Policy proposal generation system, policy proposal generation method, and policy proposal generation program

Abstract

We provide a policy proposal generation system that can generate policy proposals that reflect suggestions for policy issues. [Solution] The policy proposal generation system of the present invention includes a processor and a memory device, wherein the processor generates policy proposal data, and the processor comprises: a proposal data acquisition unit that acquires proposal data for a predetermined policy issue; an evaluation unit that evaluates the proposal data based on evaluation data associated with the proposal data; a weighting unit that weights the proposal data based on the evaluation; and a generation unit that generates policy proposal data from the proposal data based on the weighting.

Description

Technical Field

[0001] The present invention relates to a policy proposal generation system, a policy proposal generation method, and a policy proposal generation program, and particularly to a policy proposal generation system, a policy proposal generation method, and a policy proposal generation program that generate policy proposals from proposals for policy issues.

Background Art

[0002] There has been disclosed a proposal system by residents for an electronic bulletin board and a candidate navigator system that can provide a place for communication regarding policies, improve interest in elections, and improve the voting rate (Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, there is no system that presents policy issues, collects proposals for the policy issues, evaluates the proposals, and generates policy proposals from the proposals based on the evaluation.

Means for Solving the Problems

[0005] The policy proposal generation system of the present invention includes a processor and a storage device, and is a policy proposal generation system in which the processor generates policy proposal data. The processor includes a proposal data acquisition unit that acquires proposal data for a predetermined policy issue, an evaluation unit that evaluates the proposal data based on evaluation data associated with the proposal data, a weighting unit that weights the proposal data based on the evaluation, and a generation unit that generates the policy proposal data from the proposal data based on the weighting.

Effects of the Invention

[0006] According to the present invention, it is possible to generate policy proposals that reflect suggestions for policy issues. [Brief explanation of the drawing]

[0007] [Figure 1] This is a block diagram showing an example of the system configuration of the policy proposal generation system of this embodiment. [Figure 2] This block diagram shows an example configuration of the machine learning computer of this embodiment. [Figure 3] This block diagram shows an example configuration of the learning database server in this embodiment. [Figure 4] This is a block diagram showing an example configuration of the proposed / evaluation computer of this embodiment. [Figure 5] This diagram shows an example of a screen displayed on the proposed / evaluation computer's screen. [Figure 6] This figure shows an example of machine learning data, where proposed training data, evaluation data, weight data, and other feature data are associated. [Figure 7] This figure shows an example of machine learning data where policy proposal data for training, evaluation data, weight data, and other feature data are associated. [Figure 8] This flowchart shows an example of the inference method of this embodiment. [Figure 9] This sequence diagram illustrates an example of the operations involved in acquiring and preprocessing various types of data, as well as generating and storing trained models. [Figure 10] This is a sequence diagram illustrating an example of the operation of inference using a trained model and the determination of the inference result. [Modes for carrying out the invention]

[0008] An embodiment of the policy proposal generation system of the present invention will be described with reference to the drawings. Figure 1 is a block diagram showing an example of the system configuration of the policy proposal generation system of this embodiment. The policy proposal generation system 1 includes a processor and a storage device, and the processor generates policy proposal data.

[0009] As shown in Figure 1, the policy proposal generation system 1 comprises machine learning computers (PCs) 3, a learning database server 4, and a proposal / evaluation computer (PC) 5, all electrically connected and able to communicate with each other via a network 2.

[0010] Figure 2 is a block diagram showing an example configuration of the machine learning computer of this embodiment. As shown in Figure 2, the machine learning computer 3 includes a processor 21, a storage device (e.g., ROM, RAM, HDD, etc.) 22, an input device 23, an interface 24, and an output device 25.

[0011] The processor 21 is a control device such as a CPU, MPU, or GPU, and includes a proposed data acquisition unit 260, an evaluation unit 261, a weighting unit 262, a generation unit 263, a machine learning unit 27, and a trained model storage unit 28. The proposed data acquisition unit 260, the evaluation unit 261, the weighting unit 262, the generation unit 263, the machine learning unit 27, and the trained model storage unit 28 are electrically connected by a bus (not shown) and can communicate with each other.

[0012] The proposal data acquisition unit 260 acquires proposal data for a predetermined policy issue (policy issue data). The evaluation unit 261 evaluates the proposal data based on the evaluation data associated with the proposal data. The weighting unit 262 weights the proposal data based on the evaluation. The generation unit 263 generates policy proposal data from the proposal data based on the weighting.

[0013] Also, when policy plan data is generated, the evaluation unit 261 evaluates the policy plan data based on the evaluation data associated with the policy plan data. The weighting unit 262 weights the policy plan data based on the evaluation of the policy plan. The generation unit 263 generates policy issue data from the policy plan data based on the weighting.

[0014] The machine learning unit 27 generates a learned model that infers policy plan data or policy issue data by machine learning based on the proposal data and the weighting.

[0015] The input device (input unit) 23 inputs feature amount data for inference into the learned model in order to infer policy plan data or policy issue data.

[0016] FIG. 3 is a block diagram showing a configuration example of the learning database server of the present embodiment. As shown in FIG. 3, the learning database server 4 includes a processor 30, a learning database 40, a storage device (for example, ROM, RAM, HDD, etc.) 31, an input device 32, an interface 33, and an output device 34. The processor 30 is a control device such as a CPU, MPU, or GPU, and includes a data storage unit 35, a data acquisition unit 36, and a preprocessing unit 37.

[0017] The learning database 40 includes a feature amount database 41. The feature amount database 41 includes a proposal database 43, an evaluation database 44, a weight database 45, an evaluator attribute database 46, a policy plan database 48, a logical database 49, and a time database 50. The proposal database 43 includes a first proposal database 431 and a second proposal database 432.

[0018] The learning proposal data (including the first proposal data and the second proposal data) is associated with the evaluation data for the proposal data. The proposal data (including the first proposal data and the second proposal data) is associated with the weight data of the proposal data weighted based on the evaluation data.

[0019] The evaluation data is associated with the evaluator attribute data of the evaluator who performed the evaluation. The policy proposal data is associated with the evaluation data of the policy proposal data. This expands the feature data used for training. The evaluator attribute data includes the evaluator's age, gender, address / region of residence, occupation, annual income, education level, household, political party affiliation, political ideology / ideology, and at least one of the test scores, such as for critical reasoning or mathematics.

[0020] The proposal data (including the first and second proposal data) and the policy proposal data are associated with logical data. For example, as used in argument mining, the logical database 49 includes a logical structure in which the parts constituting claims, premise, support, and attack are extracted from the argumentative discourse of the proposal data (including the first and second proposal data) and the policy proposal data, and the relationships between them are structured. This logical structure may be the logical structure of the proposal data and the policy proposal data individually, the logical structure of the first and second proposal data, or the logical structure of the proposal data and the policy proposal data. This expands the feature data used for training.

[0021] The time database 50 includes the time when proposal data and policy proposal data were generated or acquired, the time when evaluation data was generated or acquired, and the time when evaluator attribute data was generated or acquired. This expands the feature data used for training.

[0022] The training data stored in the training database 40 is the data used for machine learning by the machine learning unit 27 of the machine learning computer 3.

[0023] Figure 4 is a block diagram showing an example configuration of the proposal / evaluation computer of this embodiment. Figure 5 is a diagram showing an example of a screen displayed on the display of the proposal / evaluation computer. As shown in Figure 4, the proposal / evaluation computer 5 comprises a processor 501, a storage device (e.g., ROM, RAM, HDD, etc.) 522, an input device 523, an interface 524, and an output device 525. The processor 501 is a control device such as a CPU, MPU, or GPU, and comprises a policy issue data acquisition unit 502, a proposal data input unit 503, an evaluation data input unit 504, and an evaluator attribute data input unit 505.

[0024] The policy issue data acquisition unit 502 acquires policy issue data from the learning database 40 via the interfaces (transmitting / receiving units) 524 and 33, and displays the policy issue data on the display of the output device 525.

[0025] For example, policy issues are displayed in the policy issue display section 81 of screen 80 in Figure 5. In addition, detailed information about the policy issue (background, current problems, risks and impacts, and the need for solutions, etc.) is displayed in the detailed information display section 82, and each detailed information (text, charts, videos, etc.) is displayed when each icon is selected with a cursor, etc. If there is already proposal data for the policy issue, proposal data 83-1, 83-2, and 83-3 are displayed in the proposal display section 83. Here, the proposal data being cited is referred to as the first proposal data, and the proposal data citing is referred to as the second proposal data. For example, if proposal data 83-1 is the first proposal data, then proposal data 83-2 becomes the second proposal data. Note that the citation relationship is just one example of logical data, and the first and second proposal data may be associated based on at least one of the following: policy issue data (commonality of policy issues, etc.), logical data (affirmative / negative, etc.), time data (within a specified time, etc.), evaluation data (evaluation score above a certain level, etc.), and evaluator attribute data (age, occupation, test score, etc.).

[0026] Furthermore, the proposal display unit 83 in Figure 5 displays proposal data 83-1, 83-2, and 83-3, along with their respective proposal ID data, time data, proposer attribute data, and evaluation data.

[0027] The proposal / evaluation computer 5 is operated by the operator, and the proposal data input unit 503 inputs or modifies proposal data for policy issues. The proposal data input by the proposal data input unit 503 is stored as proposal data in the evaluation database 43 of the learning database 40 via the interfaces (transmitting / receiving units) 524 and 33. For example, in Figure 5, when the proposal input icon on the input display unit 84 is selected with a cursor, a screen for inputting proposal data is displayed, and the proposal data input unit 503 inputs or modifies proposal data for policy issues.

[0028] The evaluation data input unit 504 inputs or modifies evaluations of proposal data proposed by other proposers. For example, the evaluation data input unit 504 inputs or modifies evaluation data for proposal data using an ordinal scale for multiple evaluation items (e.g., specificity, measurability, motivation, feasibility, timeliness, effectiveness, and fairness). The evaluation data input by the evaluation data input unit 504 is stored as evaluation data in the evaluation database 44 of the learning database 40 via the interfaces (transmitting and receiving units) 524 and 33.

[0029] Furthermore, the evaluation data input unit 504 inputs or modifies the evaluation of the policy proposal data. For example, the evaluation data input unit 504 inputs or modifies the evaluation of the policy proposal data using an ordinal scale for multiple evaluation items (e.g., specificity, measurability, motivation, feasibility, timeliness, effectiveness, and fairness). The evaluation data input by the evaluation data input unit 504 is stored as evaluation data in the evaluation database 44 of the learning database 40 via the interfaces (transmitting and receiving units) 524 and 33.

[0030] In Figure 5, when the evaluation input icon in the input display unit 84 is selected with a cursor, a screen for entering evaluation data is displayed, and the evaluation data input unit 504 inputs or modifies the evaluation data for the proposed data or policy draft data. Also in Figure 5, when the icon in the policy draft display unit 87 is selected with a cursor, the policy draft data generated by the generation unit 263 of the machine learning computer 3 is displayed.

[0031] The evaluator attribute data input unit 505 inputs or modifies the evaluator's attributes. For example, the evaluator attribute data input unit 505 inputs or modifies information regarding the evaluator's age, gender, address / residential area, occupation, annual income, educational background, household, political party affiliation, political ideology / ideology, and test scores such as critical reasoning and mathematics. The evaluator attribute data input by the evaluator attribute data input unit 505 is stored as evaluator attribute data in the evaluator attribute database 46 of the learning database 40 via the interfaces (transmitting / receiving units) 524 and 33.

[0032] In Figure 5, selecting the attribute input icon on the input display unit 84 with a cursor displays a screen for entering evaluator attribute data, and the evaluator attributes are entered or modified by the evaluator attribute data input unit 505. Also in Figure 5, selecting the upload icon on the input display unit 84 with a cursor allows users to upload data such as documents and files, which can then be stored in the learning database 40. Furthermore, in Figure 5, the number of proposers is displayed in the proposer count display unit 85, and the number of evaluators is displayed in the evaluator count display unit 86.

[0033] Figure 6 shows an example of machine learning data in which proposed training data, evaluation data, weight data, and other feature data are associated.

[0034] As shown in Figure 6, the first proposed data 61 is associated with the first evaluation data 62, the first weight data 63, and other feature data (first policy issue data 64, first evaluator attribute data 65, first logical data 66, and first time data 67) by the preprocessing unit (association unit) 37. Similarly, the second proposed data 68 is associated with the second evaluation data 69, the second weight data 70, and other feature data (second policy issue data 71, second evaluator attribute data 72, second logical data 73, and second time data 74) by the preprocessing unit (association unit) 37. Furthermore, the first proposed data 61 and the second proposed data 68 are associated with at least one of the other feature data (policy issue data 64, 71, evaluator attribute data 65, 72, logical data 66, 73, and time data 67, 74) by the preprocessing unit (association unit) 37.

[0035] Figure 7 shows an example of machine learning data in which training policy proposal data, evaluation data, weight data, and other feature data are associated.

[0036] As shown in Figure 7, the first policy proposal data 161 is associated with the first evaluation data 162, the first weight data 163, and other feature data (first policy issue data 164, first evaluator attribute data 165, first logical data 166, and first time data 167) by the preprocessing unit (association unit) 37. Similarly, the second policy proposal data 168 is associated with the second evaluation data 169, the second weight data 170, and other feature data (second policy issue data 171, second evaluator attribute data 172, second logical data 173, and second time data 174) by the preprocessing unit (association unit) 37. Furthermore, the first policy proposal data 161 and the second policy proposal data 168 are associated with at least one of the other feature data (policy issue data 164, 171, evaluator attribute data 165, 172, logical data 166, 173, and time data 167, 174) by the preprocessing unit (association unit) 37.

[0037] Figure 8 is a flowchart illustrating an example of the inference method of this embodiment. Figures 9 and 10 are sequence diagrams showing examples of operations performed by the inference program of this embodiment. Figure 9 is a sequence diagram showing an example of operations for acquiring and preprocessing various data, and for generating and storing a trained model. Figure 10 is a sequence diagram showing an example of operations for inference using the trained model and determining the inference result. The inference program may be stored on a recording medium.

[0038] As shown in Figures 8 and 9, in step S1, the data storage unit 35 of the learning database server 4 executes a command to store various data (111) and stores the various data input from the input device 32 in the learning database 40. The various data to be stored include proposal data stored in the proposal database 43 (first proposal data stored in the first proposal database 431 and second proposal data stored in the second proposal database 432), evaluation data stored in the evaluation database 44, weight data stored in the weight database 45, evaluator attribute data stored in the evaluator attribute database 46, policy proposal data stored in the policy proposal database 48, logical data stored in the logical database 49, and time data stored in the time database 50.

[0039] In step S2, the preprocessing unit 37 of the learning database server 4 performs preprocessing of various data (112).

[0040] In step S3, the preprocessing unit 37 executes an instruction to expand the training data (113) and expands the training data. The preprocessing unit (association unit) 37 expands the training feature data by associating at least one of the policy issue data, logical data, time data, evaluation data, and evaluator attribute data with the proposal data. The preprocessing unit (association unit) 37 also expands the training feature data by associating at least one of the policy issue data, logical data, time data, evaluation data, and evaluator attribute data with the policy proposal data. Furthermore, the preprocessing unit (association unit) 37 associates the first proposal data and the second proposal data based on at least one of the policy issue data, logical data, time data, evaluation data, and evaluator attribute data of the first proposal data and the second proposal data.

[0041] In step S4, the preprocessing unit 37 executes a filtering command for the training data (114) and performs filtering. The preprocessing unit 37 performs filtering to exclude some of the items of the feature data from the training data. At this time, the preprocessing unit 37 may perform oversampling, undersampling, or scaling.

[0042] The processor 21 of the machine learning computer 3 sends commands to acquire and transmit various data to the learning database server 4 (115). In step S5, the data acquisition unit 36 ​​of the learning database server 4 executes commands to acquire and transmit various data (116), acquires various data from the learning database 40, and transmits the various data to the machine learning computer 3 via the interface 33.

[0043] In step S6, the machine learning computer 3 executes a suggestion data acquisition command (117), and the suggestion data acquisition unit 260 acquires suggestion data for a predetermined policy issue and displays it on the display of the output device 25.

[0044] In step S7, the evaluation unit 261 of the machine learning computer 3 executes an evaluation data acquisition command (118), acquires evaluation data associated with the proposed data, and evaluates the proposed data based on the evaluation data. For example, the evaluation unit 261 evaluates the proposed data based on an ordinal scale relating to multiple evaluation items of the proposed data (e.g., specificity, measurability, motivation, realism, timeliness, effectiveness, and fairness). In this case, the evaluation unit 261 may evaluate the proposed data based on at least one of policy issue data, logical data, time data, and evaluator attribute data. For example, the evaluation unit 261 may evaluate the proposed data based on its relevance to policy issues, the justification, support, counterarguments, and citation relationships of the proposed data (first proposed data or second proposed data), the timing of the generation of the proposed data, and the evaluator's level of expertise. Alternatively, the evaluation unit 261 may evaluate the proposed data based on a predetermined function that uses these evaluation data as variables.

[0045] In step S8, the weighting unit 262 of the machine learning computer 3 executes a weighting command (119) and weights the proposed data based on the evaluation. For example, the weighting unit 262 weights the proposed data based on an ordinal scale relating to multiple evaluation items of the proposed data (e.g., specificity, measurability, motivation, realism, timeliness, effectiveness, and fairness). In this case, the weighting unit 262 may weight the proposed data based on at least one of the policy issue data, logical data, time data, and evaluator attribute data. For example, the weighting unit 262 may weight the proposed data based on its relevance to the policy issue, the justification, support, counterarguments, and citation relationships of the proposed data (first proposed data or second proposed data), the generation time of the proposed data, and the evaluator's level of expertise. Alternatively, the weighting unit 262 may weight the proposed data based on a predetermined function that uses these evaluation data as variables.

[0046] In step S9, a trained model is generated. The processor 21 of the machine learning computer 3 executes a command to generate a policy proposal model by machine learning (120), and the machine learning unit 27 generates a trained model that infers the policy proposal data by machine learning based on the proposed data and weights. The machine learning unit 27 executes a command to generate a trained model by machine learning based on the training data (including the expanded training data).

[0047] The machine learning unit 27 generates a trained model using a neural network based on training data. For example, the machine learning unit 27 generates a trained model using a Large-Scale Language Model (LLM). The LLM is based on a neural network, particularly a structure called a Transformer, and uses an encoder that processes input text and quantifies its meaning and context, and a decoder that generates new text based on the quantified data (the encoder part is omitted in GPT-based models). By using a Self-Attention Mechanism, it calculates how each word relates to other words in the overall context, thereby generating a trained model that can estimate long contexts and complex structures. The Transformer consists of many layers, and each layer learns features at different levels, enabling the estimation of more abstract and complex language patterns.

[0048] The LLM learning process involves preparing data (collecting feature data), dividing the data into tokens (units of words or strings), and inputting them into the model. The model learns through the task of predicting the next word and adjusts the weights to minimize the loss function. For example, the weights are adjusted by balancing the data, adjusting the loss function, or relabeling tokens to emphasize specific tokens or token patterns. In this process, the loss function is adjusted based on the weights assigned by the weighting unit 262.

[0049] Furthermore, after training, the model is fine-tuned with additional data to suit specific tasks. For example, reinforcement learning is used based on proposer and evaluator attributes, or custom data containing a large amount of specific tokens is used to adjust the token weights when retraining the model based on the weights assigned by the weighting unit 262.

[0050] Furthermore, the weight of the tokens may be adjusted by performing rule-based filtering or adjusting token-level embeddings based on the weights assigned by the weighting unit 262.

[0051] As described above, proposal data that receive a high evaluation (weighting) for a policy issue have a certain relationship with policy proposal data that solves the policy issue. Therefore, by generating a trained model that generates policy proposal data from proposal data based on the weighting, and inferring policy proposals that solve the policy issue from the inference feature data (prompts, etc.) input into the trained model, it is possible to infer highly-rated policy proposals with high accuracy.

[0052] In step S10, the trained model storage unit 28 stores the generated trained model in the memory device 22 by executing a command to store the trained model (policy proposal model) (123).

[0053] In step S11, if the process proceeds to the step of generating policy proposal data, it proceeds to step S12; otherwise, the process terminates.

[0054] As shown in Figures 8 and 10, in step S12, the processor 21 of the machine learning computer 3 receives feature data (prompts) and time data for inference via the input device (input unit) 23 in order to infer policy proposal data (129).

[0055] In step S13, the generation unit 263 executes a data check command to verify the data format of the feature data and time data for inference, and determines whether the data format is correct or not (131). If the data format is incorrect, the generation unit 263 generates data indicating that fact.

[0056] If the data format is correct, in step S14, the inference unit 29 executes a data preprocessing command and encodes the feature data and time data for inference according to the encoding policy (131). If the feature data and time data for inference contain values ​​that were not used during training, the inference unit 29 either excludes those values ​​or assigns predetermined values ​​according to predetermined conditions.

[0057] In step S15, the generation unit 263 executes an inference command, reads the trained model stored in the memory device 22, inputs feature data and time data for inference into the trained model, and infers predetermined policy proposal data from the feature data and time data input into the trained model (132).

[0058] For example, when the input device (input unit) 23 receives a prompt and a large-scale language model (LLM) generates policy proposal data, the policy proposal model probabilistically predicts the sequentially appearing tokens and generates tokens one by one. Through the transformer mechanism, the generated tokens are determined considering the surrounding context, and the policy proposal data is output as a continuous, natural sentence, enabling the generation of policy proposals that reflect suggestions for policy issues. Furthermore, by adjusting the weight of the proposal data based on at least one of the policy issue data, logical data, time data, and evaluator attribute data, it is possible to generate highly specialized policy proposals or policy proposals with a strong tendency toward a given characteristic. For example, it is possible to generate policy proposals similar to those of policy experts or policy proposals similar to the tendencies of a given politician.

[0059] In this case, the generation probability distribution of the model may be scaled based on the weights assigned by the weighting unit 262 using a temperature parameter, thereby changing the tendency to select tokens with higher probabilities (lower temperatures) or tokens with higher diversity (higher temperatures), and prioritizing specific tokens or patterns during the generation process. Alternatively, token generation may be adjusted based on the weights assigned by the weighting unit 262 by using top k sampling. Furthermore, token generation may be adjusted by adjusting the probability of token generation (logits score) based on the weights assigned by the weighting unit 262.

[0060] In this embodiment, an example of a large-scale language model (LLM) is shown. However, the generation unit 263 may generate policy proposal data from the proposal data based on weighting, using other methods such as N-gram language models, sequence generation models, or rule-based models. For example, in the case of a rule-based model, the generation unit 263 may generate policy proposal data from the proposal data by randomly or rule-basedly combining characteristic keywords and phrases listed based on the weighting of the proposal data.

[0061] In step S16, the evaluation unit 261 evaluates the policy proposal data based on at least one of the policy issue data, logical data, time data, evaluation data, and evaluator attribute data associated with the policy proposal data by executing a policy proposal data evaluation command (133). The policy proposal model is improved based on the evaluation of the policy proposal data. The evaluation data for the policy proposal data may be input by the proposal / evaluation computer 5, as shown in Figure 5, or it may be input directly by the machine learning computer 3.

[0062] In step S17, the processor 21 of the machine learning computer 3 causes the inference results (policy proposal data) to be displayed on the output device 25 (for example, a display) (134).

[0063] In step S18, the input device (input unit) 23 of the machine learning computer 3 specifies the determination of policy proposal data, and the generation unit 263 executes the command to determine the policy proposal data (135). Furthermore, the determined policy proposal data can be modified by the generation unit 263.

[0064] In step S19, the policy proposal data generated by the generation unit 263 is associated with feature data for inference (including policy issue data) and sent to the learning database server 4 as policy proposal data (136). The data storage unit 35 of the learning database server 4 executes a command to store the learning data (137) and stores the policy proposal data sent from the machine learning computer 3 in the policy proposal database 48 of the learning database 40.

[0065] Alternatively, the policy proposal data generated by the generation unit 263 may be associated with feature data for inference (including policy issue data) and sent to the learning database server 4 as new proposal data. The data storage unit 35 of the learning database server 4 executes a command to store learning data and stores the policy proposal data sent from the machine learning computer 3 as new proposal data in the proposal database 43 of the learning database 40. This increases the amount of learning data in the learning database 40, enabling the policy proposal data to be inferred with even greater accuracy.

[0066] Then, when the icon on the policy proposal display unit 87 (Figure 5) of the proposal / evaluation computer 5 is selected by a cursor or the like, a command to acquire / send policy proposal data is sent (138). In response, the learning database server 4 executes a command to acquire / send policy proposal data generated by the generation unit 263 of the machine learning computer 3 (139) and sends the policy proposal data to the proposal / evaluation computer 5. The proposal / evaluation computer 5 executes a command to display policy proposal data and displays the policy proposal data generated by the generation unit 263 of the machine learning computer 3 (140).

[0067] As described above, according to this embodiment, policy proposal data can be inferred with high accuracy from proposal data using machine learning, and policy proposals that reflect proposals for policy issues can be generated.

[0068] Although embodiments of the present invention have been described above, the present invention is not limited thereto and can be modified or altered within the scope described in the claims.

[0069] In the above, policy proposal data was inferred based on the weighting of the proposed data, but policy issue data may also be inferred based on the weighting of the policy proposal data. In this case, the evaluation unit 261 evaluates the policy proposal data based on at least one of the policy issue data, logical data, time data, evaluation data, and evaluator attribute data associated with the policy proposal data, the weighting unit 262 weights the policy proposal data based on at least one of the policy issue data, logical data, time data, evaluation data, and evaluator attribute data, and the generation unit 263 generates policy issue data relating to policy issues from the policy proposal data based on the weighting.

[0070] Then, the machine learning unit 27 generates a trained model that infers policy issue data for the policy proposal using machine learning, based on the policy proposal data and weights, and the input device (input unit) 23 inputs feature data for inference into the trained model in order to infer the policy issue data.

[0071] By using the flowcharts and sequence diagrams shown in Figures 8 to 10 and the above explanation, a policy issue inference system, inference method, and inference program can be realized by replacing policy proposal data with policy issue data and policy issue data with policy proposal data. Policy proposal data that receive a high evaluation (weighting) for a policy issue have a certain relationship with new policy issue data for refining the policy proposal. Therefore, a trained model is generated that generates policy issue data from policy proposal data based on the weighting, and from the inference feature data (prompts, etc.) input to the trained model, it is possible to infer policy issues for refining the policy proposal with high accuracy, and to further pursue policy issues with policy proposals that reflect proposals for policy issues.

[0072] Furthermore, as shown in Figure 7, the machine learning unit 27 may generate a trained model by machine learning based on training feature data augmented by associating at least one of the policy issue data, logical data, time data, evaluation data, and evaluator attribute data with the policy proposal data.

[0073] Furthermore, the evaluation unit 261 may evaluate the proposer of the proposal data based on at least one of the proposal data, policy proposal data, policy issue data, logical data, time data, evaluation data, and evaluator attribute data. In this case, the evaluation unit 261 may evaluate using the average value of the evaluation data of the proposal data proposed by the proposer, or it may evaluate comprehensively using the evaluation data and other data, or it may evaluate using a learning model generated by the machine learning unit 27 with the evaluation data of the proposal data proposed by the proposer as training data. Note that the training data may be training data in which at least one of the policy proposal data, policy issue data, logical data, time data, evaluation data, and evaluator attribute data is associated with the evaluation data.

[0074] In this case, the weighting unit 262 may weight the proposer attribute data based on the evaluation, and the generation unit 263 may generate candidate data from the proposer attribute data based on the weighting. Alternatively, the machine learning unit 27 may generate a trained model for inferring candidate data using machine learning based on the proposer attribute data and the weighting, and the input device (input unit) 23 may input feature data for inference into the trained model in order to infer candidate data. This makes it possible to extract proposers with high evaluations as candidates (for example, candidates for parliament). [Industrial applicability]

[0075] This invention is useful as a policy proposal generation system that can generate policy proposals that reflect suggestions for policy issues. [Explanation of Symbols]

[0076] 1…Policy proposal generation system 2…Network 3…Machine Learning Computers 3… Machine Learning Computers 4…Learning database server 5…Proposal / Evaluation Computer 21… Processor 22...Storage device 23…Input device 24… Interface 25…Output device 27…Machine Learning Department 28…Model storage 29… Reasoning part 30… Processor 32…Input device 33… Interface 34…Output device 35...Data storage unit 36...Data acquisition unit 37…Pre-treatment section 40…Learning database 41…Feature Database 43…Evaluation Database 43…Proposal Database 44…Evaluation Database 45...Weight Database 46… Evaluator Attribute Database 48…Policy Proposal Database 49…Logical Database 80... screen 81…Policy issue display section 82...Detailed information display section 83…Suggestion display area 84...Input display section 85…Number of proposers display area 86... Display section for the number of evaluators 87…Policy proposal display area 260... Proposal Data Acquisition Unit 261…Evaluation Department 262...Heavy part 263…Generation part 501… Processor 502…Policy Issue Data Acquisition Department 503... Proposal Data Input Section 504...Evaluation data input section 505... Evaluator attribute data entry section 523...Input device 524… Interface 525...Output device

Claims

1. A policy proposal generation system comprising a processor and a memory device, wherein the processor generates policy proposal data, The aforementioned processor, A proposal data acquisition unit that acquires proposal data for a given policy issue, An evaluation unit that evaluates the proposed data based on evaluation data associated with the proposed data, A weighting unit that weights the proposed data based on the above evaluation, A generation unit that generates the policy proposal data from the proposed data based on the weighting, A policy proposal generation system characterized by comprising the following features.

2. A machine learning unit generates a trained model for inferring the policy proposal data using machine learning based on the proposed data and the weightings. An input unit that inputs feature data for inference into the pre-trained model in order to infer the policy proposal data, The policy proposal generation system according to claim 1, characterized by comprising the following features.

3. The policy proposal generation system according to claim 2, characterized in that the machine learning unit generates the trained model by machine learning based on the training feature data which has been expanded by associating at least one of policy issue data, logical data, time data, evaluation data, and evaluator attribute data with the proposed data.

4. The system includes a linking unit that links the first proposal data and the second proposal data based on at least one of the policy issue data, logical data, time data, evaluation data, and evaluator attribute data of the first proposal data and the second proposal data, The acquisition unit acquires the first proposed data and the second proposed data associated with the first proposed data. The evaluation unit evaluates the first proposal data and the second proposal data based on at least one of the policy issue data, logical data, time data, evaluation data, and evaluator attribute data associated with the first proposal data and the second proposal data. The weighting unit, based on the evaluation, weights the first proposed data and the second proposed data. The policy proposal generation system according to claim 1 or 2, characterized in that the generation unit generates the policy proposal data from the first proposal data and the second proposal data based on the weighting.

5. The policy proposal generation system according to claim 1 or 2, characterized in that the evaluation unit evaluates the proposer of the proposal data based on at least one of the proposal data, policy proposal data, policy issue data, logical data, time data, evaluation data, and evaluator attribute data.

6. The evaluation unit evaluates the policy proposal data based on at least one of the policy issue data, logical data, time data, evaluation data, and evaluator attribute data associated with the policy proposal data. The weighting unit weights the policy proposal data based on at least one of the policy issue data, the logical data, the time data, the evaluation data, and the evaluator attribute data. The policy proposal generation system according to claim 1 or 2, characterized in that the generation unit generates policy issue data relating to the policy issue from the policy proposal data based on the weighting.

7. A machine learning unit generates a trained model that infers policy challenge data for the policy proposal using machine learning, based on the policy proposal data and the weightings. An input unit that inputs feature data for inference into the trained model in order to infer the aforementioned policy issue data, The policy proposal generation system according to claim 6, characterized by comprising the following features.

8. The policy proposal generation system according to claim 7, characterized in that the machine learning unit generates the trained model by machine learning based on the training feature data which has been expanded by associating at least one of the policy issue data, logical data, time data, evaluation data, and evaluator attribute data with the policy proposal data.

9. A method for generating policy proposals in which a processor generates policy proposal data, Steps include obtaining proposal data for a given policy issue, A step of evaluating the proposed data based on evaluation data associated with the proposed data, A step of weighting the proposed data based on the evaluation, A step of generating the policy proposal data from the proposed data based on the weighting, A method for generating policy proposals, characterized by comprising the following features.

10. A policy proposal generation program in which a processor generates policy proposal data, The aforementioned computer, A proposal data acquisition function that acquires proposal data for a given policy issue, An evaluation function that evaluates the proposed data based on evaluation data associated with the proposed data, Based on the above evaluation, a weighting function is provided to weight the proposed data, A generation function that generates the policy proposal data from the proposed data based on the weighting, A policy proposal generation program characterized by its ability to achieve this.

Images