A semantic retrieval method for homestay listings based on knowledge graph reasoning

Abstract

This invention discloses a semantic retrieval method for homestay listings based on knowledge graph reasoning, belonging to the field of listing retrieval technology. The method includes: constructing a semantic fulfillment graph set for listings; receiving user search queries, extracting search constraint information and target experience semantic items, generating a target check-in scenario graph and a target check-in scenario neighborhood; limiting candidate listings based on search constraint information, mapping the target check-in scenario graph to the listing semantic fulfillment graph, calculating and determining the check-in scenario semantic fulfillment potential, and forming a candidate listing semantic fulfillment status table; calculating the counter-evidence constraint based on the counter-evidence path and performing flow splitting; while keeping the core experience semantic items unchanged, performing neighborhood fill-in reasoning on non-core experience semantic items to form a candidate listing counter-evidence flow splitting table; and outputting semantic retrieval data and semantic fulfillment descriptions based on the candidate listing counter-evidence flow splitting table. This invention improves retrieval accuracy and reduces false recall.

Description

Technical Field

[0001] This invention relates to the field of property listing retrieval technology, specifically a semantic retrieval method for homestay listings based on knowledge graph reasoning. Background Technology

[0002] With the development of homestay and short-term rental platforms, property retrieval has gradually shifted from keyword matching to semantic search. Conventional methods typically construct a search index based on property titles, facility tags, price ranges, geographical location, user profiles, and historical behavior data. They also incorporate knowledge graphs to describe the relationships between properties, scenic spots, transportation hubs, surrounding amenities, and check-in rules, allowing users to express their accommodation needs in natural language and obtain a list of candidate properties. This type of method can structure the basic attributes of properties and user search criteria, and supports multi-condition queries, similar semantic recall, and ranked property display.

[0003] However, conventional semantic retrieval methods typically focus more on the matching relationship between user needs and property tags and text descriptions, lacking further judgment on whether the semantics of the property experience can be reliably delivered under the target occupancy context. This can easily lead to properties with strong promotional descriptions but insufficient historical experience support being placed in the top positions. At the same time, conventional methods are weak in utilizing negative experience evidence such as negative reviews, canceled records, and abandoned inquiries, making it difficult to perform stratification, suppression, and replacement of properties with the risk of experience distortion during the retrieval stage. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a semantic retrieval method for homestay listings based on knowledge graph reasoning, which solves the problems of existing technologies, such as difficulty in determining whether the semantics of the listing experience are stably realized in the target occupancy context and difficulty in using reverse experience evidence in listing diversion processing.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a semantic retrieval method for homestay listings based on knowledge graph reasoning, comprising: acquiring a multi-source semantic dataset of homestay listings, extracting semantic commitment items, scene triggering factors, and experience fulfillment evidence, and constructing a semantic fulfillment graph set for listings; receiving user search statements, extracting search constraint information and target experience semantic items, and generating a target check-in scenario graph and a target check-in scenario neighborhood; limiting candidate listings based on search constraint information, mapping the target check-in scenario graph to the corresponding listing semantic fulfillment graph, calculating the check-in scenario semantic fulfillment potential of each target experience semantic item, and performing check-in scenario semantic fulfillment judgment to form a candidate listing semantic fulfillment status table; calculating the counter-evidence constraint quantity of each target experience semantic item based on the counter-evidence fragments in the experience fulfillment evidence, and performing counter-evidence constraint diversion processing; while keeping the core experience semantic items unchanged, performing neighborhood supplement reasoning on non-core experience semantic items to form a candidate listing counter-evidence diversion table; sorting the final candidate listings according to the candidate listing counter-evidence diversion table, and outputting homestay listing semantic retrieval data and corresponding semantic fulfillment descriptions.

[0007] As a preferred embodiment of the knowledge graph-based semantic retrieval method for homestay listings described in this invention, the construction of the listing semantic fulfillment graph set includes: performing time unification, spatial normalization, and text processing on homestay platform data to obtain a multi-source semantic dataset of listings; extracting listing semantic commitment items from the multi-source semantic dataset of listings, and generating scene triggering factors corresponding to the listing semantic commitment items based on check-in context fragments; extracting experience fulfillment evidence corresponding to the listing semantic commitment items from the multi-source semantic dataset of listings; and establishing corresponding relationship edges with the unique listing identifier as the central node, and the listing semantic commitment items, scene triggering factors, and experience fulfillment evidence as associated nodes to construct the listing semantic fulfillment graph set.

[0008] As a preferred embodiment of the homestay listing semantic retrieval method based on knowledge graph reasoning described in this invention, the generation of the target occupancy context graph and the target occupancy context neighborhood includes: performing semantic parsing on the user's search statement, extracting search constraint information, target experience semantic items, and extended experience semantic items, determining core experience semantic items and non-core experience semantic items, and constructing the target occupancy context graph; based on the target occupancy context graph, selecting historical occupancy context fragments that match the target occupancy context from the listing semantic fulfillment graph set to form the target occupancy context neighborhood.

[0009] As a preferred embodiment of the homestay listing semantic retrieval method based on knowledge graph reasoning described in this invention, the step of mapping the target check-in scenario graph to the corresponding listing semantic fulfillment graph includes: determining candidate limiting conditions based on the retrieval constraint information in the target check-in scenario graph; selecting listing nodes that meet the basic retrieval conditions from the listing semantic fulfillment graph set under the candidate limiting conditions to form an initial candidate listing set; performing rule consistency verification on the initial candidate listing set to obtain the candidate listing set; retrieving listing semantic commitment item nodes with the same name or corresponding to the same normative semantic item in the listing semantic fulfillment graph corresponding to the candidate listings to establish a semantic mapping relationship; and recording the scene triggering factor nodes in the corresponding listing semantic fulfillment graph that correspond to both the semantic mapping relationship and the target check-in scenario neighborhood as valid mapping nodes based on the semantic mapping relationship.

[0010] As a preferred embodiment of the knowledge graph-based semantic retrieval method for homestay listings described in this invention, the calculation of the occupancy context semantic fulfillment potential of each target experience semantic item includes: for each candidate listing and each target experience semantic item, calculating the occupancy context semantic fulfillment potential based on the listing semantic commitment anchor value, the fulfillment verification density within the target occupancy context neighborhood, and the scene drift dispersion; the scene drift dispersion is obtained by statistically analyzing the number of experience fulfillment state switching times between adjacent effective neighborhood segments within the target occupancy context neighborhood of the listing's target experience semantic item, and calculating the ratio of the number of experience fulfillment state switching times to the total number of effective adjacent neighborhood segments.

[0011] As a preferred embodiment of the semantic retrieval method for homestay listings based on knowledge graph reasoning as described in this invention, the determination of the semantic fulfillment of the check-in context includes: when the semantic fulfillment potential of the check-in context is not lower than the steady-state threshold, the candidate listing for the corresponding target experience semantic item is recorded as a stable fulfillment state; when the semantic fulfillment potential of the check-in context is higher than the non-fulfillment boundary but lower than the steady-state threshold, the candidate listing for the corresponding target experience semantic item is recorded as an oscillating fulfillment state; when the semantic fulfillment potential of the check-in context is not higher than the non-fulfillment boundary, the candidate listing for the corresponding target experience semantic item is recorded as a non-fulfillment risk state.

[0012] As a preferred embodiment of the knowledge graph-based semantic retrieval method for homestay listings according to the present invention, the calculation of the counter-evidence constraint of each target experience semantic item includes: taking each target experience semantic item in the candidate listing semantic fulfillment status table as a counter-evidence retrieval object, retrieving counter-evidence paths in the counter-evidence path table that correspond to the same unique listing identifier, the same target experience semantic item, and the same valid mapping node, forming a set of counter-evidence paths to be judged; performing listing consistency verification, semantic consistency verification, and contextual consistency verification on the set of counter-evidence paths to be judged, and recording the counter-evidence paths that pass the verification as valid counter-evidence paths; calculating the counter-evidence path recurrence density based on the ratio between the number of valid counter-evidence paths and the total number of experience fulfillment evidence that can be used to verify the same target experience semantic item in the neighborhood of the target check-in context; calculating the neighborhood fit based on the number of standardized state markers consistent between the check-in context fragment corresponding to the valid counter-evidence path and the neighborhood of the target check-in context; and calculating the counter-evidence constraint based on the counter-evidence path recurrence density, the neighborhood fit, and the check-in context semantic fulfillment potential.

[0013] As a preferred embodiment of the knowledge graph-based semantic retrieval method for homestay listings described in this invention, the step of performing counter-evidence constraint and diversion processing includes: when the counter-evidence constraint amount is lower than the counter-evidence constraint threshold, the corresponding target experience semantic item is recorded as a weak constraint state; when the counter-evidence constraint amount is not lower than the counter-evidence constraint threshold and is lower than the counter-evidence cutoff boundary, and the corresponding target experience semantic item is in a stable fulfillment state, the corresponding target experience semantic item is still recorded as a weak constraint state; when the counter-evidence constraint amount is not lower than the counter-evidence constraint threshold and is lower than the counter-evidence cutoff boundary, and the corresponding target experience semantic item is in a fluctuating fulfillment state, the corresponding target experience semantic item is recorded as a restricted state; when the counter-evidence constraint amount is not lower than the counter-evidence cutoff boundary, or the target experience semantic item is in a state of loss of fulfillment risk and the counter-evidence constraint amount is not lower than the counter-evidence constraint threshold, the target experience semantic item is recorded as a cutoff state.

[0014] As a preferred embodiment of the knowledge graph-based semantic retrieval method for homestay listings described in this invention, the neighbor-supplement reasoning for non-core experience semantic items includes: if all core experience semantic items of a candidate listing are in a stable fulfillment state and none of the core experience semantic items have entered a cutoff state, then the candidate listing is recorded as a direct output class; if no core experience semantic item of a candidate listing has entered a cutoff state, but at least one core experience semantic item is in a swing fulfillment state or a restricted state, then the candidate listing is recorded as a restricted retention class; if a core experience semantic item of a candidate listing has entered a cutoff state, then the candidate listing is recorded as a supplementary replacement class, and recorded as... The core experience semantic items that trigger the supplementation are recorded; when the number of output properties in the direct output class and the limited retention class is lower than the return scale, the neighborhood supplementation reasoning is initiated; for non-core experience semantic items in the supplementation candidate properties, adjacent normative semantic items under the same accommodation experience category are retrieved in the homestay semantic mapping table, and the original non-core experience semantic items are replaced for supplementation mapping; the neighborhood supplementation reasoning takes the core experience semantic items that trigger the supplementation as the maintenance object, does not change the set of core experience semantic items, and selects candidate properties from the candidate property semantic realization status table whose core experience semantic items are in a stable realization state and whose counter-evidence constraint is lower than the counter-evidence constraint threshold as supplementation candidate properties.

[0015] As a preferred embodiment of the knowledge graph-based semantic retrieval method for homestay listings described in this invention, the output of homestay listing semantic retrieval data and corresponding semantic fulfillment descriptions includes: merging directly output listings, narrowed-down retained listings, and supplementary listings that have entered the output range through neighborhood supplementation reasoning to form a final candidate listing set; after deduplication of the final candidate listing set, performing semantic fulfillment hierarchical sorting, generating semantic fulfillment descriptions corresponding to each output listing, and outputting the homestay listing semantic retrieval data; the semantic fulfillment hierarchical sorting includes listing distribution hierarchical sorting, core... The ranking system comprises three stages: a performance-based ranking, a counter-evidence-based ranking, and an evidence-based ranking. The housing allocation ranking follows the order of direct output, limited retention, and replaceable housing. The core performance-based ranking is based primarily on the performance potential of the occupancy context corresponding to the core experience semantic item. The counter-evidence-based ranking is based on the counter-evidence-based quantity and counter-evidence-based allocation status corresponding to the core experience semantic item. The evidence-based ranking is based on the evidence duration interval, the number of consecutive evidence segments, and the time of the most recent evidence occurrence corresponding to the core experience semantic item.

[0016] The beneficial effects of this invention are as follows: by constructing a semantic fulfillment graph set of housing listings, unified reasoning of housing listing descriptions, occupancy contexts, and historical experience evidence is achieved, thereby improving retrieval accuracy; by calculating the semantic fulfillment potential of occupancy contexts, the stable fulfillment capability of housing experience semantics is identified, reducing false recall rate; and by using counter-evidence to control traffic and neighboring supplementary reasoning, distorted housing listings are suppressed and substitute housing listings are added, thereby improving retrieval reliability. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a semantic retrieval method for homestay listings based on knowledge graph reasoning.

[0019] Figure 2 A schematic diagram for constructing a semantic fulfillment atlas for housing listings.

[0020] Figure 3 A schematic diagram illustrating the mapping of target check-in scenarios and semantic fulfillment determination.

[0021] Figure 4 A schematic diagram illustrating the hierarchical sorting of traffic diversion and semantic fulfillment for the purpose of counter-evidence. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides a semantic retrieval method for homestay listings based on knowledge graph reasoning, including the following steps: S1. Obtain a multi-source semantic dataset of homestay listings, extract semantic commitment items, scene triggering factors, and experience fulfillment evidence, and construct a semantic fulfillment graph set of homestay listings.

[0026] Furthermore, it integrates data from the homestay platform, including property registration data, property details text, check-in rules text, facility tag data, price calendar data, inventory time series data, order flow data, consultation dialogue text, cancellation record text, review text, and external spatiotemporal environment data.

[0027] The external spatiotemporal environment data includes regional weather data, regional passenger flow data, surrounding facilities data, and transportation accessibility data.

[0028] The data from the homestay platform is processed in a unified time, spatial, and textual manner to obtain a multi-source semantic dataset of listings.

[0029] Furthermore, using the unique identifier of the property as the primary key, the time of execution is uniformly processed for property registration data, price calendar data, inventory time series data, order flow data, consultation dialogue text, cancellation record text, and evaluation text. Spatial normalization is performed on property address, supporting facility name, and transportation station name. Noise reduction, proper noun standardization, and privacy anonymization are performed on detail text, rule text, consultation dialogue text, cancellation record text, and evaluation text to obtain a multi-source semantic dataset of properties.

[0030] Furthermore, semantic fragments directly related to the accommodation experience are extracted from the multi-source semantic dataset of housing listings, and a semantic mapping table for homestays is established.

[0031] The semantic mapping table for homestays is organized according to semantic phrases, standardized semantic items, and semantic categories.

[0032] The semantic categories include at least the quiet category, the parent-child category, the cooking category, the office category, the landscape category, the transportation category, the long-stay category, and the check-in convenience category.

[0033] Semantic matching is performed on each of the property details text, check-in rules text, facility tag data, consultation dialogue text, and review text, mapping semantic phrases with the same meaning but different textual forms into unified and standardized semantic items.

[0034] For example, "can be lit", "allow cooking" and "have kitchen utensils" are uniformly mapped to the semantic item "can cook", and "suitable for bringing children", "family-friendly" and "convenient for children to stay" are uniformly mapped to the semantic item "family-adaptive".

[0035] After mapping is completed, the location, time of occurrence, and consecutive occurrence interval of the same standardized semantic item in different data sources are aggregated according to the unique identifier of the property to obtain a set of semantic candidate fragments.

[0036] To filter out housing listing semantic commitment terms with stable expressive meaning from the set of semantic candidate fragments, a housing listing semantic commitment anchor value is constructed, expressed as: ; in, Indicates housing resources For canonical semantic items The semantic promise anchor value of the property listing. Representing canonical semantic terms In housing The explicit density normalization value in the details text, check-in rules text, facility label data, consultation dialogue text, and review text. Representing canonical semantic items The semantic consistency coefficient among multi-source data. Representing canonical semantic terms Semantic drift suppression on consecutive time slices.

[0037] It should be noted that, The effective occurrence record is obtained by statistically analyzing the number of valid occurrence records of the standard semantic item in the multi-source semantic data of the current property listing, and then calculating the ratio of the number of valid occurrence records to the total number of text fragments that can carry the standard semantic item. The valid occurrence record refers to the text fragment containing the semantic phrase that is determined to have, allow, or support the occurrence of the corresponding standard semantic item in the current property listing after passing negative word filtering, subject consistency verification, and semantic direction verification. For example, "the room is very quiet" and "resting at night is undisturbed" can be considered as a valid occurrence of the quiet semantic item, while "not quiet", "too noisy at night", and "the landlord said there may be construction nearby" cannot be considered as a valid occurrence of the quiet semantic item. By separately analyzing the text of property details, check-in rules, facility tags, consultation dialogues, and reviews, the semantic items of the standardization were analyzed. The number of forward-supporting data sources, the number of reverse-pointing data sources, and the number of data sources not explicitly pointed to are calculated as the proportion of the number of forward-supporting data sources minus the number of reverse-pointing data sources in the total number of data sources, with a value range of [0,1]. By analyzing housing resources Chinese canonical semantic terms The continuous time slices are compared with adjacent segments. When the change rate of the number of valid occurrence records in adjacent time slices exceeds the semantic stability change boundary or the semantic direction is switched, the corresponding adjacent time slice is recorded as a drift segment. The drift segment is obtained by the ratio of the number of drift segments to the number of valid adjacent time slices, and the value range is [0,1]. The semantic stability change boundary is determined by statistically analyzing the change rate of the number of valid occurrence records of the same standard semantic item in the historical completed check-in orders within the continuous time slice, and selecting the critical change rate that can distinguish between normal text update fluctuations and semantic expression drift. The preferred value range is [0.15,0.35].

[0038] For each property listing and each specification semantic item, calculate the property listing semantic commitment anchor value. When the property listing semantic commitment anchor value is not lower than the commitment to be included in the map, write the specification semantic item into the property listing semantic commitment item set.

[0039] It should be noted that the commitment to be included in the map is determined by statistically analyzing the distribution of the semantic commitment anchor values ​​of the housing listings corresponding to the effective experience semantics and the noise semantics in the historical completed order housing listing samples, and selecting a critical value that can stably distinguish between effective experience semantics and noise semantics, with the preferred value range being [0.45, 0.70].

[0040] Furthermore, using check-in date, check-out date, number of guests, travel type, weather conditions, passenger flow, day / night time, holiday conditions, and traffic congestion conditions as contextual description dimensions, we segmented order flow data, price calendar data, inventory time series data, weather data, passenger flow data, surrounding facilities data, and transportation accessibility data into multiple sets of check-in contextual segments.

[0041] Specifically, firstly, order segments are extracted according to the same property identifier and the same order occupancy range. Then, the price calendar data and inventory time series data within the order segments are aligned by natural days, and the weather data, passenger flow data, and transportation accessibility data are assigned to the corresponding natural days according to the collection time. The slice boundaries are determined based on whether there are changes in day and night time, holiday status, weather status, passenger flow status, and traffic congestion status. When any scenario description dimension changes state, the current segment ends and the next segment begins. For each segment after segmentation, the property identifier, occupancy range, standardized state markers of the scenario description dimension, segment start and end time, and corresponding order flow node are recorded to form multiple sets of occupancy scenario segments.

[0042] For example, weather conditions belong to the context description dimension, rainy days belong to the standardized state markers corresponding to weather conditions, and "quiet semantic items, nighttime, high passenger flow, traffic congestion" together form the scene trigger factor nodes corresponding to quiet semantic items.

[0043] For each check-in scenario segment, the price fluctuation range, inventory fluctuation range, weather level, passenger flow congestion level, transportation time level, and supporting facility openness status within the corresponding time range are read, and the standardized status markers of the scenario description dimensions are mapped to scenario trigger factor nodes.

[0044] Specifically, for each check-in scenario segment, the standardized status markers of each scenario description dimension are read. The standardized status markers of the scenario description dimensions that affect the fulfillment status of the same property semantic commitment item within the same segment are combined to generate scenario triggering factors. The combination of the property semantic commitment item, the scenario description dimension name, and the standardized status markers of the scenario description dimension is used as the node naming index to establish corresponding scenario triggering factor nodes in the property semantic fulfillment graph set.

[0045] For example, quiet semantic items, nighttime, high passenger flow, and weekends can form a scene trigger factor node.

[0046] Furthermore, for each property's semantic commitment item, evidence of fulfillment of the commitment is extracted from order flow data, consultation dialogue texts, cancellation record texts, and review texts.

[0047] Specifically, semantic fragments that satisfy the semantic commitment and are confirmed in post-occupancy reviews are marked as positive fulfillment evidence; semantic fragments that are valid in some time periods but weakened in other time periods are marked as fluctuating fulfillment evidence; and semantic fragments that contradict the semantic commitment of the property and appear in cancellation records, negative reviews, or consultation abandonment texts are marked as counter-evidence.

[0048] For example, if the semantic promise of a property is quiet, then the description of "quiet at night" and "undisturbed rest" in the review text is considered positive evidence of fulfillment. The description of "quiet on weekdays and noisier on weekends" in the review text is considered fluctuating evidence of fulfillment. The description of "loud noise at night" and "excessive noise in the surroundings" in the canceled record text or negative review text is considered negative evidence.

[0049] All evidence was structured and organized according to the unique identifier of the property, the semantic commitment of the property, the context of the stay, the type of evidence, the location of the evidence text, and the time interval of the evidence to obtain a set of evidence for the fulfillment of the experience.

[0050] Furthermore, a semantic fulfillment graph set of housing is constructed, with the unique identifier of the housing as the central node and the semantic commitment items of the housing, the scene triggering factors, and the experience fulfillment evidence as the associated nodes.

[0051] In this context, the relation edges in the property semantic fulfillment graph set are stored in the form of a quadruple of start node, end node, relation type, and relation attribute. When the property semantic commitment anchor value corresponding to the property semantic commitment item is not lower than the commitment entry threshold, a commitment association is established between the property node and the property semantic commitment item node. The relation attributes of the commitment association include the property semantic commitment anchor value, the number of valid occurrence records, and the consecutive occurrence interval. When the scene trigger factor node is generated by combining the standardized state markers of the context description dimension in the property semantic commitment item and the check-in context segment, a context trigger association is established between the property semantic commitment item node and the scene trigger factor node. The relation attributes of the context trigger association include the check-in context segment number, the standardized state marker combination of the context description dimension, and the segment start and end time. When the experience fulfillment evidence and the property semantic commitment item belong to the same property and the time of the evidence occurrence can correspond to the check-in context segment or the order feedback time range, a fulfillment association is established between the property semantic commitment item node and the experience fulfillment evidence node. The relation attributes of the fulfillment association include the evidence type, the evidence text position, the evidence occurrence time, and the corresponding check-in context segment number.

[0052] S2. Receive the user's search statement, extract the search constraint information and target experience semantic items, and generate the target check-in scenario graph and the target check-in scenario neighborhood.

[0053] It receives search queries from users through the homestay search page, voice interaction portal, or intelligent question-and-answer portal, and performs text transcription, sentence segmentation, stop word removal, and homestay-related terminology normalization on the search queries to obtain standardized search text.

[0054] Entity recognition and semantic slot filling are performed on standardized search texts to extract information such as check-in area, check-in time, check-out time, number of guests, budget range, travel type, room type requirements, rule requirements, and experience preferences, forming search constraint information.

[0055] Among them, the accommodation area is identified by place name, scenic spot name, business district name and transportation station name; check-in time and check-out time are estimated by date name, relative time name and number of days of stay; budget range is identified by price name and price comparison name; and travel type is identified by semantic phrases such as family, couple, business, long stay, group travel and solo travel.

[0056] Furthermore, the retrieval constraints are divided into hard retrieval constraints and experiential retrieval constraints.

[0057] The hard search constraints include the accommodation area, check-in time, check-out time, number of guests, budget range, room type requirements, and rule requirements. The experiential search constraints include accommodation experience requirements such as quietness, proximity to scenic spots, suitability for office use, suitability for families, cooking facilities, convenient parking, comfort for long stays, and convenient transportation. For experiential search constraints that appear directly in the search query, the corresponding standardized semantic items are recorded as target experiential semantic items. For accommodation experience requirements that do not appear directly in the search query but have a stable business relationship with the travel type, supplementary identification is performed through the homestay semantic mapping table, and the standardized semantic items obtained from the supplementary identification are recorded as extended experiential semantic items.

[0058] Furthermore, a set of core experience semantic items and a set of non-core experience semantic items are generated based on the target experience semantic items and the extended experience semantic items.

[0059] Specifically, when the standardized semantic item originates from the experience retrieval constraints directly expressed in the retrieval statement, the standardized semantic item is written into the core experience semantic item set; when the standardized semantic item originates from the business association between travel type and homestay semantic mapping table, and does not conflict with hard retrieval constraints, the standardized semantic item is written into the non-core experience semantic item set.

[0060] For example, if the search query includes "bringing children" and "quiet at night", the parent-child compatible semantic item and the quiet semantic item will be written into the core experience semantic item set; when the travel type is parent-child travel, the surrounding convenience semantic item and the transportation and walking convenience semantic item will be written into the non-core experience semantic item set.

[0061] Furthermore, a target check-in scenario graph is constructed using retrieval constraint information, a set of core experience semantic items, and a set of non-core experience semantic items.

[0062] Specifically, the target check-in scenario graph is stored in the form of nodes and relation edges. Nodes include retrieval constraint nodes, core experience semantic item nodes, non-core experience semantic item nodes, and target scenario nodes; relation edges are stored in the form of quadruples of start node, end node, relation type, and relation attribute.

[0063] The following are defined as search constraint nodes: check-in area, check-in time, check-out time, number of guests, budget range, room type requirements, and rule requirements. The following are defined as core experience semantic item nodes: the standardized semantic items in the core experience semantic item set; the standardized semantic items in the non-core experience semantic item set; and the following are defined as target context nodes: the combination of check-in time, travel type, guest size structure, budget range, and check-in area. Constraint relationships are established between search constraint nodes and target context nodes; core experience semantic item nodes are established as core experience relationships with target context nodes; and non-core experience semantic item nodes are established as auxiliary experience relationships with target context nodes.

[0064] Specifically, when the retrieval constraint node originates from the standardized retrieval text and is identified as the check-in area, check-in time, check-out time, number of guests, budget range, room type requirements, or rule requirements, the retrieval constraint node is used as the starting node, and the target context node is used as the ending node. A relation edge of constraint association is established, with relation attributes including constraint name, source text fragment, and standardized constraint content. When the core experience semantic item node originates from the experience retrieval constraint directly expressed in the retrieval statement, the core experience semantic item node is used as the starting node, and the target context node is used as the ending node. A relation edge of core experience association is established, with relation attributes including the standardized semantic item name, corresponding source text fragment, and core tag. When the non-core experience semantic item node originates from the business association between travel type and homestay semantic mapping table and does not conflict with hard retrieval constraints, the non-core experience semantic item node is used as the starting node, and the target context node is used as the ending node. A relation edge of auxiliary experience association is established, with relation attributes including the standardized semantic item name, associated travel type, and auxiliary tag.

[0065] Furthermore, a target occupancy context neighborhood is generated based on the target context node.

[0066] Specifically, the check-in time, travel type, occupant structure, budget range, and check-in area in the target context node are read. Historical check-in context fragments that are consistent with the check-in area, have overlapping time ranges or are in the same holiday state, have the same or adjacent travel types, have the same or adjacent occupant structure, and have the same price range are selected from the check-in context fragments in the property semantic fulfillment graph set. The selected historical check-in context fragments are used as the initial neighborhood fragments.

[0067] The initial neighborhood segment is further narrowed based on weather conditions, passenger flow, day and night time, traffic congestion, and the availability of surrounding facilities to obtain the target occupancy context neighborhood.

[0068] Specifically, the standardized state labels of the target context node and each initial neighborhood fragment are read in terms of context description dimensions such as weather status, passenger flow status, day and night time, traffic congestion status, and the openness status of surrounding facilities. The mandatory labels are determined based on the evidence distribution of core experience semantic items in historical valid retrieval samples.

[0069] Specifically, for any core experience semantic item, the number of positive and negative evidence corresponding to the standardized state markers of each context description dimension are counted. When a change in the standardized state marker of any context description dimension causes a shift in the dominant relationship between the number of positive and negative evidence, the standardized state marker of the context description dimension is recorded as the mandatory marker for that core experience semantic item. When the mandatory marker of the initial neighborhood fragment is inconsistent with the target context node, the corresponding initial neighborhood fragment is removed. When the mandatory markers of the initial neighborhood fragments are consistent, and the standardized state markers of the other context description dimensions meet at least the neighborhood retention requirement, the corresponding initial neighborhood fragment is retained to form the target check-in context neighborhood.

[0070] It should be noted that the number of neighborhoods to be retained is determined by the number of consistent environmental states in the historical valid retrieval samples that can support the semantic fulfillment judgment of the check-in context, and the value ranges from 2 to 4.

[0071] Furthermore, to avoid the subsequent semantic fulfillment judgment of the check-in context being distorted due to an excessively wide neighborhood of the target check-in context, the context fit between each initial neighborhood segment and the target context node is calculated.

[0072] Specifically, first, a hard dimension consistency check is performed between the initial neighborhood fragment and the target context node. If any hard dimension—occupancy area, budget level, or occupancy structure—is inconsistent, the hard consistency flag is recorded as 0; if all three dimensions are consistent, the hard consistency flag is recorded as 1. The number of consistent mandatory flags related to the core experience semantic items in the initial neighborhood fragment is counted, and divided by the total number of mandatory flags to obtain the core flag fit rate. Simultaneously, the number of consistent environmental flags (excluding mandatory flags) in weather status, passenger flow status, day / night time, traffic congestion status, and the openness status of surrounding facilities is counted, and divided by the total number of remaining environmental flags to obtain the environmental flag fit rate. Finally, the context fit is calculated using the following expression: ; In the formula, Indicates the first The context fit between an initial neighborhood segment and a target context node. Indicates the first Hard, consistent labeling of an initial neighborhood segment. Indicates the first The core label fitting rate of the initial neighborhood segment Indicates the first The environmental tag fitting rate of the initial neighborhood fragment Indicates the first The normalized time offset between an initial neighborhood segment and a target context node.

[0073] It should be noted that the time offset normalization is obtained by calculating the number of days between the initial check-in start date and the target check-in start date of the initial neighborhood segment, and dividing it by the maximum allowed interval days in the neighborhood. The maximum allowed interval days in the neighborhood is obtained by statistically analyzing the time intervals of historical check-in scenario segments that maintain consistent check-in scenario semantic fulfillment judgments under the same travel type, the same holiday status, and the same core experience semantic items in historical valid retrieval samples. The preferred value range is 7 to 60 days.

[0074] When the contextual fit of an initial neighboring fragment is lower than the lower limit of neighborhood fit, the corresponding initial neighboring fragment is removed from the target occupancy context neighborhood. When the contextual fit of an initial neighboring fragment is not lower than the lower limit of neighborhood fit, the corresponding initial neighboring fragment is retained as a valid neighboring fragment in the target occupancy context neighborhood. When the number of valid neighboring fragments is lower than the lower limit of neighborhood stability, neighboring fragments are added from the removed initial neighboring fragments according to the contextual fit from high to low, until the number of valid neighboring fragments reaches the lower limit of neighborhood stability or there are no fragments to add. When the number of valid neighboring fragments is higher than the upper limit of neighborhood stability, the corresponding number of valid neighboring fragments are retained according to the contextual fit from high to low to form the target occupancy context neighborhood.

[0075] It should be noted that the upper limit of neighborhood stability is determined by statistically analyzing the distribution of the number of effective neighborhood segments in the historical valid retrieval samples after completing a check-in context semantic fulfillment judgment, and selecting the maximum number of segments that can avoid introducing weakly related segments due to an overly wide neighborhood, with a preferred value range of 40 to 120; the lower limit of neighborhood fit is determined by statistically analyzing the distribution of context fit corresponding to neighborhood segments that can support check-in context semantic fulfillment judgment in the historical valid retrieval samples, and selecting the critical fit that can stably distinguish between effective neighborhood segments and weakly related neighborhood segments, with a preferred value range of [0.45, 0.70]; the lower limit of neighborhood stability is determined by statistically analyzing the distribution of the number of effective neighborhood segments required to complete a check-in context semantic fulfillment judgment in the historical valid retrieval samples, and selecting the minimum number of segments that can ensure that the core experience semantic items have a stable fulfillment verification density, with a preferred value range of 8 to 30.

[0076] S3. Based on the retrieval constraint information, limit the candidate housing resources, map the target occupancy scenario graph to the corresponding housing semantic fulfillment graph, calculate the occupancy scenario semantic fulfillment potential of each target experience semantic item, and make a occupancy scenario semantic fulfillment judgment to form a candidate housing semantic fulfillment status table.

[0077] Using the occupancy area, check-in time, check-out time, number of guests, and budget range from the search constraints as candidate limiting conditions, the property nodes that meet the basic search conditions are selected from the property semantic fulfillment graph set to form an initial candidate property set.

[0078] The basic search criteria include: the property's location matching the accommodation area; continuous available inventory between check-in and check-out times; the property's price falling within the budget range; and the property's maximum occupancy being no less than the occupancy limit specified in the search constraints.

[0079] Furthermore, a rule consistency check is performed on the initial candidate housing set, reading the check-in rule text, room type requirement record, and rule requirement record corresponding to each initial candidate housing. When the housing rules are inconsistent with the room type requirements, pet requirements, cooking requirements, check-in time requirements, or cancellation requirements in the search constraint information, the corresponding initial candidate housing is removed from the initial candidate housing set. When the housing rules are consistent with the search constraint information, the corresponding initial candidate housing is retained, thus obtaining the candidate housing set.

[0080] Furthermore, the target occupancy scenario map is mapped to the semantic fulfillment map of each candidate property.

[0081] During mapping, the target context node, core experience semantic item node, and non-core experience semantic item node in the target check-in context graph are read. The semantic commitment item node with the same name is retrieved in the semantic fulfillment graph of the candidate property. When the target experience semantic item node and the property semantic commitment item node have the same name, a semantic mapping relationship is established between the target experience semantic item node and the property semantic commitment item node. When the names are different but they correspond to the same standard semantic item in the homestay semantic mapping table, a semantic mapping relationship is also established. When there is neither a name matching relationship nor a standard semantic item matching relationship between the two, no semantic mapping relationship is established.

[0082] Furthermore, the target occupancy context neighborhood is mapped to the scenario trigger factor node corresponding to the candidate property.

[0083] Specifically, the standardized state tag combination of each effective neighborhood segment in the target occupancy context neighborhood is read, and then the standardized state tag combination in the corresponding scenario trigger factor node of the candidate property is read. When the two are consistent in the mandatory tags related to the core experience semantic items, and the start and end times of the corresponding segments are in the same holiday state, the same day and night state, or the same occupancy duration state, the corresponding scenario trigger factor node is recorded as a valid mapping node. When the mandatory tags are inconsistent, the corresponding scenario trigger factor node is not included in the semantic realization calculation scope of the current candidate property.

[0084] Furthermore, for each candidate property and each target experience semantic item, based on the property semantic commitment anchor value, fulfillment density, and scene drift discreteness of the corresponding property semantic commitment item, the semantic fulfillment potential of the check-in context is calculated, expressed as: ; in, Indicates housing resources Target experience semantic items The semantic fulfillment potential of the check-in context. Represents the semantic item of target experience In housing The corresponding realization density within the neighborhood of the target occupancy scenario Represents the semantic item of target experience In housing The scene drift discreteness within the neighborhood of the target occupancy scenario.

[0085] It should be noted that the scene drift dispersion is obtained by statistically analyzing housing resources. Target experience semantic items The number of experience fulfillment state switching between adjacent valid neighborhood segments within the target occupancy context neighborhood is calculated by comparing the number of experience fulfillment state switching with the total number of valid adjacent neighborhood segments, with a value range of [0,1]. The fulfillment evidence density is obtained by statistically analyzing the number of positive fulfillment evidence corresponding to the target experience semantic item within the target occupancy context neighborhood of the property, and then calculating the ratio of the number of positive fulfillment evidence to the total number of positive fulfillment evidence, fluctuating fulfillment evidence, and counter-evidence that can be used to verify the target experience semantic item within the same target occupancy context neighborhood, with a value range of [0,1].

[0086] It should be noted that the semantic fulfillment potential of the check-in context adopts a form where the commitment expression and fulfillment evidence are both valid, and suppression occurs when the commitment and fulfillment deviate. This is to avoid concluding that a property can meet user needs simply because a certain experiential semantic appears frequently in the property's promotional text, and to avoid concluding that a property has stable fulfillment capabilities simply because a few historical reviews occasionally support a certain experiential semantic. When a property clearly expresses a certain experiential semantic and consistently obtains positive experiential evidence in similar check-in contexts, the semantic fulfillment potential of the check-in context increases. When the property's expression is inconsistent with historical experiential evidence, or when the experiential semantic fluctuates significantly across different check-in contexts, the semantic fulfillment potential of the check-in context is suppressed. In this way, the property's semantic expression, historical check-in experience, and contextual stability can be incorporated into the same judgment process, improving the ability to identify inflated descriptions, occasional positive reviews, and scene distortions, and reducing false recalls of properties that appear to match but have inconsistent actual check-in experiences.

[0087] Furthermore, the semantic fulfillment of the check-in context is determined based on the semantic fulfillment potential of the check-in context.

[0088] When the semantic realization potential of the occupancy scenario is not lower than the steady-state realization threshold, the housing resources will be... Target experience semantic items This is recorded as a stable realization state; when the semantic realization potential of the occupancy scenario is higher than the non-realization boundary but lower than the realization steady-state threshold, the property is considered a stable realization state. Target experience semantic items This is recorded as a swing-out realization state; when the realization potential of the occupancy scenario is not higher than the non-realization boundary, the property is... Target experience semantic items This is recorded as a status of risk of non-payment.

[0089] Specifically, for target experience semantic items in the core experience semantic item set, priority is given to recording stable redemption status and risk of non-redemption status; for target experience semantic items in the non-core experience semantic item set, priority is given to recording stable redemption status and fluctuating redemption status.

[0090] It should be noted that the steady-state fulfillment threshold is determined by statistically analyzing the semantic fulfillment potential distribution of accommodations that have completed check-in and whose review texts explicitly support the target experience semantic items in the historical valid search samples, and selecting a threshold range that can stably distinguish between accommodations with high fulfillment and accommodations with fluctuating fulfillment, with a preferred value range of [0.62, 0.82]. The non-fulfillment boundary is determined by statistically analyzing the semantic fulfillment potential distribution of accommodations that have experienced cancellation, negative reviews, or repeated rewriting of search statements in the historical valid search samples, and selecting a critical value that can stably distinguish between accommodations with fluctuating fulfillment and accommodations with non-fulfillment risk, with a value range of [0.30, 0.50].

[0091] Furthermore, the unique identifier of each candidate property, the target experience semantic item, the set of core experience semantic items or the set of non-core experience semantic items to which it belongs, the semantic mapping relationship, the effective mapping node, the semantic fulfillment potential of the check-in context, the judgment status of the semantic fulfillment of the check-in context, the fulfillment evidence density, the scene drift discreteness, and the evidence duration interval are stored accordingly to form a candidate property semantic fulfillment status table.

[0092] It should be noted that the evidence duration interval is obtained by sorting the experience fulfillment evidence by time according to the unique identifier of the property and the target experience semantic item, and grouping the experience fulfillment evidence with an interval of no more than the evidence continuity interval boundary into the same continuous evidence segment; the evidence continuity interval boundary is determined by statistically analyzing the valid evaluations of the same target experience semantic item in historical completed check-in orders, with a value range of 7 to 30 days.

[0093] S4. Calculate the counter-evidence constraint quantity of each target experience semantic item based on the counter-evidence fragments in the experience realization evidence, and perform counter-evidence constraint diversion processing. While keeping the core experience semantic items unchanged, perform neighborhood supplementary reasoning on non-core experience semantic items to form a candidate housing counter-evidence diversion table.

[0094] Furthermore, each target experience semantic item in the candidate property semantic fulfillment status table is used as the object of counter-evidence retrieval. Counter-evidence paths that correspond to the same property unique identifier, the same target experience semantic item, and the same valid mapping node are retrieved in the counter-evidence path table to form a set of counter-evidence paths to be judged.

[0095] The counter-evidence path table is obtained by structuring and organizing counter-evidence fragments in the experience fulfillment evidence. Specifically, counter-evidence fragments that are opposite to the semantic commitment of the property are extracted from cancellation record text, negative review text, consultation abandonment text, and post-check-in feedback text. Then, the counter-evidence fragments are associated with the corresponding unique property identifier, the semantic commitment of the property, check-in context fragment, counter-evidence source text, and the time of counter-evidence occurrence to form the counter-evidence path table.

[0096] More specifically, the counter-evidence path table does not record a single negative review word, but a traceable path, such as property A, quiet semantic item, high traffic situation on weekend nights, counter-evidence fragments of noise downstairs affecting rest at night; this path shows that although property A may have a quiet promise, in the context of high traffic on weekend nights, there has been experience evidence that is the opposite of quiet.

[0097] Furthermore, the set of counter-evidence paths to be determined is subjected to counter-evidence validity verification, including counter-evidence validity verification, property consistency verification, semantic consistency verification, and contextual consistency verification.

[0098] The validity verification of counter-evidence includes property consistency verification, semantic consistency verification, and contextual consistency verification. Property consistency verification confirms that the unique identifier of the property to which the counter-evidence path belongs is consistent with the candidate property. Semantic consistency verification confirms that the standard semantic item pointed to by the counter-evidence path is consistent with the target experience semantic item. Contextual consistency verification confirms that the occupancy context fragment corresponding to the counter-evidence path falls within the neighborhood of the target occupancy context. Counter-evidence paths that pass the property consistency verification, semantic consistency verification, and contextual consistency verification are recorded as valid counter-evidence paths. Counter-evidence paths that fail any of the verifications are recorded as invalid counter-evidence paths and are removed from the set of counter-evidence paths to be judged.

[0099] Furthermore, the recurrence density of the proof by contradiction and the neighborhood fit are calculated.

[0100] Specifically, the number of valid counter-evidence paths is counted, and the ratio of the number of valid counter-evidence paths to the total number of experience fulfillment evidences that can be used to verify the same target experience semantic item within the neighborhood of the target check-in context is used as the counter-evidence path recurrence density.

[0101] The number of standardized state labels that match between the occupancy context segment corresponding to the valid counter-evidence path and the neighborhood of the target occupancy context is statistically analyzed, and the ratio of the number of standardized state labels that match to the total number of standardized state labels that are compared is used as the neighborhood fit quantity.

[0102] Furthermore, for each candidate property and each target experience semantic item, the counter-evidence constraint is calculated based on the counter-evidence path recurrence density, neighborhood fit, and occupancy context semantic fulfillment potential. The expression is as follows: ; in, Indicates housing resources Target experience semantic items The counter-evidence constraint quantity. Indicates housing resources Target experience semantic items The recurrence density of the proof by contradiction path, Indicates housing resources Target experience semantic items The amount of neighboring fit.

[0103] It should be noted that the counter-evidence constraint measure adopts a form where counter-evidence reproducibility and contextual fit are both valid, and the stronger the stable fulfillment capability, the more suppressed the counter-evidence constraint is. This is to avoid a small number of irrelevant negative reviews directly affecting candidate listings, and also to avoid high-frequency counter-evidence being ignored in highly similar occupancy scenarios. When the counter-evidence path repeatedly appears in the neighborhood of the target occupancy scenario and is highly fitted to the target occupancy scenario, the counter-evidence constraint measure increases. When candidate listings have formed a high occupancy scenario semantic fulfillment potential for the same target experience semantic item, the counter-evidence constraint measure is suppressed. Through the counter-evidence constraint measure, the degree of counter-evidence reproducibility, the relevance of counter-evidence context, and the stability of positive fulfillment can be incorporated into the same judgment process, improving the ability to identify listings that seem to match but are prone to distortion in historically similar scenarios.

[0104] Furthermore, a counter-evidence diversion process is implemented.

[0105] Specifically, when the amount of counter-evidence constraint is lower than the counter-evidence constraint threshold, the corresponding target experience semantic item is recorded as a weak constraint state and retained. When the amount of counter-evidence constraint is not lower than the counter-evidence constraint threshold and is lower than the counter-evidence cutoff boundary, and the corresponding target experience semantic item is in a stable realization state, the corresponding target experience semantic item is still recorded as a weak constraint state and the corresponding counter-evidence constraint amount is retained in the candidate housing counter-evidence diversion table. When the amount of counter-evidence constraint is not lower than the counter-evidence constraint threshold and is lower than the counter-evidence cutoff boundary, and the corresponding target experience semantic item is in a swing realization state, the corresponding target experience semantic item is recorded as a restricted state. When the amount of counter-evidence constraint is not lower than the counter-evidence cutoff boundary, or the corresponding target experience semantic item is in a state of loss of realization risk and the amount of counter-evidence constraint is not lower than the counter-evidence constraint threshold, the corresponding target experience semantic item is recorded as a cutoff state. In the cutoff state, the semantic mapping relationship between the candidate housing and the corresponding target experience semantic item stops participating in subsequent ranking.

[0106] It should be noted that the counter-evidence restraint threshold is determined by statistically analyzing the distribution of counter-evidence restraint corresponding to listings that have experienced consultation abandonment, order cancellation, or negative feedback in the historical valid search samples, and selecting a critical value that can stably distinguish between listings with weak counter-evidence and listings that need to be restricted, with a value range of [0.25, 0.45]. The counter-evidence cutoff boundary is determined by statistically analyzing the distribution of counter-evidence restraint corresponding to listings whose core experience semantic items are explicitly negated in the historical valid search samples, and selecting a critical value that can stably distinguish between listings that need to be restricted and listings that need to be cut off, with a value range of [0.55, 0.75].

[0107] Furthermore, the candidate properties are further categorized by property type.

[0108] If all core experience semantic items of a candidate property are in a stable fulfillment state and none of them have entered a cutoff state, then the candidate property is recorded as a direct output class. If no core experience semantic item of a candidate property has entered a cutoff state, but at least one core experience semantic item is in a swing fulfillment state or a restricted state, then the candidate property is recorded as a restricted retention class. If a core experience semantic item of a candidate property has entered a cutoff state, then the candidate property is recorded as a replacement class, and the core experience semantic item that triggered the replacement is recorded.

[0109] Furthermore, when the number of output properties in the direct output class and the limited retention class is lower than the return size, neighborhood fill inference is initiated.

[0110] Specifically, the neighborhood replacement reasoning takes the core experience semantic item that triggers the replacement as the preservation object and does not change the set of core experience semantic items. It reads candidate properties that have not entered the direct output class and the restricted retention class from the candidate property semantic realization status table, and selects candidate properties whose core experience semantic items are in a stable realization state and whose counter-evidence restraint is lower than the counter-evidence restraint threshold as replacement candidate properties. For non-core experience semantic items in the replacement candidate properties, it retrieves the normative semantic items that belong to the same semantic category as the original non-core experience semantic items in the homestay semantic mapping table, and counts the number of times each normative semantic item and the original non-core experience semantic item co-occur in the same property multi-source semantic dataset. The normative semantic item with the highest number of co-occurrences is recorded as the adjacent normative semantic item, and the adjacent normative semantic item replaces the original non-core experience semantic item for replacement mapping.

[0111] For example, in the transportation category, when the direct transportation semantic item and the nearby scenic area semantic item co-occur most frequently, the direct transportation semantic item can be used as the adjacent standard semantic item of the nearby scenic area semantic item; in the surrounding facilities category, when the supermarket and convenience store semantic item and the surrounding restaurant and convenience store semantic item co-occur most frequently, the supermarket and convenience store semantic item can be used as the adjacent standard semantic item of the surrounding restaurant and convenience store semantic item.

[0112] It should be noted that the return size is determined by reading the display capacity of the current search page, the number of pages on the user's end, and the minimum display requirements of candidate listings on the platform. It is also calibrated by combining the average number of listings viewed by users before making a browsing decision in the historical valid search samples. The preferred value range is 10 to 50.

[0113] Furthermore, a supplementary selection criteria are determined for the candidate properties to be added.

[0114] Specifically, when the fulfillment potential of the core experience semantic item's check-in context semantic is not lower than the fulfillment steady-state threshold, the counter-evidence constraint of the core experience semantic item is lower than the counter-evidence constraint threshold, and the non-core experience semantic item has been matched with an adjacent normative semantic item under the same accommodation experience category in the homestay semantic mapping table, the replacement candidate property is written into the set of replaceable properties; when any condition is not met, the replacement candidate property is removed from the replacement candidate property.

[0115] Furthermore, the unique identifier of each candidate property, the target experience semantic item, the counter-evidence constraint quantity, the counter-evidence constraint diversion status, the property-level diversion category, the core experience semantic item that triggers the replacement, and the replacement mapping relationship are stored accordingly to form a candidate property counter-evidence diversion table.

[0116] S5. Based on the candidate listing counter-evidence diversion table, sort the final candidate listings and output the semantic retrieval data of homestay listings and the corresponding semantic fulfillment description.

[0117] Extract directly output properties, restricted and retained properties, and properties that can be supplemented into the output range through neighborhood supplementation reasoning from the candidate property counter-evidence diversion table, and merge them to form the final candidate property set.

[0118] The final candidate property list is deduplicated. When the same property unique identifier appears in multiple categories, the record corresponding to the direct output category is retained first, followed by the record corresponding to the narrowed retention category, and finally the record corresponding to the property that can be filled in is retained, ensuring that the same property is retained only once in the semantic search list.

[0119] Furthermore, a semantic fulfillment ladder ranking is performed on the final candidate housing set.

[0120] The semantic fulfillment ladder ranking includes housing allocation ladder ranking, core fulfillment ladder ranking, counter-evidence restraint ladder ranking, and evidence persistence ladder ranking.

[0121] Specifically, the final candidate housing set is sorted in a hierarchical manner, with the housing allocation category used as the sorting criterion, arranged in the order of direct output, limited retention, and supplementary housing. Among them, the direct output type of housing indicates that the core experience semantic items have not entered the cut-off state and at least one core experience semantic item is in a stable realization state; the restricted retention type of housing indicates that the core experience semantic items have not entered the cut-off state but have a swing realization state or a restricted state; and the supplementable housing indicates that the supplement is added through neighborhood supplement reasoning and the core experience semantic items still meet the stable realization requirements.

[0122] By sorting properties in a tiered manner, properties with stable fulfillment of core user experience requirements are prioritized for display, while preventing alternative properties from overshadowing those that directly meet the user's core experience requirements.

[0123] Specifically, for candidate properties within the same property-level distribution category, a core fulfillment ladder sort is performed. The core fulfillment ladder sort reads the occupancy context semantic fulfillment potential from the candidate property semantic fulfillment status table and uses the occupancy context semantic fulfillment potential corresponding to the core experience semantic item as the main sorting criterion. When a candidate property corresponds to multiple core experience semantic items, the number of core experience semantic items in a stable fulfillment state is compared first, and candidate properties with more stable fulfillment states are ranked higher. When the number of stable fulfillment states is the same, the lowest value of the occupancy context semantic fulfillment potential corresponding to the core experience semantic item is compared, and candidate properties with a higher lowest value are ranked higher.

[0124] By ranking properties according to their core features, we ensure that candidate properties do not advance to the top based on a single superior experience, but rather maintain overall stability based on the core experience semantics that users clearly care about.

[0125] Specifically, for candidate properties that are still in the same priority range after the core fulfillment ladder sorting, a counter-evidence constraint ladder sorting is performed. The counter-evidence constraint ladder sorting reads the counter-evidence constraint quantity and counter-evidence constraint distribution status from the candidate property counter-evidence distribution table. When candidate properties have the same number of core experience semantic items with stable fulfillment status, the counter-evidence constraint quantity corresponding to the core experience semantic item is compared first, and candidate properties with lower counter-evidence constraint quantity are ranked higher. When the counter-evidence constraint quantity still cannot distinguish the sorting order, the counter-evidence constraint distribution status is compared, with weak constraint status taking precedence over restrictive status, and restrictive status taking precedence over cutoff status.

[0126] By using counter-evidence to control the tiered ranking, the priority of listings with strong counter-evidence paths in similar historical occupancy scenarios is reduced, thus reducing the number of listings that superficially match the search query but whose actual accommodation experience is easily distorted from appearing in the top positions.

[0127] Furthermore, for candidate properties that remain in the same priority range after the counter-evidence constraint ladder sorting, an evidence persistence ladder sorting is performed. The evidence persistence ladder sorting is performed by comparing the evidence persistence intervals of the core experience semantic items corresponding to the candidate properties. The evidence persistence intervals are obtained by sorting the experience fulfillment evidence by time according to the unique identifier of the property and the target experience semantic item, and grouping experience fulfillment evidence with an interval between adjacent evidence occurrences not exceeding the boundary of the evidence continuity interval into the same continuous evidence segment. For the same core experience semantic item, candidate properties with a longer continuous evidence segment coverage time, a larger number of continuous evidence segments, and a more recent evidence occurrence time closer to the target check-in time are ranked higher.

[0128] By continuously ranking properties based on evidence, listings with consistent and reliable customer feedback are given priority in the display, thus reducing the impact of occasional positive reviews on the ranking.

[0129] Furthermore, the final candidate housing set, after being sorted by semantic fulfillment hierarchy, is subject to return size control.

[0130] Specifically, candidate properties in the final candidate property set are read sequentially according to the sorting order. Reading stops when the number of properties read reaches the return scale. When the number of output properties is lower than the return scale, properties that meet the supplementary admission criteria are read from the supplementary property set according to the neighborhood supplementary reasoning until the number of output properties reaches the return scale or the supplementary property set is empty.

[0131] Furthermore, a semantic fulfillment description is generated for each output property.

[0132] The semantic fulfillment description includes core experience semantic items, the semantic fulfillment potential of the corresponding check-in context for core experience semantic items, the judgment status of semantic fulfillment of check-in context, the state of counter-evidence restraint and diversion, the evidence duration interval, and the source of neighboring supplementation.

[0133] When the property belongs to the direct output category, the semantic fulfillment description marks the stable fulfillment of the core experience semantic items and the source of evidence for the corresponding experience fulfillment.

[0134] When a property is classified as a restricted or reserved property, the semantic fulfillment description marks the target experience semantic item that is in a state of fluctuating fulfillment or restricted fulfillment, and indicates that the corresponding semantic item is only for auxiliary experience reference.

[0135] When a property is a replaceable property, the semantic fulfillment description indicates that the core experience semantic items are consistently fulfilled, and the non-core experience semantic items are used to complete the neighborhood replacement reasoning through adjacent normative semantic items under the same accommodation experience category.

[0136] Furthermore, the sorted property names, prices, available booking times, core experience semantic items, property-level traffic categories, and semantic fulfillment instructions are output to form a semantic search list of homestay properties.

[0137] The semantic search list for homestay listings displays candidate listings that match the user's search query and have experience fulfillment support in the target stay context. The semantic fulfillment description displays the evidence path for candidate listings to enter the semantic search list, allowing users to view the listing matching conditions, core experience fulfillment status, and counter-evidence at the same time.

[0138] In summary, this invention improves retrieval accuracy by constructing a semantic fulfillment graph set of housing listings, enabling unified reasoning of housing listing descriptions, occupancy contexts, and historical experience evidence; it also improves retrieval accuracy by calculating the semantic fulfillment potential of occupancy contexts, thereby recognizing the stable fulfillment capability of housing listing experience semantics and reducing false recall rate; and it enhances retrieval reliability by using counter-evidence to control traffic and neighboring supplementary reasoning to suppress distorted listings and supplement alternative listings.

[0139] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A semantic retrieval method for homestay listings based on knowledge graph reasoning, characterized in that, include: Obtain a multi-source semantic dataset of homestay listings, extract semantic commitment items, scene triggering factors, and experience fulfillment evidence, and construct a semantic fulfillment graph set of homestay listings; Receive user search statements, extract search constraint information and target experience semantic items, and generate target check-in context graph and target check-in context neighborhood; Based on the retrieval constraint information, candidate properties are limited, the target occupancy scenario graph is mapped to the corresponding property semantic fulfillment graph, the occupancy scenario semantic fulfillment potential of each target experience semantic item is calculated, and the occupancy scenario semantic fulfillment is judged to form a candidate property semantic fulfillment status table. The counter-evidence constraint quantity of each target experience semantic item is calculated based on the counter-evidence fragments in the experience fulfillment evidence, and counter-evidence constraint diversion processing is performed. While keeping the core experience semantic items unchanged, perform neighborhood supplementation reasoning on non-core experience semantic items to form a candidate housing source counter-evidence diversion table; Based on the candidate listing counter-evidence diversion table, the final candidate listings are sorted, and the semantic retrieval data of homestay listings and corresponding semantic fulfillment instructions are output.

2. The semantic retrieval method for homestay listings based on knowledge graph reasoning as described in claim 1, characterized in that, The construction of the housing semantic fulfillment graph set includes: The data from the homestay platform is processed in a unified time, spatial, and text manner to obtain a multi-source semantic dataset of listings. Extract semantic commitment terms from the multi-source semantic dataset of housing listings, and generate scene triggering factors corresponding to the semantic commitment terms of housing listings based on occupancy context fragments; Extract experience fulfillment evidence corresponding to the semantic commitment items of the housing listings from a multi-source semantic dataset; Using the unique identifier of a property as the central node, and the semantic commitment items of the property, the scenario triggering factors, and the experience fulfillment evidence as the associated nodes, corresponding relationship edges are established to construct a semantic fulfillment graph set of the property.

3. The semantic retrieval method for homestay listings based on knowledge graph reasoning as described in claim 1, characterized in that, The generation of the target occupancy scenario map and the target occupancy scenario neighborhood includes: Semantic parsing of user search statements is performed to extract search constraint information, target experience semantic items and extended experience semantic items, determine core experience semantic items and non-core experience semantic items, and construct a target check-in context graph. Based on the target occupancy scenario map, historical occupancy scenario fragments that match the target occupancy scenario are selected from the property semantic fulfillment map set to form the target occupancy scenario neighborhood.

4. The semantic retrieval method for homestay listings based on knowledge graph reasoning as described in claim 3, characterized in that, The process of mapping the target occupancy scenario map to the corresponding property semantic fulfillment map includes: Based on the retrieval constraint information in the target occupancy scenario graph, candidate limiting conditions are determined. Under the candidate limiting conditions, housing nodes that meet the basic retrieval conditions are selected from the housing semantic fulfillment graph set to form an initial candidate housing set. Perform a rule consistency check on the initial candidate property set to obtain the final candidate property set; Search for semantic commitment item nodes with the same name or corresponding to the same standard semantic item in the semantic fulfillment graph of the candidate properties, and establish semantic mapping relationship; Based on semantic mapping relationships, the scene triggering factor nodes in the semantic fulfillment graph of the corresponding housing listings that correspond to both the semantic mapping relationship and the neighborhood of the target occupancy scenario are recorded as valid mapping nodes.

5. The semantic retrieval method for homestay listings based on knowledge graph reasoning as described in claim 1, characterized in that, The calculation of the contextual semantic fulfillment potential of each target experience semantic item includes: For each candidate property and each target experience semantic item, the semantic fulfillment potential of the occupancy context is calculated based on the property semantic commitment anchor value, the fulfillment density in the neighborhood of the target occupancy context, and the scene drift discreteness. The scene drift discrete quantity is obtained by statistically analyzing the number of experience fulfillment state switching between adjacent effective neighboring fragments within the target occupancy context neighborhood of the property, and then calculating the ratio of the number of experience fulfillment state switching to the total number of effective adjacent neighboring fragments.

6. The semantic retrieval method for homestay listings based on knowledge graph reasoning as described in claim 5, characterized in that, The determination of semantic fulfillment in the check-in context includes: When the semantic fulfillment potential of the check-in context is not lower than the steady-state threshold, the candidate room is recorded as a stable fulfillment state for the corresponding target experience semantic item. When the semantic fulfillment potential of the check-in scenario is higher than the non-fulfillment boundary but lower than the fulfillment steady-state threshold, the candidate room is recorded as the swing fulfillment state for the corresponding target experience semantic item. When the fulfillment potential of the check-in context semantics is not higher than the non-fulfillment threshold, the candidate property is recorded as a non-fulfillment risk state for the corresponding target experience semantic item.

7. The semantic retrieval method for homestay listings based on knowledge graph reasoning as described in claim 1, characterized in that, The calculation of the counter-evidence constraint for each target experience semantic item includes: Each target experience semantic item in the candidate housing semantic fulfillment status table is used as the object of counter-evidence retrieval. Counter-evidence paths that correspond to the same housing unique identifier, the same target experience semantic item, and the same valid mapping node are retrieved in the counter-evidence path table to form a set of counter-evidence paths to be judged. For the set of counter-evidence paths to be judged, perform property consistency verification, semantic consistency verification, and contextual consistency verification, and record the counter-evidence paths that pass the verification as valid counter-evidence paths; The recurrence density of the counter-evidence path is calculated based on the ratio between the number of valid counter-evidence paths and the total number of experience fulfillment evidence that can be used to verify the same target experience semantic item within the neighborhood of the target check-in context. Calculate the neighborhood fit based on the number of consistent standardized state labels between the occupancy scenario fragment corresponding to the valid counter-evidence path and the neighborhood of the target occupancy scenario. Based on the recurrence density of the proof-of-contrast path, the neighborhood fit quantity, and the semantic realization potential of the check-in context, the proof-of-contrast constraint quantity is calculated.

8. The semantic retrieval method for homestay listings based on knowledge graph reasoning as described in claim 7, characterized in that, The process of performing counter-evidence diversion includes: When the amount of counter-evidence constraint is lower than the counter-evidence constraint threshold, the corresponding target experience semantic item is recorded as a weak constraint state. When the amount of counter-evidence constraint is not lower than the counter-evidence constraint threshold and is lower than the counter-evidence cutoff boundary, and the corresponding target experience semantic item is in a stable realization state, the corresponding target experience semantic item is still recorded as a weak constraint state. When the amount of counter-evidence restraint is not lower than the counter-evidence restraint threshold and is lower than the counter-evidence cutoff boundary, and the corresponding target experience semantic item is in the swing realization state, the corresponding target experience semantic item is recorded as the restricted state. When the amount of counter-evidence constraint is not lower than the counter-evidence cutoff boundary, or when the target experience semantic item is already in a state of loss of realization risk and the amount of counter-evidence constraint is not lower than the counter-evidence constraint threshold, the target experience semantic item is recorded as cutoff state.

9. The semantic retrieval method for homestay listings based on knowledge graph reasoning as described in claim 8, characterized in that, The process of performing neighborhood complementation reasoning on non-core experience semantic items includes: If all core experience semantic items of the candidate property are in a stable fulfillment state and none of the core experience semantic items have entered the cut-off state, then the candidate property is recorded as the direct output class. If a candidate property does not have any core experience semantic items and enters the cut-off state, but at least one core experience semantic item is in the swing realization state or the restricted state, then the candidate property is recorded as the restricted retention class. If a candidate property has a core experience semantic item that is cut off, then the candidate property is recorded as a replacement, and the core experience semantic item that triggered the replacement is recorded. When the number of output properties in the direct output class and the restricted retention class is lower than the return size, initiate neighborhood fill-in reasoning; For non-core experience semantic items in the candidate listings, retrieve adjacent standard semantic items under the same accommodation experience category from the homestay semantic mapping table and replace the original non-core experience semantic items for the supplementary mapping; The neighborhood replacement reasoning takes the core experience semantic item that triggers the replacement as the object to be maintained, does not change the set of core experience semantic items, and selects candidate properties from the candidate property semantic realization status table whose core experience semantic items are in a stable realization state and whose counter-evidence constraint amount is lower than the counter-evidence constraint threshold as replacement candidate properties.

10. The semantic retrieval method for homestay listings based on knowledge graph reasoning as described in claim 7 or 9, characterized in that, The output semantic search data for homestay listings and the corresponding semantic fulfillment instructions include: The directly output type of housing, the restricted and retained type of housing, and the supplementary housing that enters the output range through neighborhood supplementation reasoning are merged to form the final candidate housing set. After deduplication of the final candidate listings, semantic fulfillment ladder sorting is performed to generate semantic fulfillment descriptions for each output listing and output the semantic retrieval data for homestay listings. The semantic fulfillment ladder ranking includes housing allocation ladder ranking, core fulfillment ladder ranking, counter-evidence restraint ladder ranking, and evidence continuity ladder ranking. The housing allocation hierarchy is arranged in the following order: direct output type, limited retention type, and available replacement housing. The core fulfillment ladder ranking refers to ranking based primarily on the fulfillment potential of the check-in context semantics corresponding to the core experience semantic items. The aforementioned counter-evidence constraint ladder sorting refers to sorting based on the counter-evidence constraint quantity and counter-evidence constraint diversion status corresponding to the core experience semantic item. The evidence persistence ladder sorting refers to sorting based on the evidence persistence interval, the number of consecutive evidence segments, and the time of the most recent evidence occurrence corresponding to the core experience semantic item.

Images