A method and system for searching and recommending second-hand housing information

Abstract

The application discloses a second-hand house source information retrieval and recommendation method and system, relates to the technical field of information retrieval and recommendation, and aims to solve the problem that users are trapped in repeated trial and error but cannot obtain a stable candidate house source set in the second-hand house source information retrieval and recommendation of the prior art. The method comprises the following steps: acquiring a retrieval demand input by a user, and constructing a composite demand state diagram; quantitatively evaluating the constraint strength of demand conditions corresponding to each node in the composite demand state diagram; and dividing the demand conditions into hard boundary conditions, preference boundary conditions and reference boundary conditions according to the evaluation results.

Description

Technical Field

[0001] This application relates to the field of information retrieval and recommendation technology, and in particular to a method and system for retrieving and recommending second-hand housing information. Background Technology

[0002] Existing technologies for retrieving and recommending secondhand housing information typically employ keyword search and structured filter components to help users search through massive housing databases. This approach is effective in narrowing the search scope and locating target properties when users have clear needs and some understanding of market supply characteristics. However, for users using the platform for the first time or lacking market knowledge, they often bring a holistic, lifestyle-oriented concept of housing, such as "a quiet two-bedroom apartment suitable for elderly people to take children to and from school," rather than discrete database fields. In this case, there is a significant communication barrier between the user's vague, lifestyle-oriented needs and the system's standardized housing attribute fields.

[0003] Problems arise when users attempt to forcibly disassemble this indivisible holistic concept and map it onto the platform's independent and unrelated structured filtering criteria. Because the existing system defaults to treating each filtering criterion as an equally important and rigid constraint that must be met simultaneously, users, unaware of the actual market distribution of properties behind each criterion, are highly likely to create mutually exclusive or overly restrictive combinations of conditions. For example, a user might simultaneously select "below the fifth floor" and "low total price," without realizing that this might be an empty set in a specific area, or thus incorrectly exclude highly relevant properties with elevators but on slightly higher floors.

[0004] As users repeatedly try adding or removing conditions, the system mechanically executes strict database intersection queries, causing search results to fluctuate wildly between a massive number of irrelevant listings and zero results. Existing conventional search methods struggle to identify the priorities and inherent life connections behind users' various filtering conditions, and fail to transparently convey to users the true constraints of market supply on specific combinations of conditions. This traps users in a dilemma of repeated trial and error without obtaining a stable set of candidate listings, forcing the search task to be interrupted. Summary of the Invention

[0005] This application provides a method and system for retrieving and recommending second-hand housing information. It aims to solve the problems in existing technologies for retrieving and recommending second-hand housing information, such as the connection barrier between users' vague expressions of their daily needs and the standardized housing attribute field system in the system, and the fact that users, without knowing the actual market housing distribution behind each condition, are prone to making mutually exclusive or overly tight combinations of conditions. This causes the search results to fluctuate wildly between a large number of irrelevant housings and zero results, leaving users in a dilemma of repeated trial and error without obtaining a stable set of candidate housings.

[0006] Firstly, to address the aforementioned technical problems, this invention provides a method for retrieving and recommending secondhand housing information. This method includes: acquiring user-inputted search requirements; extracting textual expressions of intent and discretized filtering conditions from the search requirements; constructing a composite demand state graph, where nodes represent demand intents and edges represent logical connections between demand intents; quantifying the constraint strength of the demand conditions corresponding to each node in the composite demand state graph; and classifying the demand conditions into hard boundary conditions, preference boundary conditions, and reference boundary conditions based on the evaluation results; performing an executability test on the search condition combination formed by the combination of hard boundary conditions and preference boundary conditions based on the current housing supply status; and locating the condition item causing the candidate housing size to exceed the preset reasonable range when the test result indicates that the candidate housing size exceeds the preset reasonable range; and performing a hierarchical reconstruction process based on the located condition item and its constraint strength, adjusting the constraint level or value boundary of some conditions in the search condition combination to form reconstructed search conditions. These reconstructed search conditions are used to initiate a search in the housing database to obtain a set of candidate housings and to generate recommendation results.

[0007] Secondly, this application provides a second-hand housing information retrieval and recommendation system, which includes: an acquisition unit, used to acquire user-input retrieval requirements, extract the textual expressions of lifelike intent and discretized filtering conditions contained in the retrieval requirements, and construct a composite demand state graph, wherein nodes represent demand intent and edges represent the logical connections between demand intents; an evaluation unit, used to quantitatively evaluate the constraint strength of the demand conditions corresponding to each node in the composite demand state graph, and divide the demand conditions into hard boundary conditions, preference boundary conditions and reference boundary conditions according to the evaluation results; a processing unit, used to perform an executability test on the retrieval condition combination formed by the combination of hard boundary conditions and preference boundary conditions according to the current housing supply status, and when the test result indicates that the candidate housing scale corresponding to the retrieval condition combination exceeds the preset reasonable range, locate the condition item that causes the candidate housing scale to exceed the preset reasonable range; and an adjustment unit, used to perform hierarchical successive reconstruction processing according to the located condition item and its constraint strength, adjust the constraint level or value boundary of some conditions in the retrieval condition combination, form reconstructed retrieval conditions, and use the reconstructed retrieval conditions to initiate a retrieval in the housing database to obtain a set of candidate housing, and use them to generate recommendation results.

[0008] This application has at least the following beneficial effects: The second-hand housing information retrieval and recommendation method disclosed in this application obtains the user's input search requirements and constructs a composite demand state graph by combining the user's lifestyle intent expression and discretized filtering conditions. Nodes represent demand intents, and edges represent the logical connections between demand intents, thereby effectively connecting the user's vague lifestyle needs with the system's standardized housing attributes. Based on this, the constraint strength of the demand conditions corresponding to each node in the composite demand state graph is quantitatively evaluated, and the demand conditions are divided into hard boundary conditions, preference boundary conditions, and reference boundary conditions according to the evaluation results. This solves the problem of rigid constraints in existing technologies that treat all filtering conditions as equally important and must be satisfied simultaneously, enabling the system to handle user needs more flexibly. Furthermore, based on the current housing supply status, the feasibility of the search condition combinations formed by hard boundary conditions and preference boundary conditions is tested. When the test results indicate that the number of candidate properties corresponding to the search condition combination exceeds a preset reasonable range, the condition item causing the number of candidate properties to exceed the preset reasonable range is identified. This effectively avoids the problem that users, without knowing the actual market housing distribution behind each condition, may make mutually exclusive or overly tight condition combinations, leading to drastic fluctuations in search results between a massive number of irrelevant properties and zero results. Finally, based on the identified condition items and their constraint strength, a hierarchical and sequential reconstruction process is performed to adjust the constraint level or value boundary of some conditions in the search condition combination, forming reconstructed search conditions. These reconstructed search conditions are used to initiate a search in the housing database to obtain a set of candidate properties and to generate recommendation results. This allows users to obtain a stable set of candidate properties that meets their needs, overcoming the dilemma in existing technologies where users are trapped in repeated trial and error without obtaining a stable set of candidate properties. In summary, this application, through the above-mentioned technical solution, effectively solves the problem of the connection between user needs and the standardized attributes of the system in the prior art, as well as the problem of poor search results caused by improper combination of conditions, and significantly improves the efficiency and user experience of second-hand housing information retrieval and recommendation. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating a method for retrieving and recommending secondhand housing information provided in this application; Figure 2 This is a flowchart illustrating another method for retrieving and recommending secondhand housing information provided in this application; Figure 3 This is a schematic diagram of the structure of a second-hand housing information retrieval and recommendation system provided in this application. Detailed Implementation

[0010] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0011] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0012] Traditional methods for retrieving and recommending secondhand housing information typically employ keyword searches and structured filter components to help users search through massive housing databases. This approach is effective in narrowing down the search scope and locating target properties when users have clear needs and some understanding of market supply characteristics. However, for users using the platform for the first time or lacking market knowledge, they often bring a holistic, lifestyle-oriented concept of living, rather than discrete database fields. At this point, there is a significant communication barrier between the user's vague, lifestyle-oriented needs and the system's standardized housing attribute fields. Problems arise when users attempt to forcibly break down this indivisible holistic concept and map it to the platform's independent, unrelated structured filter conditions. Because existing systems default to treating each filter condition as equally important and a rigid constraint that must be met simultaneously, users, unaware of the actual market housing distribution behind each condition, are prone to making mutually exclusive or overly restrictive combinations of conditions. For example, a user might simultaneously select "below the fifth floor" and "low total price," without realizing that this might be an empty set in a specific area, or thus incorrectly excluding highly relevant properties with elevators but on slightly higher floors. As users repeatedly try adding or removing conditions, the system mechanically executes strict database intersection queries, causing search results to fluctuate wildly between a massive number of irrelevant listings and zero results. Existing conventional search methods struggle to identify the priorities and inherent life connections behind users' various filtering conditions, and fail to transparently convey to users the true constraints of market supply on specific combinations of conditions. This traps users in a dilemma of repeated trial and error without obtaining a stable set of candidate listings, forcing the search task to be interrupted.

[0013] In view of the above problems, this application provides a method for retrieving and recommending second-hand housing information. The core of this method is to deeply analyze the search requirements input by users and dynamically adjust the search strategy according to the analysis results in order to adapt to the complex market supply and the ever-changing needs of users.

[0014] The following specific embodiments will provide a detailed introduction and explanation of the second-hand housing information retrieval and recommendation method provided in this application.

[0015] Reference Figure 1 This application provides a method for retrieving and recommending second-hand housing information, which may include the following steps: S101. Obtain the user's input search requirements, extract the textual expressions of lifelike intent and discrete filtering conditions contained in the search requirements, and construct a composite requirement state diagram.

[0016] In the composite demand state diagram, nodes represent demand intentions, and edges represent the logical relationships between demand intentions.

[0017] First, this method requires obtaining the user's input search requirements. Users can input their search requirements in various ways, such as by entering natural language descriptions in a text box, like "I want a quiet two-bedroom apartment suitable for elderly people to pick up and drop off children at school," or by selecting preset filter options, such as "two bedrooms and one living room," "school district housing," or "along the subway line." The system will extract the textual expressions of user intent and discrete filter conditions from these inputs. The expressions of user intent refer to the user's needs described in natural language, including life scenarios and emotional inclinations, such as "quiet" or "convenient for picking up and dropping off children." Discrete filter conditions refer to the conditions selected by the user through a structured interface, with clearly defined value ranges or options, such as "two bedrooms and one living room" or "area 90-120 square meters."

[0018] After acquiring and extracting this information, the system constructs a composite demand state graph. This graph is a graphical representation where nodes represent user intents, and edges represent the logical relationships between these intents. For example, "quiet" can be a node, and "school district housing" can also be a node. If a user expresses both needs simultaneously, there might be an edge between them, indicating that they are related within the user's overall needs. The weight of the edge can be used to represent the strength or priority of this relationship.

[0019] S102. Quantitatively evaluate the constraint strength of the demand conditions corresponding to each node in the composite demand state diagram, and classify the demand conditions into hard boundary conditions, preferred boundary conditions, and reference boundary conditions based on the evaluation results.

[0020] Next, the system will quantitatively evaluate the constraint strength of the demand conditions corresponding to each node in the composite demand state diagram. This means that the system will analyze the degree of impact of each demand condition on the final housing selection. For example, "must have an elevator" may be a condition with high constraint strength, while "preferably south-facing" may be a condition with low constraint strength. Based on the evaluation results, these demand conditions will be divided into three categories: hard boundary conditions, preferred boundary conditions, and reference boundary conditions. Hard boundary conditions are conditions that users believe must be met, such as "two bedrooms and one living room"; preferred boundary conditions are conditions that users hope to meet but can appropriately relax, such as "floor below the fifth floor"; reference boundary conditions are conditions that users only use as a reference and can adjust or ignore at any time, such as "high greening rate in the community".

[0021] S103. Based on the current housing supply status, perform an executability test on the combination of search conditions formed by the combination of hard boundary conditions and preference boundary conditions, and when the test result indicates that the size of candidate housing corresponding to the combination of search conditions exceeds the preset reasonable range, locate the condition item that causes the size of candidate housing to exceed the preset reasonable range.

[0022] Subsequently, the system performs an executability check on the search condition combinations formed by the combination of hard boundary conditions and preference boundary conditions, based on the current housing supply status. This step is to avoid situations where the search conditions are too strict, resulting in no matching properties, or too lenient, leading to an excessive number of properties that are difficult to filter. The system estimates the number of properties that meet these conditions under the current market supply. If the test results indicate that the number of candidate properties corresponding to the search condition combination exceeds a preset reasonable range (e.g., too few or too many properties), the system will identify the condition that causes the number of candidate properties to exceed the preset reasonable range. For example, if the combination of the conditions "below the fifth floor" and "low total price" results in too few properties, the system will identify these two conditions as the cause of the problem.

[0023] S104. Based on the located condition items and their constraint strength, perform hierarchical succession reconstruction processing, adjust the constraint level or value boundary of some conditions in the search condition combination, and form reconstructed search conditions.

[0024] Among them, the reconstructed search criteria are used to initiate a search in the housing database to obtain a set of candidate housing listings, and are used to generate recommendation results.

[0025] Finally, based on the identified conditions and their constraint strengths, the system performs a hierarchical restructuring process. This process aims to intelligently adjust the constraint level or value boundaries of some conditions in the combination of search conditions to form restructured search conditions. For example, if "below the fifth floor" results in too few listings, the system might adjust it to "below the seventh floor," or downgrade it from a preference boundary condition to a reference boundary condition. The restructured search conditions will then be used to initiate a search in the housing database to obtain a set of candidate listings, and ultimately used to generate recommendation results.

[0026] Through the above process, this application effectively solves the problem of the connection between users' fuzzy needs and standardized screening conditions in traditional search methods, as well as the problem of unstable search results due to improper combination of conditions. The system no longer mechanically performs strict database intersection queries, but provides users with more stable and accurate housing recommendations through intelligent demand parsing and condition reconstruction.

[0027] In some embodiments described above in this application, reference is made to Figure 2 To more accurately understand and process users' complex property search needs, it is necessary to construct a composite demand state diagram from the user's input search requirements. Specifically, the above-mentioned construction of a composite demand state diagram includes: S201. Call the preset real estate life knowledge graph, identify the life scene entities contained in the life intent expression, and map the life scene entities to multiple standardized housing attribute entities associated with the knowledge graph, so as to obtain the mapping path between the life scene entities and the standardized housing attribute entities and the association weight of each mapping path.

[0028] Among them, the mapping path and its associated weights are used as the basis for determining the edge weights between corresponding nodes in the composite demand state diagram.

[0029] S202, The discretized filtering conditions extracted from the search requirements are used as explicit standardized property attribute entities.

[0030] Its core function is to provide data sources and judgment criteria for the calculation of "structured confirmation degree". The "discrete filtering conditions" selected by the user on the interface (such as "south-facing", total price range, etc. selected from the drop-down menu) are directly used as "explicit standardized property attribute entities" to participate in the construction of the state graph. This differs from the existence of "life scene entities" and "standardized property attribute entities" mapped from natural language through knowledge graphs, which both exist as nodes in the graph, but their origins are different. When calculating the "comprehensive constraint strength value", three dimensions need to be evaluated, including "structured confirmation degree". This application explicitly defines: "The structured confirmation degree is determined based on whether the demand originates from the discrete filtering conditions". This means that if a demand condition is explicitly set by the user by clicking on each item in the structured filter, its certainty and importance are generally considered higher than demands inferred from fuzzy natural language. This application explicitly marks the "discretized screening conditions" as "explicitly standardized housing attribute entities" in order to assign a higher "structured confirmation" score to the node by tracing its source (whether it was directly selected or mapped from text) in subsequent calculations, thereby affecting the final constraint strength assessment and condition stratification.

[0031] S203. Using standardized housing attribute entities and living scenario entities as nodes of the composite demand state graph, and using the logical relationships between entities determined by semantic analysis of the expression of living intentions as edges of the composite demand state graph, and when it involves standardized housing attribute entities mapped from living scenario entities, the weight of the corresponding edge is determined according to the association weight of the mapping path to construct the composite demand state graph.

[0032] Specifically, the real estate lifestyle knowledge graph is a structured knowledge base containing a large number of real estate-related concepts, entities, and their interrelationships, such as "school district housing" and "school distance," and "subway housing" and "subway station distance." This knowledge graph helps the system understand the user's textual expression of lifestyle intent. For example, when a user expresses "wanting a school district housing," the system can identify "school district housing" as a lifestyle scenario entity through the knowledge graph and map it to standardized housing attribute entities such as "school distance" and "educational resources." The mapping path represents the specific transformation from the lifestyle scenario entity to the standardized housing attribute entity, while the association weight quantifies the strength or importance of this transformation. For example, the association weight between "school district housing" and "distance to a key primary school" may be higher than the association weight between "school district housing" and "distance to an ordinary middle school." These mapping paths and association weights provide the foundation for subsequently constructing the edge weights in the composite demand state graph.

[0033] Discretized filtering conditions refer to filtering conditions provided directly by users in a structured form, such as "price range: 3-5 million yuan", "apartment type: three bedrooms and two living rooms", "region: Haidian District", etc. Due to their explicitness and structured characteristics, these conditions can be directly regarded as explicit and standardized property attribute entities without the need for mapping through a knowledge graph.

[0034] In practical applications, the nodes of the composite demand state diagram consist of the standardized housing attribute entities and life scenario entities mentioned above. For example, "school district housing," "subway housing," "three bedrooms and two living rooms," and "Haidian District" can all be nodes in the diagram. The edges in the diagram represent the logical relationships between these entities. This logical relationship is determined through semantic analysis of the life-related intention expressions. For example, if a user expresses "wanting a school district housing in Haidian District," then there may be a logical relationship edge between "Haidian District" and "school district housing." When the entity connected by the edge contains a standardized housing attribute entity mapped from the life scenario entity, the weight of the edge is determined according to the association weight of the corresponding mapping path. For example, if "school district housing" maps to "school distance," then in the composite demand state diagram, the edge weight between the "school district housing" node and the "school distance" node will be determined by the association weight of that mapping path.

[0035] The above technical solution effectively integrates user-inputted textual expressions of everyday intent with discrete filtering conditions, forming a structured and quantifiable composite demand state graph. This graph not only clearly represents the user's demand intent but also reflects the importance and interrelationships between different needs through logical connections between nodes and edge weights. This lays a solid foundation for subsequent assessment of demand constraint strength and reconstruction of search conditions, significantly improving the understanding of complex user needs and the accuracy of property matching. Specifically, by introducing a real estate lifestyle knowledge graph, the system can deeply understand the real needs behind the user's lifestyle intent, avoiding matching biases caused by insufficient semantic understanding in traditional methods. Simultaneously, by utilizing the weights of mapping path associations, the importance of different demand intents in the composite demand state graph is quantified, further improving the refinement of demand representation.

[0036] Specifically, the above quantitative evaluation of the constraint strength of the demand conditions corresponding to each node in the composite demand state diagram, and the division of demand conditions into hard boundary conditions, preferred boundary conditions and reference boundary conditions based on the evaluation results, can be implemented in the following way.

[0037] The above quantitatively evaluates the constraint strength of the demand conditions corresponding to each node in the composite demand state diagram, and classifies the demand conditions into hard boundary conditions, preferred boundary conditions, and reference boundary conditions based on the evaluation results, including: Based on the semantic emphasis, structured confirmation, and life logic relevance of the corresponding demand conditions of each node, the comprehensive constraint strength value of each node is determined by weighted calculation. The semantic emphasis is determined by the strength of the wording or the frequency of modification when the user expresses the demand. The structured confirmation is determined by whether the demand comes from discrete screening conditions. The life logic relevance is determined by whether multiple different life scenario entities in the composite demand state diagram point to the same standardized housing attribute entity. Demand conditions with a comprehensive constraint strength value higher than the first strength threshold are classified as hard boundary conditions, demand conditions with a comprehensive constraint strength value between the first and second strength thresholds are classified as preferred boundary conditions, and demand conditions with a comprehensive constraint strength value lower than the second strength threshold are classified as reference boundary conditions.

[0038] Specifically, when quantifying the constraint strength of the demand conditions corresponding to each node in the composite demand state diagram, multiple dimensions are considered. Semantic emphasis refers to the intensity of language used by users when expressing a specific demand or the frequency with which they modify that demand. For example, if a user uses words like "must" or "certainly," or repeatedly modifies a condition, it indicates that the condition has a high semantic emphasis. Structured confirmation focuses on the source of the demand conditions. If the demand directly originates from the user's explicitly input discrete filtering conditions (such as "three bedrooms and two living rooms" or "subway access"), its structured confirmation is high, indicating that the user has a clear and definite intention. Life logic relevance is determined by analyzing the composite demand state diagram. If multiple different life scenario entities (such as "school district housing" and "convenient transportation") point to the same standardized housing attribute entity (such as "geographical location"), it indicates that this attribute has a high relevance in the user's life logic.

[0039] The above three dimensions of indicators are used to jointly determine the comprehensive constraint strength value of each requirement node through weighted calculation. The weights of each dimension can be preset or dynamically adjusted according to actual application scenarios and experience.

[0040] After obtaining the overall constraint strength value, the requirement conditions are categorized according to preset strength thresholds. Specifically, when the overall constraint strength value is higher than the first strength threshold, the requirement condition is classified as a hard boundary condition. Hard boundary conditions typically represent the user's core, non-negotiable needs, such as "must have three bedrooms." When the overall constraint strength value is between the first and second strength thresholds, the requirement condition is classified as a preference boundary condition. Preference boundary conditions represent user preferences that can be moderately adjusted, such as "preferably with cross ventilation." When the overall constraint strength value is lower than the second strength threshold, the requirement condition is classified as a reference boundary condition. Reference boundary conditions are usually supplementary information provided by the user, and their binding force is weaker, such as "there is a coffee shop nearby."

[0041] Through the above technical solution, this application enables a refined and multi-dimensional quantitative assessment of user needs, thereby overcoming the potential one-sidedness of traditional methods in understanding user needs. By dividing the needs conditions into hard boundary conditions, preference boundary conditions, and reference boundary conditions, the system can more accurately identify the core, preferences, and auxiliary information of user needs, providing a solid foundation for the feasibility testing of subsequent search condition combinations and hierarchical reconstruction processing. This hierarchical management not only improves the depth and breadth of the system's understanding of user needs, but also allows for more intelligent and flexible adjustment of search strategies when there is a shortage or surplus of housing supply. This avoids the problem of overly narrow or broad search results caused by constraints of a single dimension, significantly improving the accuracy of second-hand housing information retrieval and recommendation, and enhancing the user experience.

[0042] In some of the embodiments described above in this application, a feasibility test is proposed for combining search criteria based on the current housing supply status, and the criteria that cause the size of candidate housing units to exceed a preset reasonable range are identified. However, in practical applications, if the specific criteria that caused this abnormal size cannot be accurately and efficiently identified when the size of candidate housing units exceeds the preset reasonable range, the subsequent criteria reconstruction process may lack focus, or may even lead to unnecessary relaxation or adjustment of criteria, thereby affecting user experience and recommendation effectiveness.

[0043] In response, this application further proposes the following steps: performing an executability test on the combination of hard boundary conditions and preference boundary conditions based on the current housing supply status; and identifying the condition item causing the candidate housing size to exceed a preset reasonable range when the test result indicates that the candidate housing size exceeds a preset reasonable range. The preset cardinality estimation engine is invoked to estimate the number of properties that can be matched by the combination of search conditions under the current property supply status; Determine whether the estimated number of housing units falls within a preset reasonable range. If the estimated number of housing units is lower than the lower limit of the preset reasonable range, it is determined that the candidate housing unit scale is too narrow. If the estimated number of housing units is higher than the upper limit of the preset reasonable range, it is determined that the candidate housing unit scale is too wide. When the candidate housing stock size is too narrow, the key bottleneck conditions causing the narrowing are located by temporarily removing the preference boundary conditions in the search condition combination and observing the marginal change rate of the estimated housing stock number after removal. By querying the joint probability distribution matrix of housing stock attributes, conflict condition pairs with a coexistence probability lower than a preset probability threshold under the current supply state are identified. The key bottleneck conditions or the condition items in the conflict condition pairs are used as the condition items that cause the candidate housing stock size to exceed the preset reasonable range.

[0044] Specifically, the cardinality estimation engine can be understood as a pre-trained model or algorithm module that can quickly and accurately predict the number of properties that can be matched under a given set of search conditions, based on historical housing data and current market supply. The engine aims to efficiently assess the potential housing stock without actually performing a database search, thus avoiding resource waste caused by overly stringent or lenient search conditions. The preset reasonable range refers to a range of housing stock numbers set by the system based on experience or user preferences, such as 50 to 200 units. When the estimated number of properties is lower than the lower limit of this range, it indicates that the candidate housing stock is too narrow, potentially leaving users with no options; when the estimated number of properties is higher than the upper limit of this range, it indicates that the candidate housing stock is too wide, potentially causing users difficulty in making a choice.

[0045] In practical applications, when the pool of candidate properties is too narrow, the system employs two strategies to pinpoint problematic conditions. The first strategy is "marginal rate of change analysis," where the system temporarily removes a preference boundary condition from the current search criteria combination and re-estimates the number of properties after removal using a cardinality estimation engine. By comparing the rate of change in the number of properties before and after removal, the preference boundary condition that has the greatest impact on the property pool size—the "critical bottleneck condition"—can be identified. For example, if removing the "school district" condition significantly increases the number of properties, then "school district" may be the critical bottleneck condition causing the narrow pool. The second strategy is "conflicting condition pair identification," where the system queries a pre-defined joint probability distribution matrix of property attributes. This matrix records the probability of different property attributes coexisting in the current market. Through analysis, condition pairs with a coexistence probability below a pre-defined threshold under the current supply conditions can be identified. These condition pairs are called "conflicting condition pairs," and their existence may lead to a sharp decrease in the number of properties. For example, "low-floor" and "with elevator" might be a conflicting condition pair in some older residential areas. Ultimately, whether it is the key bottleneck condition located through marginal rate of change analysis or the condition terms in the conflict condition pairs identified through the joint probability distribution matrix, they will all be determined as the condition terms that cause the candidate housing scale to exceed the preset reasonable range, providing clear adjustment targets for subsequent condition reconstruction.

[0046] Through the aforementioned technical solution, this application effectively addresses the limitation of traditional methods in accurately identifying specific problem conditions when the number of available properties is abnormal. By introducing a cardinality estimation engine, it achieves rapid prediction of the number of properties, significantly improving system response efficiency. Furthermore, by combining marginal rate of change analysis and the joint probability distribution matrix of property attributes, this application can accurately pinpoint the key bottleneck conditions or conflicting condition pairs that lead to an excessively narrow candidate property size, considering both single-factor influence and multi-factor conflict. This refined positioning capability allows for more targeted subsequent condition reconstruction, avoiding blindly relaxing conditions. This effectively expands the candidate property set while ensuring core user needs are met, thereby improving user experience and recommendation accuracy.

[0047] In some preferred embodiments, a specific example is given below. Suppose a user inputs the search requirement: "I want a house in the city center, three bedrooms, in a good school district, preferably fully furnished, not too high up, and with an elevator." The system first extracts the hard boundary conditions (e.g., "city center," "three bedrooms") and the preference boundary conditions (e.g., "in a good school district," "fully furnished," "not too high up," "with an elevator"). The system calls the cardinality estimation engine to estimate the number of properties that meet all these conditions under the current market supply. Suppose the estimation result shows that the number of properties is only 10, far below the lower limit of the preset reasonable range (e.g., 50 properties), the system determines that the candidate property size is too narrow. At this time, the system will initiate the process of locating the problematic condition. First, a marginal rate of change analysis is performed: 1. Temporarily remove the "in a good school district" condition, the estimated number of properties becomes 20, the marginal rate of change is low. 2. Temporarily remove the "fully furnished" condition, the estimated number of properties becomes 25, the marginal rate of change is low. 3. Temporarily remove the "not too high a floor" condition (e.g., below 5 floors), and the estimated number of available units becomes 30, with a low marginal rate of change. 4. Temporarily remove the "has an elevator" condition, and the estimated number of available units becomes 80, with a significantly increased marginal rate of change. Through the above analysis, the system identifies "has an elevator" as the key bottleneck condition leading to the excessively narrow supply. Simultaneously, the system queries the joint probability distribution matrix of property attributes. The matrix shows that in the "city center" area, the coexistence probability of the property attributes "not too high a floor" (e.g., below 5 floors) and "has an elevator" is lower than the preset probability threshold. This indicates that in the city center area, low-rise properties with elevators are very scarce, forming a pair of conflicting conditions. Ultimately, the system uses "has an elevator" and "not too high a floor" as conditions causing the candidate property size to exceed the preset reasonable range, and feeds this feedback to the subsequent layered, sequential reconstruction processing module for targeted adjustments. In this way, the system can accurately identify key points where user demand does not match market supply, thereby performing intelligent condition optimization.

[0048] In some embodiments described above, a hierarchical, sequential reconstruction process for combining search criteria is proposed to adjust the constraint level or value boundaries of some criteria, thereby forming reconstructed search criteria. However, in practical applications, how to systematically and effectively perform the reconstruction process based on different types of criteria and their characteristics, so as to expand the scale of candidate properties while preserving the user's core needs to the greatest extent, is an aspect that needs further refinement and optimization. Without a refined reconstruction strategy, the adjusted search criteria may deviate significantly from the user's original intent, or the reconstruction efficiency may be low. Therefore, this application further proposes a specific method for performing hierarchical, sequential reconstruction, which can adjust the criteria according to the attribute type of the criteria and the preset strategy order.

[0049] The above-mentioned hierarchical and successive refactoring process includes: Based on the attribute type corresponding to the condition item, the condition item is divided into continuous attribute conditions and discrete attribute conditions. The continuous attribute conditions are the conditions corresponding to property attributes whose values ​​change within a continuous numerical range, and the discrete attribute conditions are the conditions corresponding to property attributes whose values ​​are selected from a discrete option set. The conditions are adjusted according to a preset strategy sequence, which includes the following strategies: a range expansion strategy that extends the value boundaries of the continuous attribute conditions outward; a strategy that replaces the discrete attribute conditions with adjacent values ​​of similar adjacent values ​​using attribute similarity in the real estate knowledge graph; a strategy that backtracks the composite demand state graph to determine the life scenario entity that generates the adjusted condition, and replaces the current mapping path with the feasible mapping path corresponding to the life scenario entity in the knowledge graph; and a constraint level dimensionality reduction strategy that downgrades the preference boundary conditions to reference boundary conditions.

[0050] Specifically, when performing hierarchical restructuring, the first step is to categorize the conditions that cause the number of candidate properties to exceed a preset reasonable range. Continuous attribute conditions refer to properties whose values ​​vary within a continuous numerical range, such as house area, price range, and floor height. Discrete attribute conditions refer to properties whose values ​​are selected from a predefined set of discrete options, such as house orientation, decoration status, and apartment type. The purpose of distinguishing between these conditions is to enable the adoption of the most appropriate adjustment strategies for different types of attributes.

[0051] Furthermore, this application employs a method of adjusting the criteria according to a preset strategy order. This preset strategy order aims to gradually relax the search criteria while minimizing deviation from user intent, thereby expanding the candidate property set while maintaining the relevance of the user's original needs as much as possible. The specific strategy items included in the preset strategy order are: As a preferred implementation, a range expansion strategy can be used for continuous attribute conditions. This strategy increases the number of properties that meet the criteria by expanding the value boundaries of continuous attribute conditions outward, for example, by raising the upper limit of the price or lowering the lower limit, or by expanding the area range. The aim is to expand the search space by fine-tuning the numerical range while keeping the core attributes unchanged.

[0052] For discrete attribute conditions, the similarity of attributes in the real estate knowledge graph can be utilized by employing an adjacent value replacement strategy. This strategy involves replacing the current discrete attribute condition with an adjacent value in the knowledge graph that has a high degree of similarity to it. For example, if a user initially requests "fully furnished" but there are too few available properties, the system can broaden the condition to "basic furnishing" or "moderate furnishing" based on the knowledge graph, as these options have a certain similarity in terms of furnishing level. The goal is to find the closest alternative among the discrete options to reduce the perceived deviation by the user.

[0053] In addition, an alternative representation strategy can be employed. This strategy involves backtracking the composite demand state graph to identify the life scenario entity that generates the adjusted condition, and then replacing the current mapping path with other feasible mapping paths corresponding to that life scenario entity in the real estate knowledge graph. For example, if a user initially expresses the intention of "school district housing" and it is mapped to the standardized attribute "near a key primary school," when this condition results in a shortage of housing listings, the system can backtrack to the life scenario entity of "school district housing" and explore other standardized attributes that it may be associated with in the knowledge graph, such as "near high-quality educational resources" or "near a well-known middle school," thereby replacing the current search condition. The aim is to find broader but still relevant housing attribute representations starting from a more macroscopic life scenario intention.

[0054] As a further optimization, a constraint hierarchy reduction strategy can be employed. This strategy downgrades preference boundary conditions to reference boundary conditions. This means that conditions originally considered as strong user preferences become reference-only conditions after reconstruction, thus no longer serving as mandatory constraints during retrieval and significantly expanding the search scope. Its purpose is to significantly increase the number of candidate properties by reducing the enforceability of conditions when necessary.

[0055] Through the above technical solutions, this application provides a more refined, intelligent, and user-friendly retrieval condition reconstruction mechanism. First, by distinguishing between continuous and discrete attribute conditions, the system can adopt the most suitable adjustment strategy for different types of conditions, avoiding a "one-size-fits-all" approach and thus improving the efficiency and accuracy of reconstruction. Second, the application of a preset strategy order makes the reconstruction process hierarchical and logical, prioritizing strategies that minimize deviation from user intent and gradually relaxing conditions. This expands the candidate housing set while maximizing the retention of relevance to the user's original needs, significantly improving the user experience. For example, the range expansion strategy can smoothly adjust numerical conditions, while the adjacent value replacement and substitution representation strategies utilize the semantic relevance of the knowledge graph to ensure that adjustments to discrete conditions still align with the user's life scenario intent. Finally, the constraint hierarchy dimensionality reduction strategy provides the system with ultimate flexibility, ensuring that feasible housing can be found even in extreme cases, while avoiding unnecessary over-relaxation through hierarchical processing. This multi-strategy, layered restructuring process enables the system to exhibit stronger robustness and adaptability when dealing with complex and ever-changing user needs and housing supply status, effectively solving the problems of large deviations from intent and low efficiency that may occur in the restructuring process of traditional methods.

[0056] In some preferred embodiments, a specific example is given below. Suppose the user's search requirements are "city center, three bedrooms and two living rooms, fully furnished, budget 5-6 million, near a top-tier primary school". After initial searching, the system finds that the number of properties meeting all the conditions is extremely small, far below the lower limit of the preset reasonable range, and determines that the candidate property pool is too narrow. At this point, the system will perform a hierarchical restructuring process.

[0057] First, the system will identify conditions that lead to an overly narrow range of options. For example, "fully furnished" and "near a top-tier primary school" may result in a scarcity of available properties, while "budget of 5-6 million" may be too restrictive.

[0058] Next, the system will adjust these conditions according to the preset strategy order: Range expansion strategy: For continuous attribute conditions such as "budget 5 million-6 million", the system may expand its value boundary outward, for example, adjust it to "budget 4.8 million-6.2 million" to increase the number of properties within the price range.

[0059] Adjacent value replacement strategy: For the discrete attribute condition "fully furnished", the system will use the attribute similarity in the real estate knowledge graph to replace it with similar adjacent values, such as "medium-level decoration" or "basic decoration", because these decoration levels can meet the user's living needs to a certain extent.

[0060] Alternative Representation Strategy: For the condition "there is a top-tier primary school nearby," the system will trace back to the real-life scenario entity "school district housing" that generated this condition. By querying the real estate knowledge graph, the system finds that "school district housing" may map not only to "near a top-tier primary school," but also to "nearby quality educational resources" or "near a well-known middle school." In this case, the system may use the broader but still relevant representation "nearby quality educational resources" to replace the original condition.

[0061] Constraint level reduction strategy: If the number of housing units is still not ideal after the above adjustments, the system may downgrade the preference boundary condition of "three bedrooms and two living rooms" to a reference boundary condition. This means that in the final recommendation results, even if there are housing units with two bedrooms and one living room or four bedrooms and two living rooms, they will be included in the scope of consideration as long as other conditions are met, but they will be presented with a lower priority.

[0062] Through the above-mentioned hierarchical and sequential reconstruction process, the system can generate a reconstruction search condition. This condition expands the candidate housing set while preserving the user's original life intentions to the greatest extent, thereby providing the user with richer and more relevant housing options.

[0063] In some embodiments described above, a hierarchical, sequential reconstruction process is proposed, which adjusts some conditions in a combination of search criteria according to a preset strategy order. However, in practical applications, simply executing strategy items in a fixed preset order may not fully consider the differences in importance of different condition items to the user's intended needs, as well as the impact of the adjustment operation on the size and diversity of the candidate property set. This fixed-order adjustment method may lead to conditions that are more important to the user being relaxed prematurely in certain situations, or conditions that have less impact on the property set being adjusted late, thereby affecting the accuracy of the reconstruction results and user satisfaction.

[0064] In response, this application further proposes that when performing the above-mentioned hierarchical and sequential reconstruction process, a relaxation cost function is maintained for each condition item. The value of the relaxation cost function is jointly determined by the comprehensive constraint strength value of the corresponding condition item, the reciprocal of the information entropy gain of the candidate housing set after the relaxation operation, and the living logic deviation penalty term. When performing the hierarchical and sequential reconstruction process, the corresponding strategy item is selected to adjust the condition item according to the relaxation cost function in ascending order.

[0065] Specifically, the relaxation cost function is designed to quantify the "cost" or "negative impact" of relaxing a certain condition. This function comprehensively considers multiple factors: the overall constraint strength value reflects the user's emphasis on the condition or its core position in their needs. For example, if a user repeatedly emphasizes a condition, or if the condition originates from explicit discrete filtering, its overall constraint strength value will be higher, indicating that relaxing the condition will bring a greater risk of user intent deviation. The reciprocal of the information entropy gain of the candidate property set after the relaxation operation is used to measure the impact of the relaxation operation on the diversity of the candidate property set. The larger the information entropy gain, the greater the increase in the diversity of the candidate property set after relaxing the condition, i.e., the search space is effectively expanded. Its reciprocal means that if the relaxation operation can bring a greater expansion of the search space (i.e., a large information entropy gain), its contribution to the cost function will be relatively small, encouraging the system to prioritize adjustments that effectively expand the search scope. The lifestyle logic deviation penalty term is used to assess whether relaxing the condition will lead to a significant deviation from the logical connection with the user's original lifestyle intent. For example, if relaxing a certain condition would lead to recommended properties that are significantly different from the user's initial intention of finding a "school district property," then that penalty would be higher. In practical applications, by calculating the relaxation cost function value of each adjustable condition after applying different strategy items (such as range expansion strategy, adjacent value replacement strategy, alternative representation inheritance strategy, or constraint hierarchy dimensionality reduction strategy), the system can obtain a quantitative "cost" assessment.

[0066] Through the above technical solution, this application overcomes the limitations of traditional fixed-strategy sequential adjustments, making the hierarchical and sequential reconstruction process more refined and intelligent. By introducing a relaxed cost function, the system can comprehensively consider the emphasis of user intent, the effectiveness of search space expansion, and the preservation of life logic. Thus, among multiple feasible adjustment strategies, it selects the solution with the least impact on user experience and the most effective solution to the housing supply problem. This not only improves the accuracy of reconstructed search conditions and user satisfaction but also enables the system to exhibit stronger adaptability and robustness when facing complex and ever-changing user needs and housing supply conditions. It effectively avoids the decline in recommendation result quality caused by inappropriate relaxation, significantly improving the overall performance of second-hand housing information retrieval and recommendation.

[0067] In some preferred embodiments, a specific example is given below. Assume a user's initial search requirement is "city center, three bedrooms and two living rooms, with a good school district, budget under 5 million." After preliminary analysis, the system identifies "city center" and "three bedrooms and two living rooms" as hard boundary conditions, "with a good school district" as a preferred boundary condition (high overall constraint strength), and "budget under 5 million" as a preferred boundary condition (medium overall constraint strength). Assume that under the current housing supply, the number of properties meeting all these conditions is too narrow and needs restructuring. The system will evaluate the cost of relaxing the preferred boundary conditions "with a good school district" and "budget under 5 million." For the "with a good school district" condition, if a "substitute representation strategy" is adopted, relaxing it to "nearby high-quality educational resources" results in a high overall constraint strength. However, if the real estate knowledge graph shows that "nearby high-quality educational resources" and "with a good school district" are highly correlated in terms of life scenario utility, the life logic deviation penalty term may be lower. Simultaneously, relaxation may significantly increase the number of candidate properties, resulting in a smaller reciprocal of the information entropy gain. If a "constraint level reduction strategy" is adopted, downgrading it to a reference boundary condition, the overall constraint strength value is relatively high, the deviation penalty term from the life logic may be moderate, and the reciprocal of the information entropy gain may be relatively small. For the "budget within 5 million" condition, if a "range expansion strategy" is adopted, relaxing it to "budget within 5.5 million", the overall constraint strength value is moderate, and the deviation penalty term from the life logic may be relatively low. Relaxation may moderately increase the number of candidate properties, and the reciprocal of the information entropy gain will be moderate. The system will calculate the relaxation cost function value for each relaxation operation. For example, if the cost function value for relaxing "with school district" to "nearby high-quality educational resources" is 0.8, while the cost function value for relaxing "budget within 5 million" to "budget within 5.5 million" is 0.6, then the system will prioritize expanding the range of "budget within 5 million" to "budget within 5.5 million" because its relaxation cost is lower, meaning that while maintaining the user's core intent and life logic, it can more economically and effectively expand the set of candidate properties. This cost function-based decision-making mechanism enables the system to intelligently select the optimal adjustment path, rather than simply following a preset fixed order, thereby providing a reconstruction result that better meets user expectations.

[0068] In some embodiments described above, this application proposes a method for retrieving and recommending secondhand housing information. This method can perform hierarchical and sequential reconstruction of search conditions based on user-input search requirements and the current housing supply status to form reconstructed search conditions, thereby obtaining a suitable set of candidate housing. However, in practical applications, when the system reconstructs the user's initial input search requirements, the constraint level or value boundaries of some condition items may be adjusted. Users may therefore be unable to intuitively understand how their original requirements have been adjusted and the reasons behind these adjustments, which may lead to decreased user trust in the recommendation results or confusion. To address this, this application further proposes an optimization scheme aimed at improving system transparency by providing users with clear explanatory information to help them understand the changes in search conditions and their preservation of their original living intentions.

[0069] The above methods also include: Compare the differences between the user's initial search requirements and the reconstructed search conditions, and extract the changed condition items; By combining the preset real estate lifestyle knowledge graph and the current housing supply status, natural language explanation information is generated for the changed conditions. The natural language explanation information is used to explain the reasons for the condition adjustment and the preservation of the original lifestyle intention after the adjustment.

[0070] Specifically, after performing hierarchical and sequential restructuring and generating restructured search criteria, the system meticulously compares the user's initial search request with the final restructured search criteria used for retrieval. This comparison process aims to identify which specific condition items (e.g., price range, apartment type, region, etc.) have changed in terms of constraint level or value boundaries during the restructuring process. Once these changed condition items are identified, the system extracts them as key information that needs to be explained to the user.

[0071] Furthermore, to provide users with easily understandable and persuasive explanations, the system combines a pre-defined real estate lifestyle knowledge graph with the current housing supply status to generate natural language explanations for these changed conditions. The real estate lifestyle knowledge graph contains rich real estate-related concepts, attributes, and their interrelationships, providing semantic support; the current housing supply status reflects the real-time distribution and quantity of properties in the market. By comprehensively utilizing this information, the system can generate context-sensitive explanations, such as explaining that a certain price range has been relaxed because there are too few properties within that price range that meet other conditions, or that the conditions for a certain area have been adjusted to match the user's deeper lifestyle intentions.

[0072] The core purpose of the natural language explanation is to clarify the reasons for the adjustment of conditions and how well the adjustment preserves the user's original life intentions. For example, the explanation can explicitly state that a certain condition was adjusted because the original conditions were too strict, leading to a scarcity of housing, or because the system identified stronger preferences for other aspects of the user's life, thus balancing these factors while ensuring the core intention remains unchanged. Simultaneously, the explanation will emphasize how the adjusted conditions retain the user's original life intentions as much as possible. For instance, although the requirement for "school district housing" has been relaxed, the system will explain that the recommended housing is still located "near high-quality educational resources," thereby ensuring that the user understands the rationale behind the adjustment and its impact on their own interests.

[0073] Through the aforementioned technical solution, this application significantly improves the user experience and transparency of the second-hand housing information retrieval and recommendation system. Specifically, by providing users with natural language explanations of adjustments to search criteria, the system effectively avoids user confusion and dissatisfaction caused by a lack of understanding of these changes, thereby enhancing user trust in the recommendation results. Furthermore, this explanation mechanism allows users to better understand the system's working principles and decision-making logic, improving its explainability. Users can clearly see how their original living intentions are intelligently maintained and optimized by the system under housing supply constraints, thus increasing user satisfaction with the entire service process.

[0074] In response, this application further proposes to monitor user interaction behavior with the candidate housing set; when the interaction behavior represents a change in the user's attention to a certain preference boundary condition, calculate the intention deviation gradient, and use a preset smoothing algorithm to dynamically adjust the comprehensive constraint strength value of the preference boundary condition in the composite demand state diagram.

[0075] Specifically, monitoring user interactions with the candidate property list means the system continuously monitors user actions such as browsing, filtering, clicking, saving, ignoring, or rejecting recommended properties. These interactions implicitly reflect the user's preference for different property attributes or conditions. For example, frequent clicks on properties that meet specific preference criteria, or prolonged lingering on a property's details page, may be seen as signals of increased attention to those preferences.

[0076] Specifically, when an interaction represents a change in a user's attention to a certain preference boundary condition, the system calculates the intent bias gradient. The intent bias gradient can be understood as a quantitative indicator of the difference between the user's current actual preference and the system's existing understanding. This gradient value indicates whether the user's emphasis on a particular preference boundary condition is increasing or decreasing, and the magnitude of the change. For example, it can be calculated by analyzing the difference in the distribution of the set of properties clicked and the set of properties not clicked by the user regarding a specific preference boundary condition.

[0077] In practical applications, using a pre-defined smoothing algorithm to dynamically adjust the comprehensive constraint strength value of preference boundary conditions in the composite demand state diagram aims to ensure the stability and rationality of the adjustment process. Smoothing algorithms can avoid over-adjustment caused by single or few interactions, thus making the change in constraint strength more gradual and robust. For example, exponentially weighted moving average (EWMA) or other time series smoothing techniques can be used to gradually update the comprehensive constraint strength value of preference boundary conditions based on the intention deviation gradient and historical adjustment records.

[0078] Through the aforementioned technical solution, this application significantly improves the intelligence level and user satisfaction of second-hand housing information retrieval and recommendation. The system no longer relies solely on the static needs initially input by the user, but can dynamically capture and respond to subtle changes in user preferences by continuously learning user interaction behavior. This allows recommendation results to more accurately match the user's current and most genuine intent, avoiding recommendation bias caused by changes in user intent or inaccurate initial expression. Especially for preference boundary conditions, the constraint strength can be flexibly adjusted based on user feedback, thereby providing users with more attractive and personalized housing options while ensuring core needs are met, greatly optimizing the user experience.

[0079] As a specific implementation method, suppose the user's initial search requirements are "three-bedroom apartment, close to the subway, preferably in a good school district". Based on the initial requirements, the system identifies "three-bedroom apartment" as a hard boundary condition, "close to the subway" as a preferred boundary condition, and "in a good school district" as a preferred boundary condition, and assigns them corresponding comprehensive constraint strength values. Based on this, the system generates and displays a set of candidate properties.

[0080] While users browse these candidate property listings, the system monitors their interaction behavior. For example, users may frequently click on properties that, while not strictly in a good school district, are very close to the subway and have a regular floor plan, while they may click less on or ignore properties that are in good school districts but are far from the subway.

[0081] At this point, the system will identify that the user's attention to the preference boundary condition "near the subway" may increase, while their attention to the preference boundary condition "with school district housing" may decrease. The system will calculate the corresponding intent bias gradient; for example, the gradient for "near the subway" is positive, while the gradient for "with school district housing" is negative. Subsequently, the system uses a preset smoothing algorithm, such as an exponentially decaying weighted average model, to gradually increase the comprehensive constraint strength value of "near the subway" in the composite demand state diagram, and gradually decrease the comprehensive constraint strength value of "with school district housing".

[0082] Through this dynamic adjustment, the system's understanding of users' true preferences becomes more accurate. In subsequent search criterion reconstruction and property recommendation processes, the system will prioritize properties that better match users' current actual preferences (e.g., emphasizing subway commuting convenience rather than strict school district attributes), thereby providing recommendations that are more tailored to users' needs.

[0083] In some embodiments described above, this application proposes to monitor user interactions with a set of candidate properties and calculate the intent bias gradient based on changes in user attention to preference boundary conditions as represented by these interactions. A pre-defined smoothing algorithm is then used to dynamically adjust the overall constraint strength of the preference boundary conditions in the composite demand state diagram. However, in practical applications, user preferences for properties are often complex and multidimensional, with different property attributes potentially substituting for each other in terms of utility within a specific living scenario. Directly lowering the constraint strength based solely on the unmet need for a single preference condition might overlook unexpected satisfaction gained by the user across other preference conditions, leading to misjudgment of the user's true intent. It could even over-relax certain key preferences, affecting the accuracy of the recommendation results.

[0084] In response, this application further proposes to dynamically adjust the comprehensive constraint strength value of the aforementioned preference boundary conditions in the composite demand state diagram using a preset smoothing algorithm, specifically including: When a user clicks on any property in the candidate property set, the clicked property is taken as the target property. The negative deviation value of the target property on the unsatisfied first preference boundary condition and the positive overflow value of the target property on the satisfied second preference boundary condition are extracted. When it is determined that the positive overflow value exceeds the preset compensation threshold, and there is a preset value exchange relationship between the first preference boundary condition and the second preference boundary condition that represents the substitutability of different housing attributes in terms of living scenario utility, the operation of reducing the comprehensive constraint strength of the first preference boundary condition is intercepted, and a local exchange rule is generated. The local exchange rule records the value level to which the first preference boundary condition is allowed to be relaxed and the compensation value level to which the second preference boundary condition must be maintained. The local exchange rule is used to be preferentially applied when the hierarchical acceptance reconstruction process is subsequently executed.

[0085] Specifically, when a user clicks on a property in the candidate property set, that property is considered the target property. The system analyzes the match between the target property and the user's initial search requirements and their preferred boundary conditions. The negative bias value refers to the degree to which the target property does not meet a user's first unmet preference boundary condition. For example, if a user expects a "large balcony," but the target property has a small balcony, there is a negative bias. The positive overflow value refers to the degree to which the target property exceeds the user's expected satisfaction of a user's second met preference boundary condition. For example, if a user expects "convenient transportation," but the target property is extremely close to a subway station, far exceeding general convenience standards, there is a positive overflow. The preset compensation threshold is an empirical or calculated value used to determine whether the positive overflow is significant enough to offset the negative bias. The preset value exchange relationship refers to the substitutability of different property attributes in the user's life scenario, predefined in the real estate life knowledge graph. For example, for some users, the value of "short commute time" can partially substitute for the value of "large house area." When these conditions are simultaneously met, the system will intercept any operation that lowers the overall constraint strength of the first preference boundary condition to avoid overly relaxing user preferences due to a single dimension's non-compliance. Simultaneously, the system will generate a local exchange rule, which details the specific value level to which the first preference boundary condition is allowed to be relaxed in the current context, and the compensation value level that the second preference boundary condition needs to maintain to achieve this relaxation. This local exchange rule will be prioritized and applied during subsequent hierarchical continuation-based refactoring to guide more intelligent condition adjustments.

[0086] Through the aforementioned technical solution, this application can more precisely understand the complex trade-offs users make when choosing housing, avoiding the system mistakenly lowering the constraint strength of unmet preferences when a user actually accepts a certain "compensatory" housing offer. This makes the dynamic adjustment of preference boundary conditions more intelligent and human-centered, effectively preventing excessive relaxation of conditions due to a single dimension's dissatisfaction. Therefore, even when housing supply is tight or user needs are ambiguous, it can still generate reconstructed search conditions that both align with the user's core intent and have high matchability. Furthermore, the generation and priority application of local exchange rules provide more guiding principles for subsequent hierarchical reconstruction processing, further improving the accuracy of search and recommendation results and the user experience.

[0087] In some embodiments, reference is made to Figure 3This application proposes a second-hand housing information retrieval and recommendation system, comprising: an acquisition unit 301, used to acquire user-inputted search requirements, extract the textual expressions of lifelike intents and discretized filtering conditions contained in the search requirements, and construct a composite demand state graph, wherein the composite demand state graph uses nodes to represent demand intents and edges to represent the logical relationships between demand intents; an evaluation unit 302, used to quantitatively evaluate the constraint strength of the demand conditions corresponding to each node in the composite demand state graph, and divide the demand conditions into hard boundary conditions, preference boundary conditions, and reference boundary conditions according to the evaluation results; and a processing unit 303, used to... Based on the current housing supply status, the feasibility of the search condition combination formed by the combination of hard boundary conditions and preference boundary conditions is tested. When the test result indicates that the candidate housing scale corresponding to the search condition combination exceeds the preset reasonable range, the condition item that causes the candidate housing scale to exceed the preset reasonable range is located. The adjustment unit 304 is used to perform hierarchical inheritance-style reconstruction processing according to the located condition item and its constraint strength, adjust the constraint level or value boundary of some conditions in the search condition combination, and form reconstructed search conditions. The reconstructed search conditions are used to initiate a search in the housing database to obtain a set of candidate housings and to generate recommendation results.

[0088] This application's secondhand housing information retrieval and recommendation system, through the collaborative work of its acquisition, evaluation, processing, and adjustment units, aims to address the disconnect between users' everyday intentions and standardized screening criteria in existing secondhand housing information retrieval and recommendation methods, as well as the instability of search results due to improper combination of criteria. Through the organic integration of these technical units, the system of this application can more accurately understand user needs and dynamically adjust search criteria based on market supply, thereby providing housing recommendations that better meet user expectations.

[0089] The above are merely embodiments of this application and are not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for retrieving and recommending secondhand housing information, characterized in that, include: The system obtains the user's input search requirements, extracts the textual expressions of lifelike intent and discretized filtering conditions contained in the search requirements, and constructs a composite demand state graph. The composite demand state graph uses nodes to represent demand intents and edges to represent the logical relationships between demand intents. The constraint strength of the demand conditions corresponding to each node in the composite demand state diagram is quantitatively evaluated, and the demand conditions are divided into hard boundary conditions, preferred boundary conditions and reference boundary conditions based on the evaluation results. Based on the current housing supply status, the feasibility of the search condition combination formed by the combination of the hard boundary condition and the preference boundary condition is tested, and when the test result indicates that the candidate housing scale corresponding to the search condition combination exceeds the preset reasonable range, the condition item that causes the candidate housing scale to exceed the preset reasonable range is located. Based on the located condition items and their constraint strength, a hierarchical succession-style reconstruction process is performed to adjust the constraint level or value boundary of some conditions in the combination of search conditions to form reconstructed search conditions. The reconstructed search conditions are used to initiate a search in the housing database to obtain a set of candidate housing properties and to generate recommendation results.

2. The method according to claim 1, characterized in that, The construction of the composite demand state diagram includes: A preset real estate lifestyle knowledge graph is invoked to identify the lifestyle scene entities contained in the lifestyle intent expression, and the lifestyle scene entities are mapped to multiple standardized housing attribute entities associated in the knowledge graph to obtain the mapping path between the lifestyle scene entities and the standardized housing attribute entities and the association weight of each mapping path. The mapping path and its association weight are used as the basis for determining the edge weight between corresponding nodes in the composite demand state graph. The discretized filtering conditions extracted from the search requirements are used as explicit standardized property attribute entities to provide data sources and judgment criteria for the calculation of structured confirmation degree; The standardized housing attribute entities and the living scene entities are used as nodes in the composite demand state graph. The logical relationships between the entities determined by semantic analysis of the expression of the living intention are used as edges in the composite demand state graph. When the standardized housing attribute entities are mapped from the living scene entities, the weight of the corresponding edge is determined according to the association weight of the mapping path to construct the composite demand state graph.

3. The method according to claim 1, characterized in that, The quantitative evaluation of the constraint strength of the demand conditions corresponding to each node in the composite demand state diagram, and the division of the demand conditions into hard boundary conditions, preferred boundary conditions, and reference boundary conditions based on the evaluation results, includes: Based on the semantic emphasis, structured confirmation, and life logic relevance of the corresponding demand conditions of each node, the comprehensive constraint strength value of each node is determined by weighted calculation. The semantic emphasis is determined based on the strength of the wording or the frequency of modification when the user expresses the demand. The structured confirmation is determined based on whether the demand originates from the discrete screening conditions. The life logic relevance is determined based on whether multiple different life scenario entities in the composite demand state diagram point to the same standardized housing attribute entity. Demand conditions where the comprehensive constraint strength value is higher than the first strength threshold are classified as hard boundary conditions, demand conditions where the comprehensive constraint strength value is between the first strength threshold and the second strength threshold are classified as preferred boundary conditions, and demand conditions where the comprehensive constraint strength value is lower than the second strength threshold are classified as reference boundary conditions.

4. The method according to claim 1, characterized in that, The step involves performing an executability test on the search condition combination formed by the combination of hard boundary conditions and preference boundary conditions based on the current housing supply status. When the test result indicates that the candidate housing size corresponding to the search condition combination exceeds a preset reasonable range, the step identifies the condition item that causes the candidate housing size to exceed the preset reasonable range, including: The preset cardinality estimation engine is invoked to estimate the number of properties that can be matched by the combination of search conditions under the current property supply status; Determine whether the estimated number of housing units falls within a preset reasonable range. If the estimated number of housing units is lower than the lower limit of the preset reasonable range, it is determined that the candidate housing unit scale is too narrow. If the estimated number of housing units is higher than the upper limit of the preset reasonable range, it is determined that the candidate housing unit scale is too wide. When the candidate housing stock size is too narrow, the key bottleneck conditions causing the narrowing are located by temporarily removing the preference boundary conditions in the search condition combination and observing the marginal change rate of the estimated housing stock number after removal. By querying the joint probability distribution matrix of housing stock attributes, conflict condition pairs with a coexistence probability lower than a preset probability threshold under the current supply state are identified. The key bottleneck conditions or the condition items in the conflict condition pairs are used as the condition items that cause the candidate housing stock size to exceed the preset reasonable range.

5. The method according to claim 1, characterized in that, The hierarchical, successive refactoring process includes: Based on the attribute type corresponding to the condition item, the condition item is divided into continuous attribute conditions and discrete attribute conditions. The continuous attribute conditions are the conditions corresponding to property attributes whose values ​​change within a continuous numerical range, and the discrete attribute conditions are the conditions corresponding to property attributes whose values ​​are selected from a discrete option set. The conditions are adjusted according to a preset strategy sequence, which includes the following strategies: a range expansion strategy that extends the value boundaries of the continuous attribute conditions outward; a strategy that replaces the discrete attribute conditions with adjacent values ​​of similar adjacent values ​​using attribute similarity in the real estate knowledge graph; a strategy that backtracks the composite demand state graph to determine the life scenario entity that generates the adjusted condition, and replaces the current mapping path with the feasible mapping path corresponding to the life scenario entity in the knowledge graph; and a constraint hierarchy reduction strategy that downgrades the preference boundary conditions to reference boundary conditions.

6. The method according to claim 5, characterized in that, The method further includes: A relaxation cost function is maintained for each condition item. The value of the relaxation cost function is determined by the comprehensive constraint strength value of the corresponding condition item, the reciprocal of the information entropy gain of the candidate housing set after the relaxation operation, and the living logic deviation penalty term. When performing the hierarchical succession-based reconstruction process, the corresponding strategy terms are selected to adjust the condition terms according to the order of the relaxed cost function from smallest to largest.

7. The method according to claim 1, characterized in that, The method further includes: Compare the differences between the user's initial search requirements and the reconstructed search conditions, and extract the changed condition items; By combining a pre-defined real estate lifestyle knowledge graph with the current housing supply status, natural language explanation information is generated for the changed conditions. The natural language explanation information is used to explain the reasons for the condition adjustment and the preservation of the original lifestyle intention after the adjustment.

8. The method according to claim 1, characterized in that, The method further includes: Monitor user interactions with the candidate property list set; When the interaction behavior represents a change in the user's attention to a certain preference boundary condition, the intention deviation gradient is calculated, and the comprehensive constraint strength value of the preference boundary condition in the composite demand state diagram is dynamically adjusted using a preset smoothing algorithm.

9. The method according to claim 8, characterized in that, The method of dynamically adjusting the comprehensive constraint strength value of the preference boundary conditions in the composite demand state diagram using a preset smoothing algorithm includes: When a user clicks on any property in the candidate property set, the clicked property is taken as the target property. The negative deviation value of the target property on the unsatisfied first preference boundary condition and the positive overflow value of the target property on the satisfied second preference boundary condition are extracted. When it is determined that the positive overflow value exceeds the preset compensation threshold, and there is a preset value exchange relationship between the first preference boundary condition and the second preference boundary condition that represents the substitutability of different housing attributes in terms of living scenario utility, the operation of reducing the comprehensive constraint strength of the first preference boundary condition is intercepted, and a local exchange rule is generated. The local exchange rule records the value level to which the first preference boundary condition is allowed to be relaxed and the compensation value level to which the second preference boundary condition must be maintained. The local exchange rule is used to be preferentially applied when the hierarchical acceptance reconstruction process is subsequently executed.

10. A second-hand housing information retrieval and recommendation system, characterized in that, include: The acquisition unit is used to acquire the user's input search requirements, extract the textual expressions of lifelike intent and discretized filtering conditions contained in the search requirements, and construct a composite demand state graph. The composite demand state graph uses nodes to represent demand intents and edges to represent the logical relationships between demand intents. The evaluation unit is used to quantitatively evaluate the constraint strength of the demand conditions corresponding to each node in the composite demand state diagram, and to divide the demand conditions into hard boundary conditions, preferred boundary conditions and reference boundary conditions based on the evaluation results. The processing unit is used to perform an executability test on the search condition combination formed by the combination of the hard boundary condition and the preference boundary condition according to the current housing supply status, and when the test result indicates that the candidate housing scale corresponding to the search condition combination exceeds the preset reasonable range, locate the condition item that causes the candidate housing scale to exceed the preset reasonable range. The adjustment unit is used to perform hierarchical reconstruction processing based on the located condition items and their constraint strength, adjust the constraint level or value boundary of some conditions in the combination of search conditions, and form reconstructed search conditions. The reconstructed search conditions are used to initiate a search in the housing database to obtain a set of candidate housings and to generate recommendation results.

Images