converting each bid in the bidding history during negotiations into a numerical vector that allows for bid identification, Using the numerical vectors obtained by transforming each of the possible bids, and the numerical vectors obtained by transforming each bid included in the history of the bid, the bid to be proposed to the other party from among all possible bids, A program that executes the command.

[0147] (Note 14) Converting each of the possible bids into a numerical vector involves causing the computer to input each of the proposed item and proposed option into a function that converts them into numerical values ​​that can identify both the proposed item and the proposed option, for each combination of proposed item and proposed option included in the bid, and then perform a linear combination of the numerical values ​​obtained by the proposed item and the numerical values ​​obtained by the proposed option to convert each of the possible bids into a numerical vector. Converting each bid included in the bidding history in the aforementioned negotiation into a numerical vector involves causing the computer to perform the conversion of each bid included in the bidding history into a numerical vector using the same conversion method as when converting each of the possible bids into a numerical vector. The program described in Appendix 13.

[0148] (Note 15) Converting each bid included in the bidding history in the aforementioned negotiation into a numerical vector means that the computer converts each bid included in the bidding history, which includes both bids made by the computer itself and bids made by the negotiating party, into a numerical vector. The program described in Appendix 13 or Appendix 14.

[0149] (Note 16) On the computer, For each of the one or more items designated as items subject to proposal, all possible bids that combine that item with one of the proposal options (which are selectable proposal options for that item) are converted into numerical vectors that allow for identifiable bids. Converting each bid in the bidding history during negotiations into a numerical vector that allows for bid identification, Using the numerical vectors obtained by transforming each of the possible bids, and the numerical vectors obtained by transforming each bid included in the history of the bid, the bid to be proposed to the other party from among all possible bids, The method for determining the bid is to be learned, A program that executes the command.

[0150] (Note 17) Converting each of the possible bids into a numerical vector involves causing the computer to input each of the proposed item and proposed option into a function that converts them into numerical values ​​that can identify both the proposed item and the proposed option, for each combination of proposed item and proposed option included in the bid, and then perform a linear combination of the numerical values ​​obtained by the proposed item and the numerical values ​​obtained by the proposed option to convert each of the possible bids into a numerical vector. Converting each bid included in the bidding history in the aforementioned negotiation into a numerical vector involves causing the computer to perform the conversion of each bid included in the bidding history into a numerical vector using the same conversion method as when converting each of the possible bids into a numerical vector. The program described in Appendix 16.

[0151] (Note 18) Converting each bid included in the bidding history in the aforementioned negotiation into a numerical vector involves causing the computer to perform the conversion of each bid included in the bidding history, which includes both bids made by the computer itself and bids made by the negotiating party, into a numerical vector. The program described in Appendix 16 or Appendix 17. [Explanation of symbols]

[0152] 100, 620 Learning devices 110 Communications Department 120 Display section 130 Operation Input Section 180 Storage section 190, 290 Processing Unit 191, 611, 621 Domain embedding section 192 Encoder section 193, 612, 622 Bit history embedding section 194 Decoder section 195 Bit Selection Section 196 Value Calculation Unit 197,624 Learning Processing Unit 200, 610 Proposal generator 361 Linear Processing Unit 362 Selection Processing Unit 613, 623 Bid determination section

Claims

1. For each of the one or more items designated as items subject to proposal, a domain embedding means converts each of all possible bids, which are combinations of that item and one of the proposal options that are selectable for that item, into a numerical vector that can identify the bid, A bid history embedding means that converts each bid included in the bid history in negotiations into a numerical vector that can identify the bid, A bid determination means that determines which bid to propose to the other party from among all possible bids, using the numerical vector obtained by transforming each of the possible bids and the numerical vector obtained by transforming each bid included in the history of the bids, A proposal generation device equipped with the following features.

2. The domain embedding means inputs each of the proposed item and proposed option included in the bid into a function that converts both the proposed item and the proposed option into numerical values ​​that can identify them, and then takes a linear sum of the numerical values ​​obtained from the proposed item and the numerical values ​​obtained from the proposed option to convert each of the possible bids into a numerical vector. The bid history embedding means converts each bid included in the bid history into a numerical vector using the same conversion method as the domain embedding means for converting bids into numerical vectors. The proposed generation apparatus according to claim 1.

3. The bid history embedding means converts each bid included in the bid history, which includes both bids from the proposal generation device and bids from the negotiating partner, into a numerical vector. The proposed generation apparatus according to claim 1 or claim 2.

4. For each of the one or more items designated as items subject to proposal, a domain embedding means converts each of all possible bids, which are combinations of that item and one of the proposal options that are selectable for that item, into a numerical vector that can identify the bid, A bid history embedding means that converts each bid included in the bid history in negotiations into a numerical vector that can identify the bid, A bid determination means that determines which bid to propose to the other party from among all possible bids, using the numerical vector obtained by transforming each of the possible bids and the numerical vector obtained by transforming each bid included in the history of the bids, A learning processing means for learning the method of determining bids by the bid determination means, A learning device equipped with the following features.

5. Computers For each of the one or more items designated as items to be proposed, all possible bids that combine that item with one of the proposed options (which are the selectable proposed contents for that item) are converted into numerical vectors that allow for identifiable bids. Each bid included in the bidding history during negotiations is converted into a numerical vector that allows for bid identification. Using the numerical vectors obtained by transforming each of the possible bids, and the numerical vectors obtained by transforming each bid included in the history of the bids, the bid to propose to the other party is determined from among all possible bids. A proposal generation method that includes the following.

6. Computers For each of the one or more items designated as items to be proposed, all possible bids that combine that item with one of the proposed options (which are the selectable proposed contents for that item) are converted into numerical vectors that allow for identifiable bids. Each bid included in the bidding history during negotiations is converted into a numerical vector that allows for bid identification. Using the numerical vectors obtained by transforming each of the possible bids, and the numerical vectors obtained by transforming each bid included in the history of the bid, the bid to propose to the other party is determined from among all possible bids. Learn the method for determining the bid. A learning method that includes this.

7. On the computer, For each of the one or more items designated as items subject to proposal, all possible bids that combine that item with one of the proposal options (which are selectable proposal options for that item) are converted into numerical vectors that allow for identifiable bids. Converting each bid in the bidding history during negotiations into a numerical vector that allows for bid identification, Using the numerical vectors obtained by transforming each of the possible bids, and the numerical vectors obtained by transforming each bid included in the history of the bid, the bid to be proposed to the other party from among all possible bids, A program that executes the command.

8. On the computer, For each of the one or more items designated as items subject to proposal, all possible bids that combine that item with one of the proposal options (which are selectable proposal options for that item) are converted into numerical vectors that allow for identifiable bids. Converting each bid in the bidding history during negotiations into a numerical vector that allows for bid identification, Using the numerical vectors obtained by transforming each of the possible bids, and the numerical vectors obtained by transforming each bid included in the history of the bid, the bid to be proposed to the other party from among all possible bids, The method for determining the bid is to be learned, A program that executes the command.

Images