Intelligent property right retrieval and matching system based on knowledge graph

Abstract

The invention relates to the technical field of intelligent retrieval and matching, in particular to a property intelligent retrieval and matching system based on a knowledge graph. The method comprises the following steps: a data acquisition and processing unit acquires property right data and preprocesses the property right data; the knowledge modeling and graph construction unit constructs the preprocessed property right data into a knowledge graph based on entity extraction, relation extraction and attribute labeling; the retrieval analysis and semantic understanding unit is used for receiving a retrieval request input by a user, converting the retrieval request into a structured retrieval statement matched with the knowledge graph based on a dynamic multi-hop reasoning intention path discovery scheme, and analyzing a retrieval intention of the retrieval request through a context modeling and semantic disambiguation method. According to the method, a semantic understanding mechanism combining a dynamic multi-hop reasoning intention path discovery scheme with context modeling and semantic disambiguation is introduced, so that a complex retrieval request input by a user in a natural language form can be accurately analyzed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent retrieval and matching, in particular to a property right intelligent retrieval and matching system based on a knowledge graph. BACKGROUND

[0002] With the increasing complexity of social and economic activities, property right related information (including property right transaction records, legal provisions, asset information, etc.) presents the characteristics of large data volume, extensive sources and heterogeneous formats. These data are usually scattered in multiple independent systems such as real estate registration, business management, financial credit, judicial judgment and contract management, forming serious "data islands", which leads to difficulties in information interconnection and association. Traditional property right retrieval systems are mostly based on relational databases and keyword matching technology, which can only support simple and explicit conditional queries and cannot effectively integrate multi-source heterogeneous data, let alone understand complex retrieval intentions expressed in natural language, such as "the house under the name of Zhang San mortgaged by the bank and not traded in the past three years" or "the stock right acquired through inheritance and actually held by his son". In addition, there are problems such as entity homonymy (e.g. multiple "Zhang San"), ambiguous reference (e.g. "he" and "the house") and various relationship expressions (e.g. "transfer", "sell" and "change ownership") in property right data. Traditional methods lack the ability of context modeling and semantic disambiguation, which can easily lead to inaccurate retrieval results or missing key information. More importantly, many property right states (such as actual control, proxy holding relationship and transfer restrictions) are not directly recorded but are implied in multi-layer legal and transaction relationship chains. Existing systems lack the ability of logical reasoning and path inference for implicit relationships and cannot discover deep associations. At the same time, the retrieval results often lack credibility evaluation and reasoning basis explanation, which makes it difficult to meet the requirements of result explainability and reliability in high-risk scenarios such as judicial review and compliance supervision. Therefore, a property right intelligent retrieval and matching system based on a knowledge graph is designed. SUMMARY

[0003] The present application aims to provide a property right intelligent retrieval and matching system based on a knowledge graph to solve the problem of low query accuracy and low intelligence level caused by the data island, weak semantic understanding, difficulty in mining complex relationships, insufficient reasoning ability and untrustworthy results of traditional property right retrieval systems.

[0004] To achieve the above-mentioned purpose, the present application aims to provide a property right intelligent retrieval and matching system based on a knowledge graph, which comprises: a data acquisition and processing unit, which acquires property right data and pre-processes the property right data; The knowledge modeling and graph construction unit constructs the preprocessed property right data into a knowledge graph based on entity extraction, relation extraction and attribute annotation. The retrieval analysis and semantic understanding unit is configured to receive a retrieval request input by a user, and convert the retrieval request into a structured retrieval sentence matched with the knowledge graph based on an intention path discovery scheme of dynamic multi-hop reasoning, and analyze the retrieval intention of the retrieval request through context modeling and semantic disambiguation. The graph query and reasoning unit is configured to receive the structured retrieval sentence, and obtain candidate results related to the retrieval intention based on the knowledge graph. The matching calculation unit is configured to perform semantic similarity calculation on the candidate results based on cosine similarity, and output an optimal retrieval result.

[0005] As a further improvement of the technical solution, the property right data at least includes property right transaction records, legal provisions and asset information.

[0006] As a further improvement of the technical solution, the knowledge modeling and graph construction unit constructs the preprocessed property right data into a knowledge graph, including the following steps: S1.1, text analysis is performed on the preprocessed property right transaction records, legal provisions and asset information, core entities related to property rights are identified and extracted, including at least property right persons, transaction parties, contract subjects, legal provision numbers, property subject names and asset categories; S1.2, based on syntactic dependency analysis and deep semantic recognition methods, property right relationships are identified from the semantic context between the core entities, including at least transfer relationships, mortgage relationships, constraint relationships and inheritance relationships; S1.3, descriptive attribute information is supplemented for the core entities and property right relationships, wherein the attributes corresponding to the property right transaction records include transaction time, transaction amount and transaction mode, the attributes corresponding to the legal provisions include provision content, applicable conditions and legal validity, and the attributes corresponding to the asset information include asset location, asset valuation and asset status; S1.4, the core entities are taken as graph nodes, the property right relationships are taken as graph edges, and a weighted directed graph structure, i.e. a knowledge graph, is generated in combination with the attribute information.

[0007] As a further improvement of the technical solution, the retrieval analysis and semantic understanding unit includes a retrieval analysis module and a semantic understanding module. The retrieval analysis module is configured to perform word segmentation, part-of-speech tagging and intention recognition on the retrieval request input by the user based on the retrieval request, extract keywords and retrieval conditions contained therein, and convert the keywords and retrieval conditions into a structured retrieval sentence based on an intention path discovery scheme of dynamic multi-hop reasoning. The semantic understanding module identifies the real intention of the user based on the structured search sentence through context modeling and semantic disambiguation methods, and maps the search request to entity, relationship and attribute expression in the knowledge graph.

[0008] As a further improvement of the technical solution, the search analysis module extracts the keywords and search conditions contained therein, and converts the keywords and search conditions into a structured search sentence based on an intent path discovery scheme of dynamic multi-hop reasoning, including the following steps: S2.1, performing word segmentation operation on the search request input by the user, dividing the search request into a word sequence; S2.2, performing part-of-speech tagging on the word sequence; S2.3, based on the word sequence and its part-of-speech tagging, generating a sorted intent path list using the intent path discovery scheme of dynamic multi-hop reasoning; S2.4, converting the intent path list into a structured search sentence of the knowledge graph.

[0009] As a further improvement of the technical solution, in S2.3, the intent path discovery scheme of dynamic multi-hop reasoning is used to generate a sorted intent path list, including the following steps: S2.31, input the word sequence into the knowledge graph generated in step S1.4 through the entity linker for comparison, establish a corresponding link relationship between the identified target word and the graph node in the knowledge graph, and obtain an anchor entity set; S2.32, input the verb and relationship phrase contained in the search request into the relationship mapping model to generate the corresponding semantic embedding vector; at the same time, encode the relationship type set defined in the knowledge graph into an embedding vector space; calculate the cosine similarity between the request relationship phrase vector and the relationship type vector, and select the candidate relationship closest to the semantics; S2.33, taking the anchor entity in step S2.31 as the starting point, combining the candidate relationship obtained in step S2.32 to perform path expansion in the local neighborhood of the knowledge graph with a limited step length, and generating a candidate path set; S2.34, based on the semantic matching degree, path confidence and jump penalty factor, score using a ranking function, and sort the candidate paths; finally select the top t paths as the intent analysis result, and form a sorted intent path list.

[0010] As a further improvement of the technical solution, the semantic understanding module identifies the real intention of the user based on the structured search sentence through context modeling and semantic disambiguation methods, and maps the search request to entity, relationship and attribute expression in the knowledge graph, including the following steps: S3.1, by semantic coding of the retrieval request and the session history of the retrieval request, constructing a context semantic vector, and associating keywords, time expressions and pronouns in the retrieval request with context entities, generating an enhanced query representation that integrates context information; S3.2, for entities, relationships and attribute candidates with ambiguity in the retrieval request, multi-factor scoring and screening are performed in combination with the context modeling results and the knowledge graph adjacency relationship; S3.3, filling the screened entities, relationships and attributes into the predefined semantic slots according to the semantic roles, and constructing logical triples, mapping the retrieval request into graph nodes, graph edges and attribute information in the knowledge graph, forming a structured query template, outputting a sorted candidate result list, and introducing evidence source credibility modeling and comparative path synthesis to optimize the sorted candidate result list.

[0011] As a further improvement of the technical solution, in S3.3, the evidence source credibility modeling and comparative path synthesis are introduced to optimize the sorted candidate result list, including the following steps: S3.31, for each path in the sorted candidate result list, extract its supporting evidence source; S3.32, set initial weights for different supporting evidence sources, and calculate the comprehensive credibility score of the candidate path ; S3.33, the candidate path with a comprehensive credibility score greater than the path threshold a is the anchor point, and the comparative path is generated in the local neighborhood of the knowledge graph; S3.34, compare the anchor point path and its comparative path dimension by dimension, and perform difference evaluation; S3.35, introduce the credibility score and the difference evaluation result into the sorting function together, and re-score and optimize the candidate result list.

[0012] As a further improvement of the technical solution, the graph query and reasoning unit includes a graph query module and a graph reasoning module; The graph query module is used to receive the structured retrieval statement output by the retrieval analysis and semantic understanding unit, and perform query operation on the knowledge graph constructed in S1.4; through entity index, relationship index and attribute index mechanism, retrieve entity nodes, relationship edges and their attribute information matching the retrieval condition from the knowledge graph, and generate a preliminary candidate result set; The graph reasoning module generates a semantic enhanced reasoning result set based on the preliminary candidate result set, in combination with the relationship chain and reasoning rules in the knowledge graph, performs logical reasoning, path inference and semantic expansion.

[0013] As a further improvement of the technical solution, the matching calculation unit performs semantic similarity calculation on the candidate results based on cosine similarity, and outputs the optimal retrieval result, including the following steps: S4.1, for the candidate results output by the graph query and reasoning unit, extracting the multi-dimensional feature representation of entities, relationships and attributes; S4.2, the semantic representation of the user search request and the feature vector of the candidate results are mapped to a unified representation space through a feature alignment method; S4.3, the semantic similarity score between the user search request representation and the candidate result representation is calculated using cosine similarity; S4.4, the candidate results are sorted from high to low according to the semantic similarity score, and the top b results are selected as the optimal retrieval result output.

[0014] Compared with the prior art, the beneficial effects of the present application are: 1. In the intellectual property intelligent retrieval and matching system based on knowledge graph, the intention path discovery scheme of dynamic multi-hop reasoning is introduced, and the semantic understanding mechanism combining context modeling and semantic disambiguation can accurately analyze the complex retrieval request input by the user in natural language form. In the entity linking, relationship mapping and path expansion process, the link confidence, semantic similarity, path confidence and context dependency are considered comprehensively, and the intention path generation is optimized through beam search and sorting function, effectively solving the problems of entity homonymy, relationship expression diversity and ambiguous reference in property data. At the same time, combined with the session history, the co-reference resolution and context enhancement are carried out, so that the system can accurately identify the reference objects such as "he" and "the property", realize the leap from surface keyword matching to deep semantic understanding, and greatly improve the analysis ability of complex and implicit query intention and the relevance and accuracy of the final retrieval result.

[0015] 2. In the intellectual property intelligent retrieval and matching system based on knowledge graph, on the basis of knowledge graph query, an inference optimization mechanism combining evidence source credibility modeling and comparative path synthesis is constructed, not only the matching results are retrieved from the graph, but also the implicit ownership relationship (such as inheritance chain, transfer path) can be found through logical reasoning, and the path comprehensive credibility is calculated through weighted evidence source (such as manual annotation, external authoritative library), so as to improve the result reliability. Further, by generating semantic related but structurally different comparative paths and performing difference evaluation, the system can identify the deviation or information missing of single path, so as to preferentially recommend the results with more complete semantic coverage and more stable logic in the sorting. This mechanism not only improves the quality and diversity of the candidate result set, but also provides traceable reasoning basis and confidence score for the final output, enhances the explainability and user trust of the system, and is especially suitable for high-risk decision-making scenarios such as property ownership determination and legal compliance review. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a whole flow chart of the present application; The meanings of various reference numerals in the drawings are as follows: 1, data acquisition and processing unit; 2, knowledge modeling and graph construction unit; 3, retrieval analysis and semantic understanding unit; 4, graph query and reasoning unit; 5, matching calculation unit. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0018] Embodiment: Please refer to Figure 1 As shown in the figure, a knowledge graph-based intellectual property right retrieval and matching system is provided, comprising: The data acquisition and processing unit 1 acquires property right data and pre-processes the property right data; In this embodiment, the property right data at least includes property right transaction records, legal provisions, and asset information.

[0019] The knowledge modeling and graph construction unit 2 constructs the pre-processed property right data into a knowledge graph based on entity extraction, relationship extraction, and attribute annotation; In this embodiment, the knowledge modeling and graph construction unit 2 constructs the pre-processed property right data into a knowledge graph, including the following steps: S1.1, text analysis is performed on the pre-processed property right transaction records, legal provisions, and asset information, core entities related to property rights are identified and extracted, at least including property right persons, transaction parties, contract subjects, legal provision numbers, property right subject matter names, and asset categories; S1.2, based on syntactic dependency analysis and deep semantic recognition methods, property right relationships are identified from the semantic context between the core entities, at least including transfer relationships, mortgage relationships, constraint relationships, and inheritance relationships; S1.3, descriptive attribute information is supplemented for the core entities and property right relationships, wherein the attributes corresponding to the property right transaction records include transaction time, transaction amount, and transaction method, the attributes corresponding to the legal provisions include provision content, applicable conditions, and legal validity, and the attributes corresponding to the asset information include asset location, asset valuation, and asset status; S1.4, the core entity is taken as a graph node, the property right relationship is taken as a graph edge, a weighted directed graph structure is generated by combining attribute information, that is, a knowledge graph, graph organization and representation of property right data are realized, the knowledge graph is stored in a graph database, an index mechanism is established to support subsequent retrieval query, logical reasoning and dynamic update.

[0020] The retrieval analysis and semantic understanding unit 3 is used for receiving a retrieval request input by a user, and converting the retrieval request into a structured retrieval sentence matched with the knowledge graph based on an intention path discovery scheme of dynamic multi-hop reasoning, and analyzing a retrieval intention of the retrieval request by a context modeling and semantic disambiguation method; In this embodiment, the retrieval analysis and semantic understanding unit 3 includes a retrieval analysis module and a semantic understanding module. The retrieval analysis module performs word segmentation, part-of-speech tagging and intention recognition on the retrieval request input by the user, extracts keywords and retrieval conditions contained therein, and converts the keywords and retrieval conditions into executable structured retrieval sentences based on an intention path discovery scheme of dynamic multi-hop reasoning, so as to ensure that the subsequent query can match the knowledge graph data format. Further, the retrieval analysis module extracts keywords and retrieval conditions contained therein, and converts the keywords and retrieval conditions into structured retrieval sentences, including the following steps: S2.1, performing word segmentation on the retrieval request input by the user, and dividing the retrieval request (the retrieval request is a continuous natural language text) into a word sequence; S2.2, performing part-of-speech tagging on the word sequence, and identifying names, places, times, numerical values and specific terms in the property right field in the word sequence; S2.3, based on the word sequence and the part-of-speech tagging, generating a sorted intention path list by using an intention path discovery scheme of dynamic multi-hop reasoning; Among them, the dynamic multi-hop reasoning intent path discovery scheme addresses the problem of parsing complex intents in user natural language retrieval requests in property rights intelligent retrieval scenarios. Specifically, when a user inputs a query involving multiple entities and relationships (such as "the property that Zhang San mortgaged to Li Si last year"), the system needs to identify core entities (such as Zhang San, Li Si, and the property) and relationships (such as mortgage) from the knowledge graph, and dynamically explore potential paths connecting these entities to accurately capture the user's potentially implicit multi-hop intents (such as assets indirectly related through transfer or inheritance relationships). This solves the challenge of traditional methods struggling to handle complex semantics and indirect relationships, ensuring that retrieval requests can be transformed into executable queries that match the knowledge graph structure. Through dynamic multi-hop reasoning and path ranking mechanisms, this scheme can effectively handle complex multi-hop queries, significantly improving the accuracy and robustness of intent parsing. Existing methods are often limited to single-hop matching or simple keyword-based mapping, which cannot deeply explore the implicit relationships in knowledge graphs. This solution combines entity linking, relationship mapping, bundle search path expansion, and multi-factor ranking (such as semantic matching degree, path confidence, and hop count penalty), and introduces evidence credibility modeling and contrast path synthesis to more comprehensively capture users' deep intentions, reduce ambiguity, support intelligent processing of indirect associations and long-path reasoning, and ultimately improve the semantic understanding ability and result relevance of the retrieval system. The intent path discovery scheme, which utilizes dynamic multi-hop reasoning, generates a sorted list of intent paths, including the following steps: S2.31. Input the word sequence into the knowledge graph generated in step S1.4 using the entity linker for comparison. (The entity linker is used to establish a correspondence between the target words in the retrieval request and the unique entity nodes in the knowledge graph. It is implemented using publicly available entity linking tools (such as DBpedia).) The Spotlight algorithm identifies target words (words already bound to graph node IDs, such as Zhang San, Li Si, and Real Estate X) and establishes corresponding links with graph nodes (with unique identifier IDs) in the knowledge graph, resulting in a set of anchor entities. During the entity link comparison process, multiple candidate entities are generated. A matching score (confidence) is calculated between each candidate entity and the input target word. A link confidence threshold is then used for filtering. When the entity linker maps Zhang San to the knowledge graph, it may find multiple entities with the same name (e.g., Zhang San (property owner A), Zhang San (property owner B)). The entity linker calculates a confidence score for each candidate entity. If the highest score is greater than the threshold, the entity is recognized as the final identification result and a corresponding relationship is established with the node ID. If the highest score is lower than the threshold, it is considered unreliable and may enter ambiguity resolution or be discarded. In this embodiment, the link confidence threshold... A link is considered successful only if its confidence score is higher than the threshold; if multiple candidate entities exist, they are sorted from highest to lowest confidence score, and the number of candidates is retained. That is, each mentioned entity can retain a maximum of three candidate nodes as anchor entities; S2.32. Input the verbs and relational phrases (such as "holding" and "mortgaged") contained in the retrieval request into the relation mapping model to generate corresponding semantic embedding vectors; simultaneously, encode the set of relation types defined in the knowledge graph into an embedding vector space; calculate the cosine similarity between the request relational phrase vector and the relation type vector (vectorize the request relational phrase and each relation type in the knowledge graph to obtain relational phrase vectors and relation type vectors, and calculate the similarity score between them in the semantic space using the cosine similarity formula; the score range is...). (The closer the value is to 1, the more semantically similar the relation). Candidate relations with the closest semantic similarity are selected. In this embodiment, the cosine similarity threshold is used. Filter candidate relationships (select those with similarity greater than the cosine similarity threshold) (of), retain the first one with the highest similarity. One candidate relation type; The relation mapping model takes verbs and relational phrases in the retrieval request and their contextual entities as input. First, it models their features through a Bi-LSTM (Bilingual Long Short-Term Memory) network. (Specifically, the verbs and relational phrases in the retrieval request are segmented and vectorized into fixed-dimensional dense vector sequences. Then, they are input into a Bi-LSTM network, which models contextual dependencies through forward and backward recurrent units to obtain the semantic encoding representation of each word in the bi-directional context. Based on this, an attention mechanism is introduced to assign higher weights to the key segments in the sequence that are most relevant to the relational semantics and perform weighted aggregation to obtain a global feature vector that can highlight the core semantics and suppress redundant noise, thus realizing deep semantic feature modeling of the relational phrases in the retrieval request.) In the intermediate layer, the expression of key relational segments is enhanced using an attention mechanism or semantic alignment method, and semantic embedding vectors aligned with the relational space of the knowledge graph are generated. S2.33. Starting from the anchor entity in step S2.31, and combining the candidate relations obtained in step S2.32, perform path expansion with a finite step length in the local neighborhood of the knowledge graph to generate a candidate path set containing multiple entities and relations. Specifically: Based on the candidate set of anchor entities in S2.31 and the candidate set of relations in S2.32, an initial search queue is constructed; for each valid anchor entity, a first-hop candidate node pair (anchor entity ID, candidate relation type, and candidate target entity ID) is generated, and a cumulative confidence score is recorded for each candidate (initially the entity link confidence score multiplied by the relation similarity of the edge); a bundle search strategy is used to control the expansion scale, with the bundle width set to... To limit the number of candidate paths reserved for each expansion layer, so as to cover high-quality paths under controllable computational load; Based on the initial search queue, path expansion is performed by layer (hop number) until the maximum hop number is reached (i.e. the maximum number of layers or steps allowed to expand along the relationship edges in the graph from the starting entity) or no new path is expandable: each expansion selects the top path from the currently reserved candidate paths, and for each path, enumerates the candidate relationships provided by S2.32 at its current endpoint and retrieves the corresponding adjacent entity nodes to generate new paths; The path confidence is calculated in real time for each newly generated path , which is defined as the product of the relationship similarity and the average of the entity link confidence in the path, i.e. ; wherein, is the confidence of the th relationship edge in the path, reflecting the credibility of the relationship in the knowledge graph, is the semantic relevance of the th node in the path to the input entity or context, is the number of relationships contained in the path, i.e. the path hop number, is the total number of nodes involved in the path; To support semantic matching, the semantic vector representation of each path (obtained by concatenating or averaging the node and relationship vectors in the path in order) is calculated, and the similarity with the semantic vector of the original query is calculated as ; After each hop, all newly generated paths are filtered and the top paths are reserved to control the search size; the final output of this step S2.33 is the set of candidate paths generated within the hop limit (at least containing the path sequence, , , hop number); S2.34, based on semantic matching degree, path confidence and hop penalty factor, use ranking function to score and sort the candidate paths (according to all candidate paths are sorted from high to low); finally, the top t paths are selected as the intent resolution result, forming the sorted intent path list; Wherein, the ranking function for scoring is: ; wherein, is the similarity between the query semantic vector and the path semantic vector, is the path confidence, is the path score, path hop count, similarity weight, value range [0.4, 0.7], empirical value, usually the highest weight, because the semantic similarity with the user query is the core, path confidence weight, value range [0.2, 0.4], empirical value, second weight, used to ensure the reliability of the path in the knowledge graph, path hop count weight, value range [-0.2, -0.05], empirical value, used to punish paths with too many hops and encourage the system to return more direct and concise answers; according to Sort from high to low, output the top t ranked intent paths as the final parsing result; each output path is accompanied by an executable structured query template (such as a SPARQL template), path score, value, entity node ID list involved and its link confidence, for the graph query and reasoning unit 4 to directly execute; for the case where no valid path is generated, the following priority is taken to adopt a fallback strategy: reduce or value retry, expand , or submit the query to the semantic classification module to adopt an alternative parsing strategy; S2.4, convert the intent path list into a structured retrieval statement executable by the knowledge graph, which is based on a query language (such as SPARQL) or equivalent logical expression construction, used to drive subsequent graph query and reasoning.

[0021] The semantic understanding module identifies the user's real intent through context modeling and semantic disambiguation methods based on the structured retrieval statement, and maps the retrieval request to entity, relationship and attribute expressions in the knowledge graph, achieving accurate understanding and intent analysis at the semantic level; The semantic understanding module identifies the user's real intent through context modeling and semantic disambiguation methods, and maps the retrieval request to entity, relationship and attribute expressions in the knowledge graph, including the following steps: S3.1, by semantic coding of the retrieval request and the session history of the retrieval request, constructing a context semantic vector, and associating the keywords, time expressions and pronouns in the retrieval request with the context entities, generating an enhanced query representation that fuses context information, specifically: in the process of constructing the context semantic vector, first segment the current retrieval request and its corresponding session history text, and then input the semantic coding model (such as bidirectional Transformer) for context modeling (the process of context modeling is: after the retrieval request and its session history are segmented and word vectorized, they are input into the semantic coding model (such as bidirectional Transformer): the model is composed of multiple layers of encoders stacked together, each layer contains multi-head self-attention mechanism and feedforward neural network, first calculate the global dependency between words in the sequence through self-attention mechanism, then pass through residual connection and layer normalization to keep information stable transmission, and then pass through the feedforward network for nonlinear feature transformation; on the data stream, the input sequence is first embedded in the vector space and added with position encoding, and then passed through multiple layers of encoders to obtain the context-enhanced semantic representation vector, where the output of each position contains not only the semantic of the word itself, but also the relevant information of the session history and the global context, providing semantic support for subsequent entity mapping and semantic disambiguation), the dependence and semantic association between the retrieval request and the historical conversation are captured through attention mechanism, thereby generating an enhanced semantic vector that fuses context information, which can represent the semantic content of the current request and its context relationship in the session history at the same time, providing semantic support for subsequent entity, relationship and attribute mapping, for example: first identify the keywords (Zhang San), time expressions (last year) and pronouns (he, the property) in the retrieval request, then combine the known entities in the session history or knowledge graph to determine the specific objects or time ranges they point to, for example, in the session history, Zhang San bought property X in 2024, and the current retrieval request is that he sold the property last year, the system maps him to Zhang San through co-reference resolution, maps the property to property X, and maps last year to 2024, thereby associating natural language expressions with context entities and ensuring the accuracy of semantic analysis; S3.2, for the ambiguous entity, relationship and attribute candidates in the retrieval request, multi-factor scoring and screening are carried out in combination with the context modeling results and the knowledge graph adjacency relationship, to ensure that the elements finally mapped into the knowledge graph are unique and accurate, specifically: first, a semantic vector is generated for each candidate, and a semantic matching degree is calculated in combination with the context semantic vector (each candidate entity, relationship or attribute is encoded into a fixed dimension vector through a pre-trained language model (such as BERT), and the retrieval request and the session history are generated into a context semantic vector through the same model, then the matching score between the candidate vector and the context vector is calculated by using a vector similarity measurement method (such as cosine similarity), to obtain the semantic matching degree), at the same time, the connection rationality of the candidate entity and the related nodes is analyzed by using the knowledge graph adjacency relationship (first, according to the node ID of the candidate entity in the knowledge graph, the relationship edges and the connected nodes directly adjacent to it are traversed, and the adjacent information related to other entities or attributes in the retrieval request is extracted, then whether these connections meet the semantic and logical constraints is evaluated, for example, whether the relationship type is matched, whether the path length is reasonable, whether there is a conflict or repeated connection, and the adjacent rationality of each candidate entity is quantified as a score value, i.e. the adjacency relationship support degree), the semantic matching degree, the adjacency relationship support degree and the entity linking confidence are weighted and fused to form a comprehensive score, then the candidates are sorted and the unique mapping result with the highest score and meeting the threshold is screened out; for example, the retrieval request is “Zhang San invests in a company located in Shanghai”, if there are multiple entities with the same name “Zhang San” in the knowledge graph, then through the semantic matching with the context session history and the combination of the adjacency relationship score of the candidate company node and “Shanghai”, “Zhang San (investor A)” can be finally determined as the unique entity mapping, at the same time, “investment” is mapped to “investment relationship”, and “company” is mapped to the specific company node, to realize ambiguity resolution; S3.3, fill the screened entities, relations and attributes into predefined semantic slots according to semantic role (first determine the slot type according to the semantic structure of the search request (such as the subject slot corresponds to entity A, the predicate slot corresponds to the relation type, and the object slot corresponds to entity B or the attribute constraint), and then match and fill the candidate entities, relations and attributes screened after ambiguity resolution and scoring according to semantic roles and corresponding slots one by one), and construct logical triples or multi-tuples (such as 〈entity A, relation, entity B〉 or 〈entity, attribute, value〉), map the search request into graph nodes in the knowledge graph (such as Zhang San, house property X), graph edges (here, the graph edge corresponds to the relation category between entities in the knowledge graph, such as holding, mortgage, which is the unique identifier of the edge type) and attribute information (the attribute value or restriction condition corresponding to the entity or relation, such as transaction amount, transaction time, asset status, which can be used for accurate matching of conditions in the search request), form a structured query template, add confidence scores and source basis to each semantic mapping result, output a sorted candidate result list, and preferentially select high-confidence templates for query execution, while labeling low-confidence results to support subsequent human-machine clarification interaction, and introduce evidence source credibility modeling and comparative path synthesis to optimize the sorted candidate result list, which is used to improve the reliability and semantic accuracy of the candidate path result; Further, evidence source credibility modeling and comparative path synthesis are introduced, mainly aiming at the core problem of uneven quality and ambiguous credibility of candidate paths in the property right search scene. In the serious field of property rights involving legal effectiveness and asset transactions, the candidate paths generated automatically by the system may be derived from data of different credibility (such as high-quality data manually annotated and noisy data automatically extracted), and a single path may not fully reflect the complex property rights (such as assets with competing ownership). This method directly solves the key problem of how to identify the most reliable and comprehensive property rights chain in multiple paths by quantitatively evaluating the credibility of evidence sources and generating comparative paths, effectively preventing search bias caused by relying on low-quality or one-sided evidence. The prior art usually only relies on semantic similarity or single confidence for sorting. This scheme realizes the qualitative change from finding relevant paths to finding reliable and comprehensive paths through the synergistic mechanism of evidence credibility quantification and comparative path synthesis. The prior art treats triples from different sources (such as manual verification and automatic extraction) in the knowledge graph equally, and is prone to local optimal paths. However, this method assigns weights to evidence sources and calculates comprehensive credibility, and preferentially adopts high-authority data. At the same time, it automatically generates different comparative paths, which can expose potential contradictions and discover complementary information, thereby systematically improving the reliability, robustness and decision support capability of the search result, especially in property disputes or complex ownership reasoning; Introducing evidence source credibility modeling, comparative path synthesis to optimize the sorted candidate result list, including the following steps: S3.31, for each path in the sorted candidate result list, extract its supporting evidence sources, including at least the triple relationship in the knowledge graph, external knowledge base reference, and historical interaction log record; S3.32, set initial weights for different supporting evidence sources (such as high weights for manually annotated knowledge bases, and low weights for automatically extracted or weakly supervised generated knowledge bases), and calculate the comprehensive credibility score of the candidate path , where is the number of evidence relied on by the path, is the weight of the th evidence (high weight for high credibility sources, such as manually annotated knowledge bases), is the credibility of the th evidence itself (between 0 and 1)); S3.33, candidate paths with a comprehensive credibility score greater than the path threshold a are anchor points, and within the local neighborhood of the knowledge graph, generate semantically related but different comparative paths, ensuring that different paths form a complement or contrast in entity coverage, relationship semantics, and jump length, specifically: first, take the key entities and relationships in the anchor path with a comprehensive credibility score greater than the path threshold a as the starting point, take out the adjacent relationships and entities of the entity, filter out candidate relationships similar but different from the original path relationship type, and combine them with the adjacent entities to generate new path extensions; then limit the number of jumps and calculate the semantic vector of the new path to ensure that the path is related to the original anchor path in semantic space while differing in entity coverage or relationship combination; next, calculate the difference score (such as entity overlap rate) for each newly generated path, and remove paths with too large semantic deviation or logical conflicts; finally, output the set of comparative paths that meet the conditions for comparison and optimization sorting with the anchor path; S3.34, compare the anchor path and its comparative paths dimension by dimension, and perform difference evaluation to measure the difference in semantic relevance, evidence source consistency, and reasoning logic rationality, first align the entity and relationship vectors at corresponding positions in the two paths in order, then calculate the difference index in each dimension, including entity ID matching, relationship type similarity, attribute constraint difference, and jump number difference; next, perform weighted summation or normalization processing on the difference results of each dimension to obtain the overall difference score of each pair of paths , which reflects the difference in entity coverage, relationship combination, and structural logic between the two paths, and the larger the value, the higher the difference, which can be directly used in the sorting function to optimize the candidate result list;​ S3.35. Introduce the credibility score and the difference assessment result into the ranking function to re-score and optimize the candidate result list; The ranking function, which incorporates the credibility score and the difference assessment results, is as follows: ; In the formula, The overall credibility score of the candidate paths. The result of the difference assessment between the anchor path and the comparison path, i.e., the overall difference score, This is a weighting coefficient for the overall confidence score, empirically chosen with a value ranging from [0.1, 0.3]. As a supplement and correction to the basic ranking function, the weight should not be too high to avoid completely overturning the dominant roles of semantics and confidence. The weighting coefficient for the difference assessment results is empirically selected, with a value range of [0.05, 0.15]. Diversity is encouraged, and the weight is usually the lowest, aiming to fine-tune the results without significantly affecting the main ranking. In this embodiment, steps S3.3 to S3.34 mainly improve the quality of the initial candidate path list. Through evidence source credibility modeling and comparative path generation, the system re-evaluates the reliability and semantic coverage of each candidate path, thereby adjusting the priority of candidate results or eliminating low-quality paths to ensure that the candidate set itself is more credible and complete. After the candidate set is optimized, it is still necessary to map the paths to the final executable query ranking. Here, the credibility score and the difference evaluation result are combined with the original semantic matching degree, path confidence, and hop count index to calculate the final path score. This is used for final path sorting and optimization.

[0022] Knowledge graph query and reasoning unit 4 receives structured search statements and obtains candidate results related to the search intent based on the knowledge graph; In this embodiment, the graph query and reasoning unit 4 includes a graph query module and a graph reasoning module; The graph query module is configured to receive a structured search statement output by the search analysis and semantic understanding unit 3, perform a query operation on the knowledge graph constructed by the interface S1.4, and retrieve entity nodes, relationship edges, and attribute information matching the search conditions from the knowledge graph through an entity index, a relationship index, and an attribute index mechanism to generate a preliminary candidate result set, thereby providing basic data support for subsequent reasoning processing. Specifically, the entity, relationship, and attribute conditions are parsed from the structured search statement, and then the entity index, relationship index, and attribute index of the knowledge graph are used to quickly locate the graph nodes and graph edges that meet the conditions. The entity index is used to find graph nodes matching the search keywords or entity IDs, the relationship index is used to match the relationship types or relationship IDs specified in the request, and the attribute index is used to filter nodes or edges that meet the attribute constraints (such as transaction amount and asset status). During the search process, Boolean matching, range query, or fuzzy matching strategies can be combined to generate candidate triples or multi-tuples for entity-relationship-attribute combinations that meet the conditions, and these candidate results are aggregated to form a preliminary candidate result set. The graph reasoning module generates a semantic-enhanced reasoning result set based on the preliminary candidate result set, in combination with the relationship chain and reasoning rules in the knowledge graph. Specifically, starting from the preliminary candidate result set, the relationship chain in the knowledge graph and the predefined reasoning rules (which are used to describe typical logical relationships in the property field, such as if an asset is registered in the name of A and A has an inheritance relationship with B, then it can be inferred that the actual control right of the asset may belong to B; if A mortgages a house to B, and B transfers the house to C, then it can be inferred that A and C have an indirect transfer relationship; if an asset has been frozen, its transfer relationship should be marked as invalid) are used for reasoning. During the reasoning process, the candidate entity nodes and their adjacent relationships are first extended and combined to identify implicit or indirect relationships (such as the ultimate ownership in the transfer chain), and low-confidence matching results are corrected. Then, the extended paths are semantically matched and their relevance is evaluated based on semantic embedding vectors to supplement missing attributes or related entities, thereby achieving semantic information expansion. Finally, the logically reasoned, path-inferred, and semantically enhanced entity, relationship, and attribute combinations are organized into a reasoning result set, each result is accompanied by a confidence score and reasoning basis, and high-quality input is provided for the matching calculation unit to improve the accuracy and intelligence level of the search results.

[0023] The matching calculation unit 5 performs semantic similarity calculation on the candidate results based on cosine similarity to output the optimal search result. In this embodiment, the matching calculation unit 5 performs semantic similarity calculation on the candidate results based on cosine similarity to output the optimal search result, including the following steps: S4.1, for the atlas query and inference unit 4 output candidate results, respectively, extract entity, relationship and attribute multi-dimensional feature representation, including text description features, semantic embedding features, numerical attribute features and context structure features; S4.2, by feature alignment method (linear transformation) to map the semantic representation of user search request and the feature vector of candidate results to a unified representation space, lay the foundation for similarity calculation and sorting; S4.3, use cosine similarity to calculate the semantic similarity score between user search request representation and candidate result representation (input the semantic vector of user search request and the feature vector of candidate result respectively into the linear mapping matrix and , get the mapped vector , , then normalize the mapping vector, so that they are in the same vector space and have the same scale); S4.4, according to the semantic similarity score from high to low, sort the candidate results, select the top b results as the optimal search results output, the optimal search results output content includes entity, relationship, attribute information and corresponding comprehensive score, so as to obtain accurate matching results or enter subsequent interactive confirmation process.

[0024] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above examples, the above examples and descriptions in the specification are only preferred examples of the present application, and are not intended to limit the present application, various changes and improvements can be made without departing from the spirit and scope of the present application, these changes and improvements all fall within the scope of the present application.

Claims

1. A property right intelligent retrieval and matching system based on a knowledge graph, characterized in that, Comprise: Data acquisition and processing unit (1), the data acquisition and processing unit (1) acquires property data, and pre-processes property data; Knowledge modeling and graph construction unit (2), the knowledge modeling and graph construction unit (2) constructs the pre-processed property data into a knowledge graph based on entity extraction, relationship extraction and attribute annotation; Retrieval analysis and semantic understanding unit (3), the retrieval analysis and semantic understanding unit (3) is used for receiving user input retrieval request, and the retrieval request is converted into a structured retrieval sentence matched with the knowledge graph based on the intention path discovery scheme of dynamic multi-hop reasoning, and the retrieval intention of the retrieval request is analyzed through context modeling and semantic disambiguation method; Graph query and reasoning unit (4), the graph query and reasoning unit (4) receives the structured retrieval sentence, and obtains the candidate results related to the retrieval intention based on the knowledge graph; Matching calculation unit (5), the matching calculation unit (5) calculates the semantic similarity of the candidate results based on cosine similarity, and outputs the optimal retrieval result. 2.The knowledge graph-based intellectual property right intelligent retrieval and matching system according to claim 1, characterized in that: The property data at least includes: property transaction records, legal provisions, asset information. 3.The knowledge graph-based intellectual property right intelligent retrieval and matching system according to claim 2, characterized in that: The knowledge modeling and graph construction unit (2) constructs the pre-processed property data into a knowledge graph, comprising the following steps: S1.1, text analysis is performed on the pre-processed property transaction records, legal provisions and asset information, core entities related to property are identified and extracted, at least including property owner, transaction party, contract subject, legal provision number, property subject name, asset category; S1.2, based on the syntactic dependency analysis and deep semantic recognition method, the property relationship is identified from the semantic context between the core entities, at least including transfer relationship, mortgage relationship, constraint relationship, inheritance relationship; S1.3, descriptive attribute information is supplemented for the core entities and property relationships, wherein the attributes corresponding to the property transaction records include transaction time, transaction amount, transaction mode, the attributes corresponding to the legal provisions include provision content, applicable conditions, legal validity, and the attributes corresponding to the asset information include asset location, asset valuation, asset status; S1.4, the core entities are taken as graph nodes, the property relationships are taken as graph edges, and the attribute information is combined to generate a weighted directed graph structure, i.e. a knowledge graph. 4.The knowledge graph-based intellectual property right intelligent retrieval and matching system according to claim 1, characterized in that: The retrieval analysis and semantic understanding unit (3) comprises a retrieval analysis module and a semantic understanding module; Wherein, the retrieval analysis module based on the user input retrieval request, the retrieval request is divided into words, word class is marked and intention is identified, the key words and retrieval conditions contained therein are extracted, and the key words and retrieval conditions are converted into a structured retrieval sentence based on the intention path discovery scheme of dynamic multi-hop reasoning; The semantic understanding module based on the structured retrieval sentence, the user's real intention is identified through context modeling and semantic disambiguation method, and the retrieval request is mapped to the entity, relationship and attribute expression in the knowledge graph. 5.The knowledge graph-based intellectual property right intelligent retrieval and matching system according to claim 4, characterized in that: The retrieval analysis module extracts the key words and retrieval conditions contained therein, and converts the key words and retrieval conditions into a structured retrieval sentence based on the intention path discovery scheme of dynamic multi-hop reasoning, comprising the following steps: S2.1, perform a word segmentation operation on the user input search request to divide the search request into a sequence of words; S2.2, perform part-of-speech tagging on the sequence of words; S2.3, based on the sequence of words and its part-of-speech tagging, generate a sorted list of intent paths using a dynamic multi-hop reasoning intent path discovery scheme; S2.4, convert the intent path list into a structured search statement of the knowledge graph. 6.The knowledge graph-based intellectual property right intelligent retrieval and matching system according to claim 5, characterized in that: In S2.3, the sorted list of intent paths is generated using a dynamic multi-hop reasoning intent path discovery scheme, including the following steps: S2.31, input the sequence of words into the knowledge graph generated in step S1.4 through an entity linker to compare and establish a corresponding link between the identified target words and the graph nodes in the knowledge graph, obtaining a set of anchor entities; S2.32, input the verbs and relationship phrases contained in the search request into a relationship mapping model to generate corresponding semantic embedding vectors; at the same time, encode the set of relationship types defined in the knowledge graph into an embedding vector space; calculate the cosine similarity between the request relationship phrase vector and the relationship type vector, and select the candidate relationship closest to the semantics; S2.33, starting from the anchor entities in step S2.31, combine the candidate relationships obtained in step S2.32 to perform path expansion within the local neighborhood of the knowledge graph with a limited step length, generating a candidate path set; S2.34, based on semantic matching degree, path confidence and jump penalty factor, use a ranking function to score and sort the candidate paths; finally, select the top t paths as the intent analysis result to form a sorted list of intent paths. 7.The knowledge graph-based intellectual property right intelligent retrieval and matching system according to claim 4, characterized in that: The semantic understanding module identifies the user's true intent through context modeling and semantic disambiguation methods, and maps the search request to entities, relationships and attribute expressions in the knowledge graph, including the following steps: S3.1, by performing semantic encoding on the search request and the conversation history of the search request, construct a context semantic vector, and associate the keywords, time expressions and pronouns in the search request with the context entities to generate an enhanced query representation that fuses context information; S3.2, for ambiguous entities, relationships and attribute candidates in the search request, combine the context modeling results and knowledge graph adjacency relationships for multi-factor scoring and filtering; S3.3, fill the filtered entities, relationships and attributes into the predefined semantic slots according to the semantic roles, and construct logical triples to map the search request to graph nodes, graph edges and attribute information in the knowledge graph, form a structured query template, output a sorted list of candidate results, and introduce evidence source credibility modeling and comparative path synthesis to optimize the sorted list of candidate results. 8.The knowledge graph-based intellectual property right intelligent retrieval and matching system according to claim 7, characterized in that: In S3.3, the sorted list of candidate results is optimized by introducing evidence source credibility modeling and comparative path synthesis, including the following steps: S3.31, for each path in the sorted list of candidate results, extract its supporting evidence sources; S3.32, set initial weights for different support evidence sources, and compute a combined confidence score for the candidate path ; S3.33, with the integrated confidence score The candidate path greater than the path threshold a is an anchor point, and a contrast path is generated in the local neighborhood of the knowledge graph. S3.34, compare the anchor path and its comparative path dimension by dimension, and perform difference evaluation; S3.35, the credibility score is introduced into the ranking function together with the difference evaluation result, and the candidate result list is rescored and optimized. 9.The knowledge graph based intellectual property right intelligent retrieval and matching system according to claim 3, characterized in that: The graph query and reasoning unit (4) comprises a graph query module and a graph reasoning module; The graph query module is configured to receive the structured search statement output by the search analysis and semantic understanding unit (3), perform a query operation on the knowledge graph constructed in S1.4, retrieve entity nodes, relationship edges and attribute information matching the search condition from the knowledge graph through an entity index, a relationship index and an attribute index mechanism, and generate a preliminary candidate result set; The graph reasoning module performs logical reasoning, path inference and semantic expansion based on the preliminary candidate result set, in combination with the relationship chain and reasoning rules in the knowledge graph, to generate a semantic-enhanced reasoning result set. 10.The knowledge graph based intellectual property right intelligent retrieval and matching system according to claim 1, characterized in that: The matching calculation unit (5) performs semantic similarity calculation on the candidate results based on cosine similarity, and outputs the optimal search result, comprising the following steps: S4.1, for the candidate results output by the graph query and reasoning unit (4), extract the multi-dimensional feature representation of entities, relationships and attributes; S4.2, map the semantic representation of the user search request and the feature vector of the candidate results to a unified representation space through a feature alignment method; S4.3, calculate the semantic similarity score between the user search request representation and the candidate result representation using cosine similarity; S4.4, sort the candidate results in descending order of semantic similarity score, and select the top b results as the optimal search result.

Images