Voting aggregation system based on large language model and Aeroconditional analysis
By combining a large language model and Arrow conditional analysis, the voting aggregation system solves the problems of inconsistent performance and lack of interpretability of traditional voting systems in different scenarios, realizes efficient and transparent collective decision-making, supports evaluation of multiple scenarios and rules, and improves user experience and decision quality.
Patent Information
- Application Number
- CN202510828281.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-20
- Publication Date
- 2025-11-07
AI Technical Summary
Traditional voting systems perform poorly across different scenarios, lack interpretability, struggle to handle fuzzy preference information, and the application of large language models in structured decision-making has not been fully explored.
Design a voting aggregation system based on large language models and Road condition analysis, including data input processing, multi-mechanism aggregation calculation and result analysis output units. It integrates multiple voting rules and large language models, verifies Road conditions in real time, and provides transparent decision reasoning and rule recommendation.
It significantly simplifies the collective decision-making process, improves the transparency of results and user trust, expands the scope of application, supports any number of voters and candidates, provides scientific evidence and multi-faceted rule evaluation, lowers the threshold for use, and enhances the system's flexibility and applicability.
Smart Images
Figure CN120912403A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of big data voting, and particularly relates to a voting aggregation system based on a large language model and Arrovian condition analysis. BACKGROUND
[0002] As a discipline for studying the aggregation of individual preferences into collective decisions, the core of social choice theory is to design a voting system that meets certain normative conditions. Since the mid-20th century, Kenneth Arrow's impossibility theorem has driven significant theoretical progress in this field. The theorem proves that there is no perfect voting aggregation rule that meets the rationality conditions of unrestricted domain, transitivity, consistency, independence of irrelevant alternatives, and non-dictatorship. Traditional voting systems are mainly divided into three categories: positional voting rules, elimination rules, and pairing rules, which are widely used in political elections, committee decisions, and award selection scenarios. However, according to Arrow's theorem, these traditional voting rules will inevitably violate certain rationality conditions in some situations in practical applications.
[0003] Currently, traditional voting systems face many key challenges: different voting rules perform differently in various scenarios, making it difficult for users to adapt to specific scenarios; traditional voting rules based on strict mathematical foundations lack explanatory power for the aggregation process; the system cannot feedback on the normative conditions that may be violated; it is difficult to handle fuzzy preferences or incomplete information of voters; and in recent years, large language models in the field of artificial intelligence have made breakthroughs, although they have shown strong capabilities in information extraction and content generation in many scenarios, their application in structured decision-making problems is still in its early stages, and the potential of using large language models as a voting aggregation mechanism has not been fully explored. To combine traditional voting theory with modern AI technology, we propose a voting aggregation system based on a large language model and Arrovian condition analysis. SUMMARY
[0004] To solve the above technical problems, the application is realized by the following technical solutions:
[0005] The application is a voting aggregation system based on a large language model and Arrovian condition analysis, which includes a data input processing unit, a multi-mechanism aggregation calculation unit, and a result analysis output unit. The data input processing unit includes a front-end interactive interface, a preference collection module, a data verification unit, and a distributed storage module.
[0006] The multi-mechanism aggregation calculation unit includes a core calculation engine module, a traditional rule execution module, a large language model aggregation module, and a service communication module.
[0007] The result analysis output unit includes an Arrovian condition verification module, a theoretical analysis engine module, and a result comparison module.
[0008] The workflow of the voting aggregation system is as follows:
[0009] Step S1: Receive voter preference data through manual input, batch import or API interface, support any number of voters and candidates;
[0010] Step S2: Divide the preference data by sliding window, verify the integrity and consistency of the preferences, and generate error correction suggestions;
[0011] Step S3: Call no less than 13 voting rule libraries, including position rules, elimination rules and pairing rules, perform parameterized calculation and output the ranking results;
[0012] Step S4: Integrate multi-model API, guide the generation of social preference ranking and decision reasoning through structured prompt templates, and parse into standardized output;
[0013] Step S5: Real-time verification of whether each rule violates transitivity, consistency, independence of irrelevant alternatives and non-autocracy, and calculation of violation probability through Monte Carlo simulation;
[0014] Step S6: Calculate the ranking distance index of different mechanism results, generate rule performance dashboard, and recommend the optimal rule based on preference characteristics.
[0015] Further, the front-end interactive interface output end is unidirectionally connected with the preference collection module input end, the preference collection module output end is unidirectionally connected with the data verification unit input end, the data verification unit output end is unidirectionally connected with the core calculation engine module input end, the core calculation engine module output end is unidirectionally connected with the traditional rule execution module and the large language model aggregation module input end, the traditional rule execution module output end and the large language model aggregation module output end are both unidirectionally connected with the Arrow condition verification module input end, the Arrow condition verification module output end is unidirectionally connected with the theoretical analysis engine module input end, the theoretical analysis engine module output end is unidirectionally connected with the result comparison module input end, the result comparison module output end is unidirectionally connected with the front-end interactive module input end, the distributed storage module is bidirectionally connected with the core calculation engine module, and the service communication module input end is unidirectionally connected with the large language model aggregation module output end.
[0016] Further, the front-end interactive interface is based on Web technology to build an interactive operation interface, providing users with a preference data input channel, a voting rule selection entrance, and a visual display platform for aggregation results and theoretical analysis, realizing direct interaction between users and the system;
[0017] The preference collection module is responsible for collecting and managing voter preference data, supporting manual input, batch import (such as CSV, Excel, JSON, etc.), and API submission, and can adaptively process voter group data of different sizes, while having preference data verification, visualization display, and simulation generation functions;
[0018] The input preference data is automatically checked for integrity, consistency and validity. Once data anomalies are found, the user is provided with error correction suggestions in a timely manner to ensure that the subsequent processed preference data is accurate and reliable;
[0019] The distributed storage module is responsible for building a database system for storing voter preference data, rule aggregation calculation results, and historical voting records, and realizing persistent storage and management of data;
[0020] The core computing engine module is used to coordinate the operation of each functional module, responsible for core computing tasks such as data processing, rule execution, large language model calling, and condition verification, so that the overall operation logic of the system is executed;
[0021] The traditional rule execution module has at least 13 classic voting rules (including position rules, elimination rules, and pairing rules) built-in, which can aggregate the input preference data, support parameter adjustment of some rules (such as Borda weight modification and elimination threshold setting), and provide working principle explanation and calculation process visualization for each rule;
[0022] The large language model aggregation module uses a large language model as an independent preference aggregation mechanism, integrates multiple large language model APIs (such as OpenAI GPT series and Anthropic Claude series), guides the model to generate high-quality aggregation results through optimization of prompt engineering, parses the model output into structured social preference rankings, and unifies the format with traditional rule results, while having error handling and multi-round optimization functions;
[0023] The service communication module provides an API interface layer to realize communication with external systems and large language model services, ensuring smooth data interaction and function calling between the system and external resources;
[0024] The Arrow condition verification module implements an accurate test algorithm for the Arrow conditions (transitivity, consistency, independence of irrelevant alternatives, and non-dictatorship), estimates the probability of each aggregation rule violating the conditions through the Monte Carlo simulation method, integrates theoretical research results, and conducts real-time condition testing on the current preference configuration, as well as comparative analysis and sensitivity analysis;
[0025] The theoretical analysis engine module is used to comprehensively analyze the aggregation results of traditional rules and large language models, evaluate the consistency of the results, generate a rule performance dashboard, display the decision reasoning process of the large language model, conduct sensitivity tests, and recommend the most suitable aggregated rules to the user based on preference configurations and theoretical performance;
[0026] The result comparison module is used to compare the results of different aggregation mechanisms in multiple dimensions, calculate the ranking distance index, identify the ranking difference points, analyze the difference reasons, intuitively display the theoretical performance of each aggregation mechanism, present the decision reasoning process of the large language model, and help the user understand the performance differences of different rules under the same conditions, providing a basis for rule selection.
[0027] Further, in step S1, the voter preference data is received through manual input, batch import (in CSV / Excel / JSON format), or API interface. The system supports processing any number of voters and candidates, covering small committee decisions to large election scenarios. After data input, the data is automatically entered into the distributed storage module for persistent management of preference data.
[0028] Further, in step S2, the sliding window technique is used to segment the preference data, and the integrity, consistency, and validity of the data are automatically checked (such as checking for contradictory rankings or missing items). If abnormalities are found, visual error correction suggestions are generated and returned to the user. At the same time, preference data can be generated based on IC / IAC / Mal lows theory distribution simulation for rule performance testing.
[0029] Further, in step S3, at least 13 classic voting rule libraries are called, including the following steps:
[0030] Step S31, Position Rules: Plurality, Anti-plurality, Borda, Dowdall;
[0031] Elimination Rules: Plurality Runoff, Hare, Coombs, Baldwin, Nanson;
[0032] Pairing Rules: Pairwise Majority, Copeland, Simpson-Kramer, Ranked Pairs;
[0033] Step S32, parameterized adjustment of rules (such as Borda weight, elimination threshold) is supported. After aggregation calculation, the structured social preference ranking is output, and the rule working principle and calculation process are displayed through the visual interface. According to the theoretical properties, the rules are classified to help users locate the rule type, and professional users are allowed to add custom rules to extend the system functions.
[0034] Further, in step S4, an integrated multi-model API (such as the GPT series and the Claude series) is designed to guide the model to generate aggregation results by designing a structured prompt template specifically for preference aggregation tasks, generate JSON format output containing social preference ranking and decision reasoning, convert unstructured text into standardized ranking results, implement robust error detection and recovery mechanisms, handle API call failures and output format error exceptions, the system automatically parses the model response, and supports optimizing output certainty by adjusting temperature and top-p parameters, while having API call error handling and multi-round result optimization mechanisms.
[0035] Further, in step S5, the satisfaction of each aggregation mechanism to the Arrow condition is verified in real time, including the following steps:
[0036] Step S51, transitivity: verify whether the social preference ranking conforms to the transitive logic;
[0037] Consistency: check whether the consistent preferences of all voters are respected;
[0038] Independence of irrelevant alternatives: verify whether the ranking only depends on the preferences of the target option pair;
[0039] Non-autocracy: check whether there is a single option that dominates the decision;
[0040] Step S52, estimate the probability of violating each rule by Monte Carlo simulation, generate a confidence interval combined with theoretical research results, and support different preference distributions (such as IC and IAC);
[0041] Step S53, by integrating published theoretical research results, provide violation probabilities for parameter combinations, real-time verify whether each rule violates a specific condition for the current preference configuration, compare the violation probabilities of different rules under the same condition, highlight the relative advantages, and evaluate the sensitivity of violation probability to parameter changes.
[0042] Further, in step S6, calculate the ranking distance indicators (such as Kendall's Tau and Spearman's correlation coefficient) of different aggregation mechanism results, generate a visual rule performance dashboard, intuitively display the probability of violating the Arrow condition for each rule and the performance under the current preference configuration, based on preference characteristics (such as polarization degree and diversity) and the importance of the condition to the user, generate optimal rule recommendations through a multi-objective recommendation algorithm, and show the decision reasoning process of the large language model and the logic comparison with traditional rules.
[0043] The present application has the following beneficial effects:
[0044] 1. This invention is the first to integrate traditional voting rules, large language model aggregation, and Arrow conditional analysis into a single platform, providing users with a "one-stop" decision-making solution. It significantly simplifies the collective decision-making process, can handle any number of voters and multiple candidate configurations, and is adaptable to various scenarios from small team decision-making to large organization voting, greatly expanding the scope of application.
[0045] 2. In this invention, users can select different aggregation rules according to specific needs, or apply multiple rules simultaneously for comparison, which enhances the flexibility and applicability of the system. It not only provides aggregation results, but also provides detailed theoretical performance analysis and decision reasoning based on large language models, making the decision-making process more transparent and improving users' understanding and trust in the results. Based on preference configuration features and theoretical performance analysis, the system can intelligently recommend aggregation rules suitable for specific situations, simplifying the user's rule selection process.
[0046] 3. This system is the first to achieve quantitative estimation of the probability of Law condition violation in a practical voting platform, extending theoretical performance analysis from the research field to practical applications. It provides users with a scientific basis for rule selection, innovatively using large language models as an independent preference aggregation mechanism and evaluating their probability of violating Law conditions, opening up a new field of application for large language models. The system combines the rigorous analysis of social choice theory with the convenience of a practical system, bridging the gap between theoretical research and practical application. It supports simulation analysis based on theoretical distributions and the processing of real-world preference data, providing users with multi-faceted rule performance evaluation.
[0047] 4. By providing theoretical performance analysis and comparison of multiple aggregation mechanisms, this invention helps organizations choose decision-making methods more suitable for their specific situations, improving the quality of collective decision-making. By explicitly demonstrating the probability of different rules violating Arrow's conditions, the system helps users understand and mitigate the risk of potential social choice paradoxes. The system has advantages in handling non-standard preference data, especially by utilizing the understanding capabilities of large language models, enabling it to adapt to more real-world decision-making scenarios. It provides an experimental platform for social choice theorists to test new rules, analyze theoretical phenomena in real data, and promote theoretical innovation.
[0048] 5. By providing transparent theoretical analysis and decision reasoning, this invention enhances users' trust and acceptance of the aggregation results. The system presents complex voting theories and rules in a user-friendly manner, lowering the barrier to entry for advanced voting systems. The system can be used as an educational tool to help users understand social choice theory and the characteristics of different voting rules, improving collective decision-making literacy. The decision reasoning provided by the large language model makes the aggregation results more interpretable, helping users understand the logic behind complex decisions. Users can customize the analysis according to the conditions and performance indicators they are interested in, and obtain targeted rule evaluations and suggestions.
[0049] Of course, implementing any product of the application does not necessarily require achieving all the advantages described above at the same time. BRIEF DESCRIPTION OF DRAWINGS
[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.
[0051] Fig. 1 The figure is a schematic diagram of the voting aggregation system based on the large language model and the Arrow condition analysis of the present application.
[0052] Fig. 2 The figure is a schematic diagram of the voting aggregation system based on the large language model and the Arrow condition analysis of the present application. DETAILED DESCRIPTION
[0053] The technical solutions of the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0054] Please refer to Figs. 1-2 The voting aggregation system based on the large language model and the Arrow condition analysis of the present application includes a data input processing unit, a multi-mechanism aggregation calculation unit and a result analysis output unit, wherein the data input processing unit includes a front-end interactive interface, a preference collection module, a data verification unit and a distributed storage module.
[0055] The multi-mechanism aggregation calculation unit includes a core calculation engine module, a traditional rule execution module, a large language model aggregation module and a service communication module.
[0056] The result analysis output unit includes an Arrow condition verification module, a theoretical analysis engine module and a result comparison module.
[0057] The working process of the voting aggregation system is as follows:
[0058] Step S1: receiving voter preference data through manual input, batch import or API interface, supporting any number of voters and candidates;
[0059] Step S2: dividing the preference data by sliding window, verifying the integrity and consistency of the preference, and generating error correction suggestions.
[0060] Step S3: Call no less than 13 voting rule libraries, including position rules, elimination rules and pairing rules, perform parameterized calculation and output ranking results;
[0061] Step S4: Integrate multi-model API to guide the generation of social preference ranking and decision reasoning through structured prompt templates, and parse into standardized output;
[0062] Step S5: Real-time verification of whether each rule violates transitivity, consistency, independent of irrelevant alternatives and non-dictatorship, and calculation of violation probability through Monte Carlo simulation;
[0063] Step S6: Calculate the ranking distance index of different mechanism results, generate a rule performance dashboard, and recommend the optimal rule based on preference characteristics.
[0064] Further, the front-end interactive interface output end is unidirectionally connected with the preference collection module input end, the preference collection module output end is unidirectionally connected with the data verification unit input end, the data verification unit output end is unidirectionally connected with the core calculation engine module input end, the core calculation engine module output end is unidirectionally connected with the traditional rule execution module and the large language model aggregation module input ends, the traditional rule execution module output end and the large language model aggregation module output end are both unidirectionally connected with the Arrow condition verification module input end, the Arrow condition verification module output end is unidirectionally connected with the theoretical analysis engine module input end, the theoretical analysis engine module output end is unidirectionally connected with the result comparison module input end, the result comparison module output end is unidirectionally connected with the front-end interaction module input end, the distributed storage module is bidirectionally connected with the core calculation engine module, and the service communication module input end is unidirectionally connected with the large language model aggregation module output end.
[0065] Further, the front-end interactive interface is based on Web technology to build an interactive operation interface, providing users with a preference data input channel, a voting rule selection entrance, and a visualization display platform for aggregated results and theoretical analysis, realizing direct interaction between users and the system;
[0066] The preference collection module is responsible for collecting and managing voter preference data, supporting manual input, batch import (such as CSV, Excel, JSON, etc.), API submission and other diversified input methods, and can adaptively process voter group data of different sizes, while having preference data verification, visualization display and simulation generation functions;
[0067] The input preference data is checked for integrity, consistency and validity, and once data anomalies are found, error correction suggestions are provided to the user in a timely manner to ensure that the subsequent processed preference data is accurate and reliable;
[0068] The distributed storage module is responsible for building a database system for storing voter preference data, rule aggregation calculation results and historical voting records, realizing persistent storage and management of data.
[0069] The core computing engine module is used to coordinate the operation of each functional module, responsible for core computing tasks such as data processing, rule execution, large language model calling and condition verification, so that the overall operation logic of the system is executed.
[0070] The traditional rule execution module has at least 13 classic voting rules (including position rules, elimination rules, pairing rules, etc.) built-in, which can aggregate the input preference data, support parameter adjustment of some rules (such as Borda weight modification, elimination threshold setting, etc.), and provide working principle explanation and calculation process visualization for each rule.
[0071] The large language model aggregation module uses large language models as an independent preference aggregation mechanism, integrates multiple large language model APIs (such as OpenAI GPT series, Anthropic Claude series, etc.), guides the model to generate high-quality aggregation results through optimized prompt engineering, parses the model output into structured social preference rankings, and unifies the format with traditional rule results, while having error handling and multi-round optimization functions.
[0072] The service communication module provides an API interface layer to realize communication with external systems and large language model services, ensuring smooth data interaction and function calling between the system and external resources.
[0073] The Arrow condition verification module implements an accurate verification algorithm for the Arrow conditions (transitivity, consistency, independence of irrelevant alternatives, non-dictatorship), estimates the probability of each aggregation rule violating each condition through Monte Carlo simulation method, integrates theoretical research results, conducts real-time condition verification on the current preference configuration, and carries out comparative analysis and sensitivity analysis.
[0074] The theoretical analysis engine module is used to analyze the aggregation results of traditional rules and large language models, evaluate the consistency of the results, generate a rule performance dashboard, display the decision-making reasoning process of the large language model, conduct sensitivity testing, and recommend the most suitable aggregation rule for the user based on the preference configuration characteristics and theoretical performance.
[0075] The result comparison module is used to compare the results of different aggregation mechanisms in multiple dimensions, calculate the ranking distance index, identify the ranking difference points, analyze the difference reasons, intuitively display the theoretical performance of each aggregation mechanism, present the decision-making reasoning process of the large language model, and help users understand the performance differences of different rules under the same conditions, providing a basis for rule selection.
[0076] Further, in step S1, the voter preference data is received by manual input, batch import (CSV / Excel / JSON format), or API interface. The system supports processing any number of voters and candidates, covering small committee decision-making to large election scenarios. After data input, it automatically enters the distributed storage module for persistent management of preference data.
[0077] Further, in step S2, the preference data is segmented using the sliding window technique, and the integrity, consistency, and validity of the data are automatically checked (such as checking for contradictory ordering or missing items). If abnormalities are found, visual error correction suggestions are generated and returned to the user. At the same time, it supports generating preference data based on IC / IAC / Mallows theory distribution simulation for rule performance testing.
[0078] Further, in step S3, at least 13 classic voting rule libraries are called, including the following steps:
[0079] Step S31, Position Rules: Plurality, Anti-plurality, Borda, Dowdall;
[0080] Elimination Rules: Plurality Runoff, Hare, Coombs, Baldwin, Nanson;
[0081] Pairing Rules: Pairwise Majority, Copeland, Simpson-Kramer, Ranked Pairs;
[0082] Step S32, support rule parameterization adjustment (such as Borda weight, elimination threshold), execute aggregation calculation, output structured social preference ordering, and show rule working principle and calculation process through visual interface. According to the theoretical nature, classify the rules to help users locate the rule type, allow professional users to add custom rules, and extend the system function.
[0083] Further, in step S4, integrate multi-model API (such as GPT series, Claude series), design structured prompt templates specifically for preference aggregation tasks, guide the model to generate aggregation results, generate JSON format output containing social preference ordering and decision reasoning, convert unstructured text to standardized ordering results, implement robust error detection and recovery mechanism, handle API call failure, output format error exception, system automatically parses model response, and supports adjusting temperature, top-p parameters to optimize output certainty, while having API call error handling and multi-round result optimization mechanism.
[0084] Further, the step S5 of real-time checking whether each aggregation mechanism satisfies the Arrow condition comprises the following steps:
[0085] Step S51, transitivity: verifying whether the social preference order conforms to the transitive logic;
[0086] Consistency: checking whether the consistent preferences of all voters are respected;
[0087] Independence of irrelevant alternatives: verifying whether the order only depends on the preferences of the target option pair;
[0088] Non-autocracy: checking whether there is a single voter who dominates the decision;
[0089] Step S52, estimating the probability of violating each rule condition through Monte Carlo simulation, combining the theoretical research results to generate a confidence interval, and supporting different preference distributions (such as IC, IAC);
[0090] Step S53, by integrating the published theoretical research results, providing violation probabilities for parameter combinations, real-time checking whether each rule violates a specific condition under the current preference configuration, comparing the violation probabilities of different rules under the same condition, highlighting the relative advantages, and evaluating the sensitivity of the violation probability to parameter changes.
[0091] Further, in the step S6, the ordering distance indicators (such as Kendall's Tau and Spearman's correlation coefficient) of the results of different aggregation mechanisms are calculated to generate a visual rule performance dashboard, which intuitively displays the probability of violating the Arrow condition and the performance under the current preference configuration. Based on the preference characteristics (such as polarization degree and diversity) and the importance of the condition to the user, the optimal rule suggestion is generated through a multi-objective recommendation algorithm, and the decision-making reasoning process of the large language model is compared with the logic of traditional rules.
[0092] One specific application of the embodiment is:
[0093] 1. System architecture implementation
[0094] Data input processing unit:
[0095] Front-end interactive interface: use the React framework to build a web interface, provide a preference input form (support drag-and-drop ordering of candidate options), a rule selection drop-down menu (contain 13 preset rules), and a result visualization panel;
[0096] Preference collection module: receive user-uploaded CSV files (column structure: voter ID, option A > option B > option C), and automatically parse them into a JSON-formatted preference order list;
[0097] Data verification unit: detect preference contradictions (e.g., the same voter prefers A>B and B>A), and generate error correction suggestions for missing items (e.g., "Candidate D for voter 3 is not ranked, please add");
[0098] Distributed storage module: use MongoDB shard cluster to store 100,000 historical voting records;
[0099] Multi-mechanism aggregation computing unit:
[0100] Core computing engine: coordinate task flow and dynamically schedule computing resources;
[0101] Traditional rule execution module:
[0102] Built-in 13 voting rule library, including position rule, elimination rule and pairing rule;
[0103] Support parameter customization function (such as adjusting the weight sequence of Borda count method);
[0104] Large language model aggregation module:
[0105] Integrate multiple large language model APIs;
[0106] Design structured prompt template to guide model to generate social ranking and explanation;
[0107] Equipped with error recovery mechanism to handle API exceptions;
[0108] Service communication module: connect external systems through standardized API gateway;
[0109] Result analysis output unit:
[0110] Aro condition verification module:
[0111] Real-time inspection of transitivity, independent of irrelevant alternatives, consistency and non-autocracy;
[0112] Generate 10,000 sets of theoretical distribution data through Monte Carlo simulation, and calculate the probability of rule violation;
[0113] Theoretical analysis engine: generate rule performance heat map and recommend optimal rule based on preference characteristics;
[0114] Result comparison module: calculate ranking difference index and visualize decision logic comparison;
[0115] 2、Work flow:
[0116] Scenario: 5 experts of a transportation company rank 4 logistics schemes (A / B / C / D);
[0117] S1 Data input: API receives JSON format preference data, containing the complete ranking sequence of each expert;
[0118] S2 Data verification: sliding window (window size = 3) detects that expert 3 has a circular contradiction (A > B and B > A), and the system generates a highlighted visual correction report;
[0119] S3 Traditional rule calculation:
[0120] Borda rule (weight [3, 2, 1, 0]): scheme B total score 9 points first;
[0121] Pairwise ranking method: lock the dominant pair B > A (3:2) to form the final ranking B > A > C > D
[0122] S4 Large language model aggregation:
[0123] Large language model outputs structured ranking results and decision reasons;
[0124] Adjust the temperature parameter to control the randomness of the output;
[0125] S5 Aro condition verification: quantify the probability of violating the key condition of each rule;
[0126] S6 Rule recommendation: based on the data polarization feature (polarization index 0.7), recommend Copeland method as the optimal rule (violation probability <5%).
[0127] In the description of the specification, the description of the terms "one embodiment", "example", "specific example" and the like means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In the specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.
[0128] The preferred embodiments of the application disclosed above are only used to help explain the application. The preferred embodiments do not describe all the details and limit the application to the specific embodiments described. Obviously, many modifications and changes can be made according to the content of the specification. The specification selects and describes these embodiments in order to better explain the principles and practical applications of the application, so that those skilled in the art can well understand and utilize the application. The application is limited only by the claims and their entire scope and equivalents.
Claims
1. A voting aggregation system based on large language models and Arrow condition analysis, characterized in that: The application relates to a voting aggregation system, comprising a data input processing unit, a multi-mechanism aggregation calculation unit and a result analysis output unit, wherein the data input processing unit comprises a front-end interactive interface, a preference collection module, a data verification unit and a distributed storage module; The multi-mechanism aggregation calculation unit comprises a core calculation engine module, a traditional rule execution module, a large language model aggregation module and a service communication module; The result analysis output unit comprises an Arrow condition verification module, a theoretical analysis engine module and a result comparison module; The working process of the voting aggregation system is as follows: Step S1: receiving voter preference data through manual input, batch import or API interface, supporting any number of voters and candidate items; Step S2: dividing the preference data by using a sliding window, verifying the preference integrity and consistency, and generating error correction suggestions; Step S3: calling no less than 13 voting rule libraries, including position rules, elimination rules and pairing rules, performing parameterized calculation and outputting the sorting results; Step S4: integrating multi-model API, guiding the generation of social preference sorting and decision reasoning through a structured prompt template, and analyzing the same into standardized output; Step S5: verifying whether each rule violates the transitivity, consistency, independent of irrelevant alternatives and non-autocraticity, and calculating the violation probability through Monte Carlo simulation; Step S6: calculating the sorting distance index of the results of different mechanisms, generating a rule performance instrument panel, and recommending the optimal rule based on the preference characteristics.
2. The voting aggregation system based on large language model and Arrow condition analysis according to claim 1, wherein, The front-end interactive interface output end is unidirectionally connected with the preference collection module input end, the preference collection module output end is unidirectionally connected with the data verification unit input end, the data verification unit output end is unidirectionally connected with the core calculation engine module input end, the core calculation engine module output end is unidirectionally connected with the traditional rule execution module and the large language model aggregation module input ends, the traditional rule execution module output end and the large language model aggregation module output end are both unidirectionally connected with the Arrow condition verification module input end, the Arrow condition verification module output end is unidirectionally connected with the theoretical analysis engine module input end, the theoretical analysis engine module output end is unidirectionally connected with the result comparison module input end, the result comparison module output end is unidirectionally connected with the front-end interactive module input end, the distributed storage module is bidirectionally connected with the core calculation engine module, and the service communication module input end is unidirectionally connected with the large language model aggregation module output end.
3. The voting aggregation system based on large language model and Arrow condition analysis according to claim 1, wherein, The front-end interactive interface is constructed based on Web technology and realizes interactive operation interface, provides a preference data input channel, a voting rule selection entrance and a visual display platform of aggregation results and theoretical analysis for users, and realizes direct interaction between the user and the system; The preference collection module is responsible for collecting and managing voter preference data, supports diversified input modes such as manual input, batch import and API submission, can adaptively process data of different sizes of voter groups, and has the functions of verifying, visualizing and simulating the preference data; The input preference data is automatically checked for integrity, consistency and validity, error correction suggestions are provided for the user in time once data anomalies are found, and the subsequent processed preference data is ensured to be accurate and reliable; The distributed storage module is responsible for constructing a database system for storing voter preference data, rule aggregation calculation results and historical voting records, and realizing persistent storage and management of data. The core computing engine module is used to coordinate the operation of each functional module, and is responsible for core computing tasks such as data processing, rule execution, large language model calling and condition verification, so as to execute the overall operation logic of the system. The traditional rule execution module has at least 13 built-in classic voting rules, which aggregate the input preference data, support parameter adjustment of some rules, and provide working principle explanation and calculation process visualization display for each rule. The large language model aggregation module uses a large language model as an independent preference aggregation mechanism, integrates multiple large language model APIs, guides the model to generate high-quality aggregation results through optimization of the prompt engineering, parses the model output into structured social preference ranking, and unifies the format with the traditional rule results, while having error handling and multi-round optimization functions. The service communication module provides an API interface layer to realize communication with external systems and large language model services, and ensures smooth data interaction and function calling between the system and external resources. The Arrow condition verification module realizes an accurate testing algorithm for the Arrow condition, estimates the probability of each aggregation rule violating the conditions through the Monte Carlo simulation method, integrates theoretical research results, performs real-time condition testing on the current preference configuration, and conducts comparative analysis and sensitivity analysis. The theoretical analysis engine module is used to analyze the aggregation results of traditional rules and large language models, evaluate the consistency of the results, generate a rule performance dashboard, display the decision-making reasoning process of the large language model, conduct sensitivity testing, and recommend the most suitable aggregation rule for the user based on the preference configuration characteristics and theoretical performance. The result comparison module is used to compare the results of different aggregation mechanisms in multiple dimensions, calculate the ranking distance index, identify the ranking difference points, analyze the difference reasons, intuitively display the theoretical performance of each aggregation mechanism, present the decision-making reasoning process of the large language model, and help users understand the performance differences of different rules under the same conditions, providing a basis for rule selection.
4. The voting aggregation system based on large language model and Arrow condition analysis according to claim 1, wherein, In step S1, voter preference data is received through manual input, batch import or API interface, and the system supports processing any number of voters and candidates, covering small committee decision-making to large election scenarios. After data input, the preference data is automatically entered into the distributed storage module for persistent management.
5. The voting aggregation system based on large language model and Arrow condition analysis according to claim 1, wherein, In step S2, the sliding window technique is used to segment the preference data, and the integrity, consistency and validity of the data are automatically checked. If abnormalities are found, visual error correction suggestions are returned to the user. The system also supports generating preference data based on IC / IAC / Mallows theory distribution simulation for rule performance testing.
6. The voting aggregation system based on large language models and analysis of Arrow conditions according to claim 1, characterized in that, In step S3, at least 13 classic voting rule libraries are called, including the following steps: Step S31, position rules: Plurality, Anti-plurality, Borda, Dowdall; Elimination rules: Plurality Runoff, Hare, Coombs, Baldwin, Nanson; Pairwise rules: Pairwise Majority, Copeland, Simpson-Kramer, Ranked Pairs; Step S32, support rule parameter adjustment, output structured social preference ranking after aggregation calculation, and show rule working principle and calculation process through visual interface, classify rules according to theoretical properties, help users locate rule type, allow professional users to add custom rules and expand system functions.
7. The voting aggregation system based on large language models and analysis of Arrow conditions according to claim 1, characterized in that, In step S4, integrate multi-model API, design structured prompt templates specifically for preference aggregation tasks, guide model to generate aggregation results, generate JSON format output containing social preference ranking and decision reasoning, convert unstructured text into standardized ranking results, implement robust error detection and recovery mechanism, handle API call failure, output format error exception, system automatically parses model response, and supports optimizing output certainty by adjusting temperature and top-p parameters, while having API call error handling and multi-round result optimization mechanism.
8. The voting aggregation system based on large language models and analysis of Arrow conditions according to claim 1, characterized in that, In step S5, real-time check whether each aggregation mechanism satisfies the Arrow condition, including the following steps: Step S51, transitivity: verify whether the social preference ranking meets the transitivity logic; Consistency: check whether all voters' consistent preferences are respected; Independence of irrelevant alternatives: verify whether the ranking only depends on the preferences of the target option pair; Non-dictatorship: check whether a single voter dominates the decision; Step S52, estimate the probability of violating the condition for each rule through Monte Carlo simulation, generate confidence interval combined with theoretical research results, and support different preference distributions; Step S53, by integrating published theoretical research results, provide violation probability for parameter combinations, real-time check whether each rule violates specific conditions for current preference configuration, compare violation probabilities of different rules under the same conditions, highlight relative advantages, and evaluate the sensitivity of violation probability to parameter changes.
9. The voting aggregation system based on large language model and Arrow condition analysis according to claim 1, wherein, In step S6, calculate the ranking distance index of different aggregation mechanism results, generate a visual rule performance dashboard, intuitively display the probability of violating the Arrow condition for each rule and the performance under the current preference configuration, generate optimal rule recommendations based on preference characteristics and user's emphasis on conditions through multi-objective recommendation algorithm, and show the decision reasoning process of large language models and the logic comparison with traditional rules.