Business rule-based person-house matching system for indemnificatory rental house

Abstract

The invention discloses a person-house matching system for indemnificatory rental houses based on business rules. The system comprises a multi-modal information extraction and feature fusion module; an attribute-level matching degree calculation module; the multi-rule dynamic weighting engine is connected with the attribute-level matching degree calculation module and comprises a configurable rule base and a rule conflict detection and resolution unit; the configurable rule base is used for storing and executing priority rules, constraint rules and optimization rules; a mixed distribution strategy execution module; and an interpretable output and strategy tracing module. Through the multi-modal information extraction and feature fusion module, the system can analyze and fuse semantic description, structured attributes and policy rules of an applicant at the same time, converts unstructured policy provisions into computable features, lays a rich and aligned data foundation for subsequent accurate matching, and improves the accuracy of policy analysis. And the limitations of single information dimension and insufficient utilization of a traditional mode are overcome.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of people-house matching, and particularly relates to a people-house matching system for security rental housing based on business rules. BACKGROUND

[0002] With the continuous advancement of urbanization process and the continuous improvement of housing security system in China, the supply scale of security rental housing is increasing, and the number of applicant groups and the diversity of demand are also significantly increasing.

[0003] The traditional people-house matching method mainly relies on manual review, simple sorting or semi-automatic means such as drawing lots, which has been difficult to cope with massive, multi-dimensional and dynamic application and housing information. Specifically, the existing technology generally has the following deficiencies: first, the information utilization is not sufficient, it is difficult to effectively integrate the text description of the applicant, the structured data and the complex policy provisions, resulting in single matching basis; second, the matching mode is rigid, usually using fixed scoring rules or waiting order, which cannot dynamically adjust the weight according to the individual factors such as the family structure and the employment status of the applicant, and the matching result is not strong in pertinence; third, it is difficult to handle rule conflicts, housing security policies often contain priority, constraint, optimization and other types of rules, the existing system lacks effective conflict detection and automatic resolution mechanism, and often relies on manual intervention, which is low in efficiency and inconsistent in standard; fourth, the allocation strategy is single, it is difficult to flexibly switch between intelligent recommendation, fair drawing, demand guidance and other multi-allocation modes according to the housing supply and demand situation and policy goals; fifth, the decision-making process is not transparent, the matching result lacks explainability, it is difficult to trace the influence of specific rules and data, which is not conducive to public supervision and policy evaluation.

[0004] Therefore, there is an urgent need for a security rental housing people-house matching system that can deeply integrate multi-source information, support individualized dynamic matching, intelligently handle rule conflicts, execute hybrid allocation strategies and provide full-link explainability, in order to improve the allocation efficiency, fairness, accuracy and public credibility. SUMMARY

[0005] The application provides a security rental housing people-house matching system based on business rules, which aims to solve the problems of single matching basis and rigid matching mode in the prior art.

[0006] In a first aspect, the security rental housing people-house matching system based on business rules comprises:

[0007] A multi-modal information extraction and feature fusion module extracts semantic features and structured features from applicant information and housing information, and converts housing security policies into computable rule features to generate an applicant feature set, a housing feature set and a policy rule set;

[0008] An attribute-level matching degree calculation module is connected with the multi-modal information extraction and feature fusion module, and is configured to calculate a single-item matching degree score of any applicant and any house source in each preset attribute dimension, and dynamically generate a personalized weight of each attribute dimension for different applicants based on an attention mechanism. A weighted sum is performed to obtain a comprehensive matching degree score;

[0009] A multi-rule dynamic weighting engine is connected with the attribute-level matching degree calculation module, and includes a configurable rule base, a rule conflict detection and resolution unit. The configurable rule base is configured to store and execute priority rules, constraint rules and optimization rules. The engine performs a weighted comprehensive decision on the allocation scheme based on the comprehensive matching degree score and the activated rules, and outputs a recommended allocation scheme.

[0010] A hybrid allocation strategy execution module is connected with the multi-rule dynamic weighting engine, and is configured to call and execute at least one allocation strategy including an intelligent recommendation mode, a lottery balancing mode, an order-based construction response mode and a resale regulation market mode according to the house source, the application queue and the policy target, and generate a final allocation instruction.

[0011] An explainability output and strategy tracing module is configured to generate a visual explanation report for each allocation result, and record system full-link operation and decision logs.

[0012] Optionally, in the multi-modal information extraction and feature fusion module, a sentence embedding model is used to extract semantic feature vectors of applicant text description information and house source text description information.

[0013] The structured information of the applicant and the house source is standardized and encoded to generate a structured feature vector.

[0014] Through a pre-defined policy label system, the policy provisions of affordable housing are converted into assignable and calculable policy rule features.

[0015] Optionally, in the attribute-level matching degree calculation module, the preset attribute dimensions include commuting convenience, living adaptability, rent pressure and policy compliance.

[0016] The attention mechanism dynamically generates an attribute weight vector for the current applicant-house source pair by calculating the interaction between the applicant feature vector and the house source feature vector, so that the importance of each attribute dimension in the final matching degree calculation can be personalized adjusted for applicants with different family structures or needs.

[0017] Optionally, in the multi-rule dynamic weighting engine, the priority rules are used to give higher allocation priority to specific policy groups or long-waiting-time applicants.

[0018] The constraint rules are used to set the upper limit of income, the matching relationship between family population and house type, and other hard conditions that cannot be violated;

[0019] The optimization rules are used to achieve the global goal of the highest overall housing turnover rate and balanced community population structure;

[0020] The rule conflict detection and resolution unit solves the decision-making conflicts between different rules through preset rule priority levels, weight voting, or manual arbitration.

[0021] Optionally, in the mixed allocation strategy execution module, the intelligent recommendation mode is used to recommend the top K sets of housing with the highest matching degree to the applicants for their selection;

[0022] The lottery balancing mode is used to determine the winning applicants in a random manner among multiple applicants whose matching degree scores fall within a preset adjacent interval, to ensure fairness;

[0023] The order-based construction response mode is used to analyze accumulated location and house type demand data, and when the number of pre-applications reaches a preset threshold, triggers the planning and construction process of new or renovated housing;

[0024] The resale adjustment market mode provides a compliant housing replacement platform within the system for families that have been in residence for a specified period of time and meet the conditions.

[0025] Optionally, the visual explanation report generated by the explainability output and strategy traceability module includes: displaying the composition of individual matching degree scores and dynamic weight allocation in the form of charts;

[0026] Listing the key rules affecting this allocation result and their contribution in the form of a list;

[0027] Tracing the complete decision-making path from application to allocation result determination in the form of a flowchart.

[0028] Optionally, the full-link logs recorded by the explainability output and strategy traceability module include operation logs, system decision logs, and strategy effect review databases;

[0029] The operation log records all manual operations and data access to the system;

[0030] The system decision log records the complete input, execution parameters, intermediate results, and final output of each round of allocation tasks;

[0031] The strategy effect review database is used to store allocation results and long-term residence feedback data in association, supporting policy and rule effect evaluation and tuning.

[0032] In a second aspect, a method for intelligent adaptation and dynamic allocation of security rental housing comprises the following steps:

[0033] S1: Extract and fuse the multi-modal information of the applicant and the housing source and the policy rules to generate a feature set and a rule set;

[0034] S2: Calculate the attribute-level matching degree of the applicant-housing source pair, and dynamically weight to obtain the comprehensive matching degree;

[0035] S3: Based on the multi-rule engine, the allocation scheme is weighted and comprehensively judged to generate a recommended scheme;

[0036] S4: According to the preset strategy, one or more allocation modes are called to execute the final allocation;

[0037] S5: Output the explainability report of the allocation result and record the full-link audit log.

[0038] Compared with the prior art, the present application has at least the following beneficial effects:

[0039] Through the multi-modal information extraction and feature fusion module, the system can analyze and fuse the semantic description, structured attributes and policy rules of the applicant at the same time, convert the unstructured policy provisions into computable features, lay a rich and aligned data foundation for subsequent accurate matching, and overcome the limitations of single information dimension and insufficient utilization in traditional methods.

[0040] The attribute-level matching degree calculation module introduces an attention mechanism, which can dynamically generate personalized weights for each attribute dimension (such as commuting, house type, rent, policy) according to the interaction between the applicant's features and the housing source's features, so that the comprehensive matching degree score can more truly reflect the core concerns and differentiated needs of different applicants, improving the relevance and rationality of the matching.

[0041] The multi-rule dynamic weighting engine builds a configurable rule library containing priority, constraint and optimization rules, and is equipped with a rule conflict detection and resolution unit. The system can automatically identify decision conflicts between rules, and resolve them according to the preset priority level, weight voting or manual arbitration when necessary, ensuring the compliance, consistency and efficiency of allocation decisions in complex policy environments.

[0042] The mixed allocation strategy execution module integrates intelligent recommendation, lottery balancing, order-based construction response, resale adjustment market and other modes, which can automatically select or combine execution according to real-time housing conditions, application queue and policy targets. This mixed strategy takes into account efficiency, fairness, demand guidance and resource revitalization, enabling the allocation system to upgrade from passive adaptation to active and flexible resource dynamic allocation network. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 The module connection diagram of the business rule-based security rental housing person-house matching system provided in the application is shown in the figure. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical scheme and advantages of the application more clear, the application is further described in detail below with reference to the accompanying drawings and embodiments. Figure 1 and embodiments.

[0045] The business rule-based security rental housing person-house matching system provided in the application comprises:

[0046] The multi-modal information extraction and feature fusion module extracts semantic features and structured features from the applicant information and the housing source information, and converts the housing security policy into a calculable rule feature, to generate an applicant feature set, a housing source feature set and a policy rule set;

[0047] The workflow of the multi-modal information extraction and feature fusion module comprises the following steps:

[0048] Step one: deep analysis of applicant information, the system receives various materials submitted by the applicant, such as application forms, identity proofs, income proofs, work proofs, etc. For text description information: for example, the applicant fills in “I have an elderly person at home, and need an elevator and low floor” or “I hope to have a short commuting time” in the “special needs” column. For this kind of free text, a semantic understanding model (such as Sentence-BERT) is used for interpretation. This model can understand the deep meaning behind the sentence, not just keyword matching, so as to convert a paragraph into a fixed-length digital sequence (semantic feature vector) that can represent its semantics;

[0049] For structured information: for example, family size, monthly income, place of household registration, application date, etc. The above information itself is an explicit number or category, and the system directly standardizes and encodes it into a structured numerical feature;

[0050] Step two: multi-dimensional description of housing source information, the system obtains comprehensive data of the housing source from the apartment management background. For housing description information, such as “the housing source is located in a mature community, surrounded by a vegetable market and a park” or “the room faces south and has excellent lighting”. Similarly, a semantic understanding model is used to convert these descriptions into semantic feature vectors;

[0051] For housing attribute information, such as specific building address (convertible to latitude and longitude coordinates), house type (number of rooms and number of halls), building area, floor, monthly rent, list of furniture and appliances, etc. These information is extracted and converted into structured numerical or vector;

[0052] Step three: build a policy tag system, the system administrator will enter the relevant policies of affordable housing into the system, and the module will "translate" them into computer executable logic;

[0053] For example, the "talent priority" policy: the system will define a rule to automatically scan the applicant's materials. If the applicant's career belongs to the identified talent directory and provides relevant proof, the system will give the applicant a "talent tag" and assign a corresponding priority weight in subsequent calculations.

[0054] For example, "low-income family priority": the system will compare the applicant's reported and verified income data with the set income line, and automatically give a "low-income family tag" to families that meet the conditions.

[0055] For example, "multi-child family priority": the system will verify the number of children in the family and give a "multi-child family tag" to families that meet the policy.

[0056] Other policies such as "waiting time points" and "local residence bonus" are converted into computable rule features in a similar manner. These "policy tags" themselves also serve as a strong feature, integrated into the applicant's overall feature vector.

[0057] Step four: feature alignment and fusion. After the above processing, each applicant and each set of housing are represented as a set of feature vectors containing multiple dimensions (semantic, structured, policy tag). The final task of this module is to prepare for subsequent matching calculations, ensuring that the applicant's "workplace" feature can be correctly calculated with the "geographical location" feature of the housing when calculating the "commuting distance matching degree"; and the applicant's "family structure" feature can be compared with the "house type" feature of the housing when calculating the "family adaptation degree".

[0058] The output of this module is the cleaned, structured, and quantified applicant feature set and housing feature set, as well as a set of activated and digitized policy rules. These results will be directly delivered to the next "attribute-level matching degree calculation module" as fuel and scales for its fine matching operations.

[0059] The attribute-level matching degree calculation module is connected to the multi-modal information extraction and feature fusion module, and calculates the single-item matching degree score of any applicant and any housing in each pre-set attribute dimension. It also dynamically generates personalized weights for different applicants in each attribute dimension based on the attention mechanism, and calculates the comprehensive matching degree score by weighted summation.

[0060] Specifically, the attribute-level matching degree calculation module includes the following steps:

[0061] Step 1: Establish a multi-dimensional scorecard for each pair of "applicant i - housing j" combination, and calculate the base score for the following key dimensions:

[0062] Commute distance score: The system calculates the commuting time or distance from housing j to applicant i's workplace based on map services. The shorter the time, the higher the score.

[0063] House type adaptation score: The system compares the number of bedrooms, bathroom configuration, and whether the elevator is equipped in housing j with the family size, age structure (such as whether there are old people or children) of applicant i. For example, a three-generation family matches a three-bedroom house type, and the score will be high; while matching a one-bedroom house, the score will be low;

[0064] Rent income ratio score: The system calculates the proportion of the monthly rent of housing j to the monthly income of applicant i's family. The lower the ratio, the stronger the rent affordability, and the higher the score; at the same time, the system will ensure that the ratio does not exceed the safety line specified by the policy;

[0065] Policy compliance score: The system checks the policy tags carried by applicant i, and if housing j is in a project or area with preferential policies for certain tags, this score will be significantly improved. For example, if applicant i has a "talent tag" and housing j belongs to a "talent apartment" project, this matching degree will be full marks;

[0066] Step 2: Dynamic weighting, simulating the flexible thinking process of human allocation officers, making the matching result more personalized and reasonable. The system analyzes the overall characteristics of the applicant and dynamically determines which item is more important (high weight) and which item is relatively less important (low weight) in this matching, for example: for a single applicant who has just graduated and works in the city center, the system will automatically increase the weight of "commute distance" because he may be most sensitive to commuting time. At the same time, it may appropriately reduce the weight of "house type adaptation" because he usually has low requirements for the number of rooms;

[0067] For example: for a family with multiple school-age children, the system will automatically increase the weight of "house type adaptation" and "surrounding school district", and the tolerance for commuting may be relatively high;

[0068] Step 3: Comprehensive Calculation and Generation of Final Matching Score. After scoring all sub-items and assigning dynamic weights to each sub-item, the system performs a weighted summation calculation. It multiplies the base score of each sub-item by its corresponding dynamic weight, then adds them together to obtain a comprehensive score representing the overall matching degree between applicant i and property j. For all applicants currently in the queue and all available rental properties, this module outputs a large matching matrix. Each number in this matrix precisely quantifies the degree of compatibility between a particular applicant and a particular property. This matrix will become the core basis for the next module to provide the final allocation decision.

[0069] A multi-rule dynamic weighted engine, connected to the attribute-level matching degree calculation module, includes a configurable rule base and a rule conflict detection and resolution unit; the rule base is used to store and execute priority rules, constraint rules, and optimization rules; the engine performs a weighted comprehensive decision on the allocation scheme based on the comprehensive matching degree score and the activated rules, and outputs a recommended allocation scheme.

[0070] The multi-rule dynamic weighting engine builds and maintains a dynamically editable rule base, in which rules are divided into three categories according to function and nature:

[0071] Priority rules: used to determine the order of allocation in the sequence.

[0072] Example of rule logic:

[0073] IF Applicant.Identity Identifier IN['Category A Talent', 'Category B Talent'] THEN Priority Weight = P1 (Highest)

[0074] If the applicant's waiting time is greater than 24, the applicant's basic matching score is increased by Q = (waiting time - 24) * Δ, where Δ is a preset scoring coefficient.

[0075] IF Applicant.Preferential Treatment Recipient Identifier = Truthen Priority Weight = P2

[0076] Constraint rules: These are used to set mandatory preconditions for allocation; failure to meet them results in a veto.

[0077] Example of rule logic:

[0078] If the applicant's average monthly family income per capita exceeds the policy threshold or income limit, then their eligibility for this allocation will be cancelled.

[0079] If applicant's family structure is 'single' and property's number of bedrooms is greater than 1, then matching is prohibited.

[0080] If the property's ownership status is not equal to 'available for rent', then it is prohibited from participating in the allocation process.

[0081] Optimization rule: used to guide the system to achieve certain global or long-term management goals under the premise of meeting priorities and constraints.

[0082] Rule logic example:

[0083] Objective function: MAXIMIZE (total vacancy reduction of this batch of allocated housing resources) to improve housing turnover efficiency.

[0084] Objective function: MINIMIZE (| community. The proportion of elderly families - target equilibrium value |) to promote community population structure balance.

[0085] Objective function: MAXIMIZE (applicant. The average commuting efficiency from work to the geographical location of the housing resource) to promote job-housing balance.

[0086] The multi-rule dynamic weighting engine has a rule conflict detection and resolution mechanism. Conflict detection is a static logic analysis and dynamic operation detection before executing rule logic. When two or more rules make mutually exclusive judgments on the same attribute of the same application object (for example, rule A requires priority allocation of a certain housing resource, and rule B prohibits allocation of the housing resource), a conflict flag is triggered.

[0087] Resolution strategies include: preset priority resolution, which presets a global priority level for all rules (such as: constraint rule > priority rule > optimization rule). When there is a conflict, the high-level rule overrides the low-level rule.

[0088] Weight arbitration resolution: for conflicts between rules of the same level (especially between multiple optimization targets), the system will perform weighted calculation according to the weight preset by the administrator for each rule, and output a compromise result.

[0089] Manual intervention channel: when automatic resolution fails or the conflict involves major policy interpretation, the system suspends the automatic allocation process, generates a conflict report and submits it to the administrator terminal, requesting manual adjudication.

[0090] The multi-rule dynamic weighting engine provides a rule management configuration interface, providing administrators with a graphical rule management interface, supporting the following dynamic configuration operations:

[0091] Rule enable / disable: administrators can activate or disable a rule with one click to address temporary policies or conduct effectiveness tests.

[0092] Rule parameter adjustment: administrators can modify the threshold and coefficient within the rule (such as P1, Δ, income upper limit value, target equilibrium value, etc.) through the interface.

[0093] Rule weight adjustment: administrators can adjust the relative importance (weight value) of each optimization rule in the weighted calculation in real time through a slider or input box.

[0094] Rule version management, the system records all the historical modified versions of the rules, the effective time and the operator, supports quick rollback to any historical version;

[0095] The final output of the multi-rule dynamic weighting engine is a compliance allocation scheme sequence evaluated and sorted by multiple rules, which reflects the policy priority and global optimization goal under the premise of meeting all hard constraints, and has solved potential rule conflicts. This sequence will be used as the input of the subsequent mixed allocation strategy module;

[0096] The mixed allocation strategy execution module is connected with the multi-rule dynamic weighting engine, and according to the housing situation, the application queue and the policy target, at least one of the allocation strategies including intelligent recommendation mode, lottery balance mode, order type construction response mode and resale / regulation market mode is called and executed to generate final allocation instructions;

[0097] Among them, the intelligent recommendation mode, i.e. normal precise matching, is suitable for the normal allocation period when the daily housing supply is relatively sufficient and the applicant's demand is diversified;

[0098] The working mode of intelligent recommendation mode is: according to the score obtained by the "matching degree calculation module", the system filters and lists the top 3-5 housing sources (i.e. Top-K recommendation list) with the highest matching degree for each applicant in the queue. The applicant will receive a notification through mobile phone message or APP: "Dear XX, according to your situation, the system has recommended the following X housing sources for you. Please log in to the system within 72 hours to confirm or select", the applicant can choose the one that he is most satisfied with in the recommendation list. If no selection is made within the specified time, the system will consider it as accepting the first recommended order, or the round will be postponed; the intelligent recommendation mode respects the choice of the applicant, improves the satisfaction and occupancy rate;

[0099] The application scenario of the lottery balance mode is when the popular housing sources in a batch are in short supply, and the matching scores of multiple applicants are very close and difficult to distinguish. This is an important fairness supplement to pure score sorting; the working mode of the lottery balance mode is: the system will define a "matching score interval" (for example, all applicants with scores between 85 and 90), and applicants in this score interval will be automatically included in a "candidate pool". Under public supervision (or through a reliable random algorithm), the system randomly selects the winners from the candidate pool. Applicants with small differences in scores and in front of the order no longer have absolute priority; the lottery balance mode avoids unfairness caused by small errors or boundary conditions in the model, and increases the public credibility and acceptance of the system;

[0100] Applicable scenario of order-based construction response mode: This is a proactive strategy to cope with medium and long-term demand and optimize resource allocation. It is triggered when the system discovers through big data analysis that the sustained application demand in a certain area (such as an industrial park) or a certain type of house (such as large houses suitable for three-child families) is much larger than the current stock of housing. The working mode of order-based construction response mode is: the system will regularly generate a "demand heat report" to identify "hot areas" and "shortage of housing types" that need to add housing sources. The housing management department can start a new affordable housing construction project or transform the existing property according to this report. The system can open a "pre-application" channel for this purpose, and applicants can "pre-book" these under-construction or planned housing sources in advance. When the number of registrations reaches the pre-set threshold for project start (for example, the recovery rate reaches 70%), the project is officially confirmed to start construction, effectively avoiding the risk of empty houses after completion. The order-based construction response mode upgrades the allocation system from "passive adaptation of stock" to "active guidance of increment", achieving data-driven precise construction and efficient use of resources;

[0101] The applicable scenario of the resale / adjustment market mode is to meet the housing needs of families who have moved into the system due to changes in family size, job changes, etc., while ensuring that housing sources are always circulating within the system. The working mode of the resale / adjustment market mode is: the system provides an internal "housing adjustment platform" for families that have lived for a certain period of time (such as 5 years) and meet the policy requirements. Families who want to change their homes can post "outbound" information on the platform and look for their desired "inbound" housing sources. All adjustment transactions must be completed on the platform and comply with government guidelines and eligibility checks. The system automatically completes contract changes, right transfers, and record registration to ensure that transactions are transparent, transparent, and compliant. The resale / adjustment market mode meets the reasonable life cycle housing needs of residents while strictly controlling the properties of affordable housing, making the entire housing security system "alive" and improving the long-term use efficiency of housing sources;

[0102] The mixed allocation strategy execution module automatically selects the mode according to real-time conditions. For example, for the first 80 of the 100 new housing sources, intelligent recommendation may be used; for the last 20 particularly popular ones, a lottery balance is started in the high-score interval. At the same time, the back-end continuously runs the order-based construction response analysis and provides adjustment market services for existing residents; this hybrid strategy makes the allocation system no longer a single pipeline, but a three-dimensional, dynamic, and flexible resource allocation network;

[0103] The interpretable output and strategy tracing module generates a visual explanation report for each allocation result and records the system's full-link operation and decision log;

[0104] Specifically, the explainable output and strategy traceability module automatically generates a structured explanation report for each generated allocation scheme (or recommendation list). This report is presented in both visual interface and data table form, and includes the following core content:

[0105] Matching degree composition analysis, showing the applicant's and target housing's scores in each attribute dimension in chart form;

[0106] Rule influence weight explanation, detailing the key rules that affect this allocation ranking or result and their impact in list form;

[0107] Allocation path traceability, restoring the entire allocation decision-making process for the applicant in timeline or flowchart form;

[0108] The explainable output and strategy traceability module records all key operations and state changes in an unalterable log, building a complete audit tracking chain to provide full-process audit logs;

[0109] Full-process audit logs include operation logs, which record the operator ID, operation time, operation terminal IP, operation module, execution action (such as "manually adjust rule weight", "reject allocation scheme"), data snapshots before and after the operation, and access control for all sensitive data and configuration functions;

[0110] System decision logs, which record the unique ID of each allocation task (Batch), task execution time, range of applicants and housing involved, adopted allocation strategy mode, strategy execution parameters, generated intermediate results (such as matching degree matrix), and final output scheme;

[0111] Decision logs and final explanation reports are strongly associated by application number or task ID, supporting one-key query;

[0112] Strategy effect review database, regularly associates and stores allocation results with actual feedback data (such as rent-out rate, satisfaction survey, community population structure change), providing analysis tools for administrators to evaluate the long-term effects of different rule weights or allocation strategies in actual operation (e.g., whether "increasing community balance weight" effectively promotes community integration), and provides data basis for future policy and rule optimization;

[0113] The output of the explainable output and strategy traceability module includes:

[0114] For applicants / public, personalized explanation reports can be pushed through the APP or portal, making them clearly understand the basis behind the matching results;

[0115] For the administrator / auditor, a full-featured audit query background can be provided to support penetration query and export of logs according to time, personnel, application number, rules and other dimensions;

[0116] For decision makers, regular policy effect review analysis reports can be provided to present policy implementation effectiveness and system operation efficiency in the form of charts and data.

[0117] In one embodiment, a method for intelligent adaptation and dynamic allocation of security rental housing is provided, comprising the following steps:

[0118] S1: Extract and fuse the multi-modal information of the applicant and the housing source and the policy rules to generate a feature set and a rule set;

[0119] S2: Calculate the attribute-level matching degree of the applicant-housing source pair, and dynamically weight to obtain the comprehensive matching degree;

[0120] S3: Based on the multi-rule engine, the allocation scheme is weighted and comprehensively judged to generate a recommended scheme;

[0121] S4: According to the preset strategy, one or more allocation modes are called to execute the final allocation;

[0122] S5: Output the explainability report of the allocation result and record the full-link audit log.

[0123] The technical features of the above embodiments can be combined arbitrarily. In order to make the description simple, not all possible combinations of the technical features in the above embodiments are described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

Claims

1. A system for matching people with housing in a guaranteed rental housing based on business rules, characterized in that, The application comprises: A multi-modal information extraction and feature fusion module extracts semantic features and structured features from applicant information and housing information, and converts housing security policies into computable rule features to generate an applicant feature set, a housing feature set, and a policy rule set; An attribute-level matching degree calculation module connected with the multi-modal information extraction and feature fusion module calculates the single-item matching degree score of any applicant and any housing in each preset attribute dimension, and dynamically generates individualized weights for different applicants in each attribute dimension based on an attention mechanism. The comprehensive matching degree score is obtained by weighted summation; A multi-rule dynamic weighting engine connected with the attribute-level matching degree calculation module includes a configurable rule base, a rule conflict detection and resolution unit. The configurable rule base is used to store and execute priority rules, constraint rules, and optimization rules. The engine performs weighted comprehensive decision on the allocation scheme based on the comprehensive matching degree score and the activated rules, and outputs a recommended allocation scheme; A hybrid allocation strategy execution module connected with the multi-rule dynamic weighting engine calls and executes at least one allocation strategy including intelligent recommendation mode, lottery balance mode, order type construction response mode, and resale adjustment market mode according to the housing situation, application queue, and policy target to generate final allocation instructions; An interpretable output and strategy tracing module generates a visual explanation report for each allocation result and records the system full-link operation and decision log.

2. The business rules based secure rental housing matching system of claim 1, wherein, In the multi-modal information extraction and feature fusion module, a sentence embedding model is used to extract semantic feature vectors of applicant text description information and housing text description information; The structured information of the applicant and the housing is standardized coded to generate a structured feature vector; Through a pre-defined policy label system, the policy provisions of affordable housing are converted into assignable and computable policy rule features.

3. The business rules based secure rental housing matching system of claim 1, wherein, In the attribute-level matching degree calculation module, the preset attribute dimensions include commuting convenience, residential adaptability, rent pressure, and policy compliance; The attention mechanism dynamically generates an attribute weight vector for the current applicant-housing pair by calculating the interaction between the applicant feature vector and the housing feature vector, so that the importance of each attribute dimension in the final matching degree calculation can be personalized adjusted for applicants with different family structures or needs.

4. The business rules based secure rental housing matching system of claim 1, wherein, In the multi-rule dynamic weighting engine, the priority rules are used to give higher allocation priority to specific policy groups or long-waiting-time applicants; The constraint rules are used to set the upper limit of income, the matching relationship between family population and house type, and other hard conditions that cannot be violated; The optimization rules are used to achieve the global goal of the highest overall housing turnover rate and balanced community population structure; The rule conflict detection and resolution unit solves the decision conflicts among different rules through preset rule priority levels, weight voting, or triggering manual arbitration.

5. The business rules based secure rental housing matching system of claim 1, wherein, In the hybrid allocation strategy execution module, the intelligent recommendation mode is used to recommend the top K housing with the highest matching degree to the applicant for selection. The draw balance mode is used to determine the winning applicant in a random manner among multiple applicants whose matching scores fall into a preset adjacent interval, so as to ensure fairness. The order-based construction response mode is used to analyze accumulated data of location and house type requirements, and when the number of pre-applications reaches a preset threshold, the planning and construction process of new or renovated housing sources is triggered. The resale adjustment market mode provides a housing replacement platform in compliance with the system for families that have lived in the housing for a specified period of time and meet the conditions.

6. The business rules based secure rental housing matching system of claim 1, wherein, The visual explanation report generated by the explainability output and strategy traceability module includes: displaying the composition of individual matching scores and dynamic weight allocation in the form of a chart. Listing the key rules affecting the allocation results and their contribution in the form of a list. Tracing the complete decision-making path from application to allocation result determination in the form of a flowchart.

7. The business rules based secure rental housing matching system of claim 1, wherein, The full-link log recorded by the explainability output and strategy traceability module includes operation logs, system decision logs, and policy effect review databases. The operation log records all manual operations and data access to the system. The system decision log records the complete input, execution parameters, intermediate results, and final output of each round of allocation tasks. The policy effect review database is used to store the allocation results and long-term occupancy feedback data, supporting the evaluation and optimization of policies and rules.

8. A method for intelligent adaptation and dynamic allocation of security rental housing based on the system of any one of claims 1-7, characterized in that, The method comprises the following steps: S1: Extract and fuse the multi-modal information of applicants and housing sources and policy rules to generate feature sets and rule sets; S2: Calculate the attribute-level matching degree of applicant-housing source pairs and dynamically weight to obtain the comprehensive matching degree; S3: Make a weighted comprehensive decision on the allocation scheme based on a multi-rule engine to generate a recommended scheme; S4: Call one or more allocation modes according to the preset strategy to perform the final allocation; S5: Output the explainability report of the allocation results and record the full-link audit logs.

Images