Intelligent legal service searching and pushing system based on user requirements

Abstract

The invention discloses a legal service intelligent searching and pushing system based on user requirements, particularly relates to the field of information processing, and is used for solving the problems of missing evidence binding and insufficient version anchoring in existing retrieval and touch. Executing structured positioning according to the retrieval intention to generate a fragment anchor identifier, querying a leatherwise table by the fragment anchor identifier to complete version coincidence judgment and paraphrasing closing judgment to generate an effective fragment pointing and rebounding entry, reading a touch node to bind and generate touch content, embedding the touch content into the rebounding entry, recovering a rebounding drop point, checking consistency and correcting a record, and performing rebounding. And a closed evidence chain of retrieval, judgment, reaching and checking is realized.

Description

Technical Field

[0001] This invention relates to the field of information processing, and more specifically, to a system for intelligent search and push of legal services based on user needs. Background Technology

[0002] In the existing legal search service system, after the homepage search bar accepts the input of the cause of action, it guides users to the details of the service package and the selection of merchants, and then to the payment and case progress entry points. During this process, there is a continuous demand for retrieval and outreach. On the progress page, users expect to receive push notifications that match the key points of the cause of action, and to be able to locate still valid legal provisions or document fragments with one click during communication and appeals. At the same time, an auditable chain of evidence is retained for review and tracking. The scenario covers continuous links such as retrieval, comparison, order placement, acceptance, and review, which is consistent with the theme of "intelligent search and push of legal services based on user needs".

[0003] However, current practices lack evidence binding and version anchoring when returning answers and delivering content: answers are often presented as summary text or full-text links, making it impossible to directly jump back to the corresponding clause level; after the evolution of regulations or interpretations, historical responses and subsequent push notifications point to inconsistent text locations; progress-triggered reminders lack version verification and evidence consistency checks, making it difficult for users and the service side to verify the basis at the same time and place, creating a break in the evidence chain and forcing repeated communication and processing. The above problems are not simply semantic matching deviations, but rather stem from the fact that the retrieval, answer generation, and push links do not set fragment-level evidence and temporal validity as rigid constraints, directly affecting the consistency and credibility of on-demand retrieval and on-demand delivery.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a legal service intelligent search and push system based on user needs. This system reads case keywords to form a search intent and writes it into a semantic source table; performs structured positioning in the legal and document database based on the search intent to generate fragment anchor identifiers; uses fragment anchor identifiers to query the evolution table to complete version compatibility and interpretation closure determinations, generating valid fragment pointers and jump-back entry points; reads case progress reach nodes, binds valid fragment pointers to generate reach content, embeds jump-back entry points, retrieves jump-back landing points to verify consistency, and triggers a re-search and re-adjudication to correct pointers and jump-back entry points when inconsistencies are found, writing them into a correction record table. This addresses the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] Semantic solidification module: Reads case keywords to form search intent and writes the case keywords into the semantic source table;

[0008] Fragment Anchoring Module: Based on the search intent, it performs structured positioning in the legal and document database, extracts sentence and segment granularity and generates fragment anchor tags, which are defined by the document identifier, hierarchical path and sentence / segment order as the pointing location;

[0009] Version Adjudication Module: Uses fragment anchor identifiers to query the evolution table to complete version compatibility and interpretation closure judgments. Generates a ruling conclusion pointing to the version first and then the interpretation, which is either adoption, adoption of the previous volume, or backtracking correction. Outputs the corresponding valid fragment pointers and jump-back entries and records the ruling reason.

[0010] The outreach binding module reads the case progress outreach nodes, binds the valid fragments to the metadata of the outreach nodes, generates outreach content and embeds the jumpback entry, while maintaining a consistent mapping of fragment anchors in retrieval, adjudication and outreach.

[0011] Back-to-back verification module: Checks the consistency between the back-to-back landing point and the valid segment pointer. If there is a discrepancy, it triggers a re-search and re-decision with the segment anchor identifier as the only entry point, corrects the pointer and back-to-back entry point, and writes the correction record to the correction record table.

[0012] Furthermore, the semantic solidification module reads case keywords from the search bar, performs word segmentation and invalid word cleanup, completes synonym merging according to the jurisdiction thesaurus and case keyword thesaurus to obtain the search intent, and writes the case keywords and search intent into the semantic source table simultaneously, recording the source time, call entry point and call stage. The word segmentation operation splits continuous strings into independent word units, removes invalid words to refine the content, and maps similar or equivalent words to unified standard terms to form standardized expressions.

[0013] Furthermore, the semantic solidification module reads case keywords from the search bar, listens for submission signals in the search bar using an event-driven mechanism, extracts complete strings as case keywords, performs word segmentation to generate a word list, iterates through and matches against an invalid thesaurus and deletes matching items, queries synonym groups according to the jurisdiction thesaurus and case keyword thesaurus and replaces them with standard representative words, recombines the word list to form a standardized expression, integrates the standardized word set to construct the search intent, inserts the case keywords and search intent as field values ​​into the corresponding columns of the semantic source table, and obtains the current timestamp to insert into the source time column.

[0014] Furthermore, the segment anchoring module performs structured positioning in the legal documents database based on the search intent. It establishes a positioning path according to the hierarchical structure of chapters, clauses, provisions, and sentences in the legal documents database. It drills down the hierarchy in the legal documents database to locate the sentences and generates segment anchors for the positioning results. The segment anchors consist of document identifiers, hierarchical paths, and sentence order, and are used to uniquely point to the text location. A one-to-one mapping is established between the segment anchors and the search intent in the index table. The positioning path, corpus source, and jumpable parameters are recorded and used as a unified entry point for subsequent historical queries, adjudication, and access binding. The hierarchical drilling starts from the chapter level and gradually delves into the sentences and segments to ensure that the extracted sentences and segments directly correspond to the key points of the case.

[0015] Furthermore, the segment anchoring module reads the search intent from the semantic source table, loads the metadata directory of the regulations and documents database, parses the hierarchical tree structure of each document, assigns a unique path identifier to each node, connects the node sequence numbers using dot separators, records the start and end offsets of the path, stores the path in a temporary path cache and associates it with the search intent, drills down the hierarchical level in the regulations and documents database, selects relevant chapters by performing relevance sorting at the chapter level, filters relevant clauses and provisions within the selected chapters, and finally locates the sentence segment. It generates a segment anchor identifier for the located sentence segment by concatenating the document identifier, hierarchical path and sentence segment sequence, establishes a one-to-one mapping in the index table, records the location path, corpus source and bounce parameters, and serves as a unified entry point for historical query and adjudication and access binding.

[0016] Furthermore, the version adjudication module queries the evolution table using fragment anchor identifiers. The evolution table records the effective status, effective time, ineffective time, revision association, and interpretation association fields to obtain a set of version candidates related to the fragment and establish a time sequence mapping. First, it completes the version compatibility judgment, judging based on the effective status of the version in which the fragment belongs and the continuity of the evolution chain. If they match, it proceeds to the next judgment; if they do not match, it generates a backtracking correction pointing to the adjudication conclusion and backtracks to the most recent effective version in time sequence, generating the corresponding valid fragment pointer and jump entry. Then, it completes the interpretation closure judgment, referencing and interpreting links within the fragment's internal parsing bar, constructing a single-layer reference parsing relationship, and checking whether each item lands on a specific sentence segment under the same effective status. When all land, it is recorded as closed; when there is a dangling or cross-version landing, it is recorded as unclosed. It generates pointing to the adjudication conclusion in the order of judging the version first and then the interpretation. If they match and are closed, it outputs adoption; if they match but are not closed, it outputs up-roll adoption and promotes the pointer to the next level fragment and generates the corresponding jump entry. At the same time, it records the conclusion type, adoption level, unclosed reference position, and the reason for backtracking or up-rolling in the adjudication record table.

[0017] Furthermore, the version adjudication module uses fragment anchors to query the evolution table, constructs SQL query statements to extract matching version records, collects field values ​​to form a version candidate set, establishes a time sequence mapping by sorting by effective and ineffective times, checks the effectiveness status and evolution chain continuity for the current version, marks it if they match and enters the interpretation closure judgment, and if they do not match, it traverses backward along the time sequence mapping to the most recent effective version to generate a valid fragment pointer and a jump-back entry. Within the fragment, regular expressions are used to match the references and interpretation links within the segment, constructing a directed graph as a single-layer reference resolution relationship, checking edge by edge whether it falls into the same effective state and recording it as closed or not closed, and generating the pointer adjudication conclusion in order.

[0018] Furthermore, the reach binding module reads reach nodes from the case progress. The metadata of the reach nodes includes node category, reach conditions, display entry point, and relative order. The effective fragment pointer is bound to the metadata of the reach node. Based on the node category, the title, core sentences, and necessary prompts are extracted to form reach content. A jump-back entry point is embedded in the reach content. The fragment anchor identifier, effective fragment pointer, and summary of the adjudication record are written to the reach link to ensure that the same reach node and the same fragment anchor identifier correspond to the same landing point, avoiding the generation of new semantic sources and duplicate evidence. The binding operation creates an associated object, with the node category as the key corresponding to the effective fragment pointer.

[0019] Furthermore, the outreach binding module queries the case progress database to retrieve the progress status of currently active cases, uses the case ID to filter and extract outreach node records, parses metadata fields and assigns them to node categories, outreach conditions, display entry points, and relative order variables, verifies that the outreach conditions are met, selects matching parts from valid fragment pointers to create binding objects, maps metadata fields to pointer associations one by one, uses hash verification to verify the uniqueness of the binding, queries predefined templates based on node categories, parses titles, core sentences, and necessary prompts from valid fragment pointers to fill the templates, forms the outreach content string, generates a bounce-back entry URL to embed the outreach content, and inserts fragment anchor identifiers, valid fragment pointers, and adjudication record summaries into the outreach link to ensure that they correspond to the same landing point.

[0020] Furthermore, the jumpback verification module retrieves the jumpback landing point after the user generates a view or reply action. It checks whether the landing point and the valid fragment pointer are consistent according to the index table. If they are consistent, it writes a pass record and accumulates a reliable landing point sample. If they are inconsistent, it triggers a re-search and re-decision with the fragment anchor identifier as the only entry point. It reuses the paths of the fragment anchoring module and the version adjudication module to generate a new valid fragment pointer and jumpback entry point, and writes the correction reason and the landing points before and after the correction into the correction record table. The correction record table is used as a priori signal in the subsequent adjudication stage. When the same fragment anchor identifier and the same citation pattern appear, the roll-up or backtracking strategy is executed first to reduce the probability of deviation again. At the same time, the reach content and jumpback entry point are updated synchronously on the reach side to maintain consistency of the basis. The retrieved jumpback landing point includes the actual located document identifier, hierarchical path and sentence / segment order.

[0021] The technical effects and advantages of this invention's intelligent search and push system for legal services based on user needs are as follows:

[0022] This invention uses case keywords as a single semantic source to drive retrieval, locating sentences and generating segment anchors. It then determines the unique valid segment pointer and jump-back entry based on version compatibility and semantic closure. When a reach node is triggered, a push notification carrying a jump-back entry directly leads to the evidence location. The consistency between the landing point and the valid pointer is checked, triggering correction. This integrates retrieval, adjudication, reach, jump-back, and verification into a closed chain of evidence and a stable pointing relationship, avoiding version mismatches and ambiguous interpretations. It ensures that the pushed content is always aligned with the handling stage, reducing the cost of back-and-forth communication and repeated retrieval, improving the verifiability of evidence and the effectiveness of reach, and developing adaptive convergence capabilities through continuous feedback, thus enhancing the continuity and credibility of legal services. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the intelligent search and push system for legal services based on user needs, as described in this invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Example 1: Figure 1 This invention presents a user-demand-based intelligent search and push system for legal services, comprising:

[0026] Semantic solidification module: Reads case keywords to form search intent and writes case keywords into the semantic source table.

[0027] Fragment Anchoring Module: Based on the search intent, it performs structured positioning in the legal and document database, extracts sentence and segment granularity and generates fragment anchor tags, which are defined by the document identifier, hierarchical path and sentence / segment order as the pointing location.

[0028] Version Adjudication Module: Uses fragment anchor identifiers to query the evolution table to complete version compatibility and interpretation closure judgments. Generates a ruling conclusion pointing to the version first, followed by the interpretation, in the order of adoption, up-volume adoption, or backtracking correction. Outputs the corresponding valid fragment pointers and jump-back entries and records the ruling reason.

[0029] The outreach binding module reads the case progress outreach nodes, binds the valid fragments to the metadata of the outreach nodes, generates outreach content and embeds the jumpback entry, while maintaining a consistent mapping of fragment anchors in retrieval, adjudication and outreach.

[0030] Back-to-back verification module: Checks the consistency between the back-to-back landing point and the valid segment pointer. If there is a discrepancy, it triggers a re-search and re-adjudication with the segment anchor identifier as the only entry point and corrects the pointer and back-to-back entry point. The correction record is written to the correction record table.

[0031] In existing legal service systems, after users enter keywords related to their case type in the homepage search bar, they are often guided to service package details, merchant selection, payment, and case progress entry points. This process continuously generates search and outreach needs. Simultaneously, on the progress page, users expect to receive push notifications that align with the key points of their case, and during communication and appeals, they can easily locate still valid legal provisions or document fragments, retaining an auditable chain of evidence for review and tracking. This continuous process covers scenarios such as search, comparison, order placement, acceptance, and review, highly consistent with the themes of intelligent search and push notifications for legal services based on user needs. Through this mechanism, users can obtain more accurate legal service responses, improving overall processing efficiency and trust. However, current practices lack evidence binding and version anchoring when returning answers and reaching content. This results in answers often being presented as summary text or full-text links, making it impossible to directly jump back to the corresponding clause level. Furthermore, as regulations or interpretations evolve, historical responses and subsequent push notifications may point to inconsistent text locations. Progress-triggered reminders lack version verification and evidence consistency checks, making it difficult for users and the service provider to verify the evidence at the same time and place, creating a break in the chain of evidence. To address these issues, this invention starts with a semantic solidification module. By reading case keywords to form search intent and writing these keywords into a semantic source table, it ensures the solidification and consistency of semantic evidence, avoiding semantic drift and redundant inferences in subsequent stages, thereby laying a stable foundation for the entire search, adjudication, outreach, and verification process.

[0032] Sub-step 1.1: Read the keywords of the cause of action.

[0033] The original text entered by the user in the search bar serves as the starting point for the entire retrieval process and needs to be accurately captured to prevent initial biases from affecting subsequent links. First, capture the input string from the search bar interface as the initial case keyword. This capture process includes real-time monitoring of input events and extracting the complete string after the input is completed and confirmed to ensure that the captured case keyword fully reflects the user's intention. Use the event-driven mechanism to monitor the submission signal of the search bar. Once the signal is triggered, immediately extract the string from the input field and temporarily store it in the memory buffer for subsequent processing. After processing, summarize that the original case keyword has been obtained, providing a direct input for word segmentation and cleaning to ensure the accurate starting point of the semantic source.

[0034] Sub-step 1.2: Perform word segmentation and cleaning of invalid words.

[0035] Original case keywords often contain continuous strings and potential redundant elements. To ensure refined content to support the formation of accurate intentions, they must first be split and filtered. Based on the captured case keyword, perform word segmentation operations to split the continuous string into independent word units, and then remove invalid words to refine the content. Specifically, it includes using the Jieba word segmentation algorithm to segment the case keyword to generate a list of words. Subsequently, traverse the list of words and compare it with a predefined invalid word library (including stop words such as "of", "is", and general words irrelevant to the law), match and delete the matching items one by one to ensure that the remaining list of words only retains the core elements related to legal cases. After processing, summarize that a refined list of words has been obtained, laying a clean foundation for synonym merging and avoiding noise interference in subsequent intention formation.

[0036] Sub-step 1.3: Complete synonym merging according to the legal domain word list and case word list.

[0037] There may be expression variations or non-standard terms in the refined list of words. To achieve semantic normalization for subsequent system matching, mapping adjustments must be made through a dedicated word list. Using the refined list of words obtained in the previous step, compare it with the legal domain word list (covering standard terms in legal fields such as civil, criminal, and administrative) and the case word list (including specific case standard expressions such as "contract disputes", "tort liability", etc.), and perform synonym merging operations to map similar or equivalent words to unified standard terms. First, load the legal domain word list and case word list as dictionary structures. Then, for each word in the refined list of words, query the synonym group in the dictionary. If there is a match, replace it with the standard representative word of the group; if there is no match, retain the original word. After merging, recombine the list of words to form a standardized expression. After processing, summarize that a standardized set of words has been generated, which serves as the core component of the retrieval intention to ensure semantic consistency and comparability.

[0038] Sub-step 1.4: Obtain the retrieval intention.

[0039] The standardized term set already possesses a basic structure, but to enhance the operability of retrieval, further weighting and organization are needed to form a complete intent representation. The standardized term set is integrated to construct the retrieval intent as the semantic core for subsequent retrievals. The standardized term set is rearranged according to logical order (e.g., subject-verb-object structure); then, the TF-IDF algorithm is applied to calculate the inverse exponent of document frequency for each term (where T is the term frequency, D is the total number of documents, and I is the number of documents containing the term) to highlight keywords; finally, the weighted term set is encapsulated into a retrieval intent object, containing the keyword sequence and weight distribution. After processing, the retrieval intent is summarized as generated, providing a complete semantic representation for writing into the semantic source table, ensuring semantic stability in subsequent stages.

[0040] Sub-step 1.5: Simultaneously write the case keywords and search intent into the semantic source table.

[0041] Search intent and original case keywords must be bound and stored together to solidify the semantic source and add contextual information, thereby preventing drift in subsequent processes. The original case keywords and the search intent formed in the previous step are simultaneously recorded in the semantic source table, along with the source time, call entry point, and call stage. A database insert statement is created, inserting the case keywords as field values ​​into the "Original Keywords" column and the search intent as JSON format into the "Intent Representation" column; simultaneously, the current system timestamp is obtained and inserted into the "Source Time" column, and the call entry point (e.g., "Homepage Search Bar") and call stage (e.g., "Initial Search") are extracted from the system logs and inserted into the corresponding columns; after executing the insert operation, the transaction is confirmed to commit, ensuring data persistence. After processing, the semantic source table is summarized as updated, and subsequent searches, rulings, and outreach are only read from it, avoiding semantic drift and duplicate inferences, thus maintaining the consistency of the basis throughout the entire legal service chain.

[0042] The semantic solidification module establishes a solidified foundation for semantic sources by reading case keywords and forming search intents. The original input is processed through word segmentation, invalid word cleanup, and synonym merging to transform it into standardized search intents. This process ensures the accuracy and standardization of user expressions, avoiding the impact of initial deviations on subsequent steps. Search intents and case keywords are synchronously written into the semantic source table, along with the source time, call entry point, and call stage, providing a unified semantic basis. The storage mechanism of the semantic source table prevents drift and duplicate inferences, supporting the continuity of adjudication, access, and verification. This design addresses dynamic legal service scenarios, ensuring the construction of auditable anchor points from the very beginning of input, improving overall consistency of evidence and service reliability.

[0043] Based on the semantic solidification module, which solidifies the semantic source, the module then moves to the fragment anchoring module. By performing structured positioning in the legal and document database according to the search intent, it extracts sentence and segment granularity and generates fragment anchor identifiers to ensure the uniqueness and traceability of the pointing location. This provides a solid technical anchor point for subsequent version determination and access binding, avoids link breakage caused by ambiguous positioning, and enhances the verifiability and continuity of the overall service.

[0044] Sub-step 2.1: Establish a location path according to the hierarchical structure of the chapters, clauses, terms, and paragraphs of the regulations.

[0045] Regulatory texts are typically presented in a multi-layered nested structure. To achieve precise drill-down positioning, a hierarchical path framework must be pre-constructed to match the semantic requirements of the search intent. The search intent is read from the semantic source table as a guiding principle for positioning. First, for documents in the regulatory and document database, positioning paths are established according to the hierarchical structure of chapters, clauses, provisions, and paragraphs. Specifically, the metadata directory of the regulatory and document database is loaded, and the hierarchical tree structure of each document is parsed, where chapters correspond to first-level nodes, clauses to second-level nodes, provisions to third-level nodes, and paragraphs to fourth-level nodes. Then, a unique path identifier is assigned to each node, for example, using a dot separator to connect node numbers, such as "Chapter 1. Clause 2. Provision 3. Paragraph 4". Simultaneously, the start and end offsets of each path are recorded to support fast indexing. Finally, these paths are stored in a temporary path cache, associated with the search intent, for easy use in subsequent hierarchical drill-down. After processing, the positioning path is established, providing a hierarchical framework for performing structured positioning and ensuring accurate navigation of the regulatory and document database.

[0046] Sub-step 2.2: Drill down to the sentence / segment in the legal and document database according to the search intent.

[0047] The established location path provides a navigation foundation, but to truly match the user's case, it is necessary to perform layer-by-layer filtering and positioning in conjunction with the search intent to achieve precise extraction at the sentence / segment level. Based on the search intent, a hierarchical drill-down operation is performed in the legal and document database, starting from the chapter level and progressively delving into the sentence / segment. Specifically, firstly, using the keyword sequence and weight distribution in the search intent, the BM25 algorithm is executed at the chapter level to rank relevance, selecting the top K relevant chapters (K is dynamically set based on the database size); then, within the selected chapters, the BM25 calculation is repeated at the clause level to filter relevant clauses; next, the same algorithm is applied at the item level within the clauses to further narrow down the scope; finally, within the item, the specific sentence / segment is located, and the final match is confirmed by calculating the cosine similarity between the sentence / segment and the search intent (based on word vector embedding), ensuring that the extracted sentence / segment directly corresponds to the key points of the case. The entire drill-down process is logged in case of backtracking needs. After processing, it is summarized as completed sentence / segment positioning, providing specific text locations for generating fragment anchor tags, ensuring that the fragments extracted from the wide-area database are highly relevant.

[0048] Sub-step 2.3: Generate fragment anchor identifiers for the positioning results.

[0049] Although the segment location result is accurate, a composite identifier must be synthesized to achieve unique pointing and subsequent binding, using a document identifier, hierarchical path, and segment sequence number to jointly define the location. A segment anchor identifier is generated for the segment located in the previous step. This identifier consists of a document identifier (a unique document code, such as a law number), a hierarchical path (the dot-separated path established in the previous step), and a segment sequence number (the segment's number at the end of the path). Specifically: the document identifier of the document to which the located segment belongs is extracted, for example, "Civil Code 2020 Edition"; then, the hierarchical path from the previous step is concatenated, such as "Chapter 1. Clause 2. Article 3"; next, the segment's sequence number within that path is calculated, for example, an integer count starting from 1; finally, the three are concatenated using a separator (such as "|") to form a segment anchor identifier string, for example, "Civil Code 2020 Edition|Chapter 1. Clause 2. Article 3|4", and its uniqueness is verified through a hash check to avoid collisions; this identifier is used to uniquely point to the text location. After processing, the fragment anchor identifiers have been generated, providing a unified entry point for mapping establishment and ensuring the stability and immutability of the pointing positions.

[0050] Sub-step 2.4: Establish a one-to-one mapping between fragment anchor identifiers and search intents within the index table.

[0051] Although the segment anchor identifier has been formed, it must be bound to the search intent to facilitate subsequent historical queries and reach binding, and additional parameters must be recorded to support jumpback. A one-to-one mapping between segment anchor identifiers and search intents is established in the index table, while recording the location path, corpus source, and jumpable parameters. A database insert statement is created, inserting the segment anchor identifier as the primary key into the "Anchor Identifier" column and the search intent as JSON format into the "Intent Mapping" column. Then, the location path from the previous step is recorded in the "Path Details" column, the corpus source (the specific library name of the regulations or document repository) is inserted into the "Source Identifier" column, and the jumpable parameters (including offset and URL template) are inserted into the "Jumpback Configuration" column. After the insert is executed, the unique constraint of the index table is confirmed to be effective, ensuring no duplicate mappings. This mapping serves as a unified entry point for subsequent historical queries, adjudication, and reach binding. In summary, the mapping has been established and fully recorded, ensuring consistent use of segment anchor identifiers throughout the process and improving the continuity of retrieved adjudication.

[0052] The fragment anchoring module establishes a precise mapping from semantics to specific text locations by performing structured localization based on search intent and generating fragment anchor tags. This process transforms the hierarchical structure of regulations and document databases into actionable paths, ensuring the accuracy of sentence-level extraction and achieving unique targeting through composite tags. Structured drill-down operations avoid the limitations of generalized searches, and the fragment anchor tag composition integrates document identifiers, hierarchical paths, and sentence-level order, providing a stable foundation for subsequent evolutionary queries. The mapping mechanism of the index table strengthens the continuity of the link, recording the localization path, corpus source, and bounce-back parameters, facilitating seamless inheritance of adjudication and reach. This design addresses the challenges of the dynamic evolution of legal texts, ensuring the traceability and consistency of the pointing location, and improving the verification efficiency and reliability of the service link.

[0053] Based on the semantic source and fragment anchor identifier established in the semantic solidification module and fragment anchoring module, the module then transitions to the version adjudication module. By querying the evolution table using the fragment anchor identifier, it completes the version compatibility judgment and interpretation closure judgment, generates the ruling conclusion in sequence, and outputs the valid fragment pointer and jump-back entry to ensure the timeliness and completeness of the basis, thereby avoiding version mismatch and interpretation ambiguity, providing reliable output for subsequent reach and binding, and enhancing the adaptability and continuity of the legal service link.

[0054] Sub-step 3.1: Query the evolution table using the fragment anchor identifier.

[0055] The fragment anchor identifier, as the unique entry point generated in the previous step, provides a precise query basis for processing regulatory evolution, avoiding biases caused by generalized searches. The fragment anchor identifier is read from the index table and used as the query key. A matching operation is performed in the evolution table, which pre-stores fields related to the regulatory fragment, including effective status (valid or invalid), effective date (start date timestamp), invalidation date (end date timestamp), revision association (link ID pointing to the revision version), and interpretation association (link ID pointing to the interpretation document). By constructing an SQL query, records in the evolution table are filtered using the fragment anchor identifier to extract all matching version records. Then, the extracted records are traversed, and field values ​​are collected to form a version candidate set. Subsequently, this set is sorted by effective date and invalidation date to establish a time sequence mapping, where each version is associated with a timestamp sequence for subsequent backtracking. Finally, the set is verified to be non-empty; if empty, a query failure log is recorded and the process is terminated. After processing, the version candidate set and time sequence mapping have been obtained, providing an ordered data foundation for version matching determination and ensuring the complete tracing of the evolution chain.

[0056] Sub-step 3.2: Complete version compatibility determination.

[0057] The chronological mapping of the version candidate set lays the foundation for the judgment. However, to confirm the validity of the current fragment, the continuity and consistency of its versions must be evaluated to address the failure issues caused by regulatory revisions. Based on the version candidate set, the effective status and chronological chain continuity are checked first for the current version corresponding to the fragment anchor identifier. The position of the current version is located from the chronological mapping. Then, the effective status is checked. If invalid, the chronological chain continuity is checked by comparing whether the revision association points to consecutive versions (without breakpoints). If they match (effective status and chain continuity), they are marked as matched and enter the interpretation closure judgment. If they do not match, a backtracking correction judgment is generated, traversing backward along the chronological mapping to the most recent effective version (based on the most recent effective version at the time of failure), and generating a corresponding effective fragment pointer (updated fragment anchor identifier variant) and a jump-back entry (URL template containing offset) for this version. The entire judgment process records intermediate state logs to support debugging. After processing, it is summarized that the version matching judgment has been executed. If they do not match, a backtracking correction is output to ensure the timeliness of the effective fragment pointer and avoid using invalid criteria.

[0058] Sub-step 3.3: Complete the interpretation closure judgment.

[0059] After confirming the basic validity through version compatibility, to further verify the completeness of the explanation, it is necessary to parse the internal references of the fragment to ensure that all associations are implemented in the same state, preventing cross-version dangling from affecting credibility. Under the condition of compatibility in the previous step, the internal references and explanatory links of the fragment are parsed to construct a single-level reference resolution relationship. Regular expressions are used to match the reference patterns (such as "see Article X") and explanatory links (embedded association IDs) in the fragment text; then, a single-level reference resolution relationship is constructed as a directed graph, where nodes are sentences and edges are reference links; each edge is checked item by item to see if it is implemented in the same effective state (compare the effective time of the reference target with the current version) to a specific sentence; if all edges are implemented (no dangling or cross-version), it is marked as closed; if there is dangling (no target) or cross-version implementation (inconsistent time), it is marked as not closed; the parsing process is limited to a single level to avoid excessive recursion depth. After processing, it is summarized that the interpretation closure judgment has been completed, providing a closure status marker to lay the foundation for generating the ruling conclusion and ensuring the integrity of the explanation chain.

[0060] Sub-step 3.4: Generate the ruling conclusion and output.

[0061] The interpretation closure status is combined with the version compatibility result to form a final ruling in a specified order, outputting usable pointers and entry points, and recording details for easy tracking. Following the order of judging the version first and then the interpretation, the closure status from the previous step is integrated with the version compatibility judgment to generate a pointer ruling conclusion. Specifically, if the version is compatible and the interpretation is closed, an adoption conclusion is output, using the currently valid fragment pointer and jump-back entry point; if the version is compatible but the interpretation is not closed, an upper-volume adoption conclusion is output, the pointer is promoted to the next higher-level fragment (from a sentence / segment back to the payment or clause, adjusted through hierarchical path), and a corresponding jump-back entry point is generated; for retrospective correction (directly output from the version judgment), its conclusion is retained; then, a record is inserted into the ruling record table, including the conclusion type (adoption, upper-volume adoption, or retrospective correction), adoption level (current or upper-volume level), unclosed reference position (list of dangling links), and the reason for retrospective or upper-volume correction (specific evolutionary chain breakpoint or dangling description); after insertion, the transaction is confirmed to commit, ensuring record persistence for subsequent verification and tracking. After processing, the results are summarized as follows: the ruling conclusion has been generated and the valid fragment pointing to the jump-back entry has been output; the ruling record table has been updated, supporting the consistency of the basis of the reach process and the audit requirements.

[0062] The version adjudication module verifies the validity of versions and interpretations through querying and order determination driven by fragment anchor identifiers, outputting a unique valid fragment pointer and jumpback entry. This process integrates evolution table fields and parsing relationships to ensure the rigor of the adjudication conclusion and avoid version mismatches and dangling issues. The application of time sequence mapping and a single-layer reference graph enhances the traceability of the adjudication, and the storage mechanism of the adjudication record table further improves the adaptive capability of the chain. This design provides a core guarantee for a closed chain of evidence to address the complexity of legal evolution, optimizing the efficiency and reliability of subsequent access and verification.

[0063] This invention, based on the semantic solidification module and the version adjudication module, which solidify semantics, locate fragments, and adjudicate valid points, then moves to the reach binding module. By reading the case progress reach nodes, it binds the valid fragment points to the metadata of the reach nodes, generates reach content, and embeds a jump-back entry, ensuring consistent mapping of fragment anchor identifiers. This achieves accurate reach and direct access to evidence, improves the continuity and effectiveness of legal services in the progress stage, and provides a stable foundation for subsequent verification.

[0064] Sub-step 4.1: Read the reach nodes from the case progress.

[0065] Case progress, as the core of the dynamic processing chain, provides the timing basis for triggering push notifications. It is essential to accurately extract nodes to match the user's current stage. The process involves querying the case progress database to retrieve the progress status of currently active cases and reading the reach nodes. The metadata of each node includes node category (e.g., acceptance or review), reach conditions (trigger thresholds such as progress percentage), display entry point (user interface location such as the notification bar), and relative order (the node's sequence number in the chain). A query statement is constructed, using the case ID as the key to filter progress records; matching reach node records are extracted; then, the metadata fields are parsed and assigned to the node category, reach conditions, display entry point, and relative order variables respectively; the reach conditions are verified (comparing the current progress with the threshold); if not, the process is stopped and logged; finally, the read reach nodes are temporarily cached for later binding. After processing, it is summarized that the reach nodes and metadata have been read, providing a progress context for the binding operation and ensuring synchronization between push notifications and the case processing stage.

[0066] Sub-step 4.2: Bind the valid fragment pointer to the metadata of the reached node.

[0067] The metadata of the reach nodes read in the previous step provides a push framework, but to incorporate legal basis, it must be integrated with the valid fragment pointers adjudicated in the previous step to form a composite structure. Based on the node category, a matching portion is selected from the valid fragment pointers output by the version adjudication module and bound to the metadata of the reach node. Specific implementation methods include: loading valid fragment pointers (including updated fragment anchor identifier variants); then, creating a binding object, using the node category as the key, and associating it with the valid fragment pointer; mapping the reach conditions, display entry points, and relative order to this object one by one, ensuring that each metadata field corresponds to a pointer; using a hash algorithm to verify the uniqueness of the binding to avoid duplication; after binding, storing it in a memory structure as input for generating reach content. After processing, it is concluded that the binding is complete, establishing the association between metadata and valid fragment pointers, ensuring that the basis for generating reach content is correctly embedded.

[0068] Sub-step 4.3: Generate the content to be reached and embed the bounce entry.

[0069] The bound object integrates progress and basis information. To generate user-visible push notifications, key elements must be extracted based on node categories, and interactive entry points must be embedded to support direct access. Based on the node categories bound in the previous step, the title (first sentence summary), core sentence (complete matching text), and necessary prompts (historical warnings) are extracted from the valid fragment pointers to form the reach content, and a jump-back entry is embedded in it. A predefined template is queried according to the node category (e.g., the "Case Update" template corresponding to the acceptance category); then, the title (the first 20 characters are truncated as a summary), core sentence (complete sentence text), and necessary prompts (warning text generated based on the ruling reason) are parsed from the valid fragment pointers; these elements are combined to fill the template to form the reach content string; next, a jump-back entry URL is generated, and a jump-back parameter is concatenated with the valid fragment pointer (e.g., "jump: / / [fragment anchor identifier]?offset=[offset]"); the jump-back entry is embedded as a hyperlink at the end of the reach content; finally, the content length is verified to not exceed the threshold, and if it does, it is truncated and an ellipsis is added. After processing, the results show that the content has been generated and embedded with a bounce-back entry, providing users with a one-click location function to ensure the practicality and interactivity of the push notifications.

[0070] Sub-step 4.4: Write the fragment anchor identifier and the summary of the valid fragment pointer and adjudication record to the reach link.

[0071] Although the reached information has been formed, to maintain consistency across the entire chain, key identifiers and summaries must be recorded to avoid subsequent deviations. The fragment anchor identifier, valid fragment pointer, and summary of the adjudication record are written to the reached information chain database to ensure that the same reached node and the same fragment anchor identifier correspond to the same endpoint. Specifically, an insert statement is created to insert the fragment anchor identifier into the "Anchor Identifier" column and the valid fragment pointer into the "Pointer Details" column; then, a summary (a compressed description of the conclusion type and reason) is extracted from the adjudication record table and inserted into the "Adjudication Summary" column; the associated reached node is used as a foreign key to ensure the correspondence; after the insert is executed, the transaction is confirmed to be committed, and it is checked whether a new semantic source has been generated (if so, it is rolled back); this write serves as a constraint to avoid duplicate evidence. After processing, the reached information chain is summarized as follows: it has been updated, maintaining a consistent mapping of fragment anchor identifiers in retrieval, adjudication, and reached information, thus improving the stability of the evidence chain and auditability.

[0072] The outreach binding module reads outreach nodes and binds valid fragment pointers to generate outreach content with embedded bounce-back entry points, which is then written to the link record, achieving precise binding for progress-triggered push notifications. This process ensures seamless integration of metadata and adjudication output, avoiding issues related to new semantic sources. The outreach link writing mechanism strengthens consistent mapping and supports adaptive feedback. For dynamic case progress, it optimizes the effectiveness of outreach and the verifiability of evidence, improving the overall continuity of the legal service link.

[0073] This invention, based on the semantic solidification module and the reach binding module solidifying semantics, locating fragments, adjudicating effective directions, and generating reach content, then enters the bounce verification module. By recovering the bounce landing point to check consistency, and triggering re-retrieval and re-adjudication when there is inconsistency, it ensures the consistent mapping of fragment anchor identifiers, thereby achieving adaptive closure of the evidence chain, improving the reliability and continuity of legal services in the user interaction process, and providing prior signals for subsequent feedback optimization.

[0074] Sub-step 5.1: Retrieve the bounce landing point.

[0075] User-side viewing or replying actions serve as the feedback starting point. To capture the actual landing point for verification, interaction data must be monitored and extracted. After a user-side viewing or replying action occurs, the jump-back landing point is captured from the log interface. This landing point includes the actual document identifier, hierarchical path, and sentence / segment position. An event listener is deployed to monitor the user's click on the jump-back entry point. Once triggered, the landing point coordinates are extracted, including the document identifier (document unique code), hierarchical path (dot-separated node number), and sentence / segment position (path end number). These coordinates are then combined into a landing point string and temporarily stored in a buffer. Simultaneously, the action timestamp and user ID are recorded to associate the case. If capture fails, a retry is performed and a log warning is issued. After processing, it is concluded that the jump-back landing point has been recovered, providing actual user-side data for consistency verification and ensuring the accuracy of the feedback chain's start.

[0076] Sub-step 5.2: Check whether the landing point and the valid segment point are consistent according to the index table.

[0077] The previously retrieved bounce points provide user-perspective data. To verify the match with the preset pointers, a comparison with the index table is necessary to identify potential discrepancies. The index table is used to query the valid segment pointers corresponding to the segment anchor identifiers, and a string comparison is performed with the retrieved bounce points. This includes using the segment anchor identifier as the key to extract valid segment pointers (including document identifiers, hierarchical paths, and sentence / segment positions) from the index table; then, the points are compared component by component: first, the document identifier is matched; if they do not match, they are marked as inconsistent; if they match, the hierarchical path and sentence / segment position are matched; if all match, the result is recorded in the log table, and the point is added as a reliable point sample (appended to the sample library for training and optimization); if any component does not match, it is marked as inconsistent and enters the correction process. After processing, the consistency check is summarized as follows: if consistent, the sample is saved to support stable verification of the evidence chain; if inconsistent, further processing is prepared.

[0078] Sub-step 5.3: Trigger re-retrieval and re-adjudication using the fragment anchor identifier as the sole entry point.

[0079] Inconsistent flags triggered a correction requirement. To restore accurate pointing, the core operations must be re-executed using the previous steps. Using the fragment anchor identifier as the sole entry point, a re-retrieval and re-adjudication are triggered, reusing the structured positioning of the fragment anchoring module and the version compatibility and interpretation closure judgments of the version adjudication module. This includes loading the retrieval intent and positioning path associated with the fragment anchor identifier from the index table; then, reusing the hierarchical drill-down of the fragment anchoring module to relocate the sentence / segment in the regulations and document database and generate temporary fragment anchor identifiers; next, reusing the query history table of the version adjudication module to complete the version compatibility judgment (checking the effective status and the continuity of the history chain) and the interpretation closure judgment (constructing a single-level reference resolution relationship by parsing references within the parsing clause); based on the judgment, a new pointing adjudication conclusion is generated, and the updated valid fragment pointing and jump-back entry are output; the entire reuse process is logged to track iterations. After processing, it is concluded that a re-retrieval and re-adjudication have been triggered, generating new valid fragment pointing and jump-back entry, ensuring that the correction is based on the latest regulatory status.

[0080] Sub-step 5.4: Correct the pointer and the jumpback entry.

[0081] The newly generated valid fragment pointers and jump-back entries from the previous step serve as the basis for correction. To apply this to the link, relevant records must be updated to maintain consistency. The new valid fragment pointers and jump-back entries are obtained from the reuse path in the version adjudication module, replacing the original inconsistencies. The new valid fragment pointers (updated document identifiers, hierarchical paths, and sentence / segment order combinations) are extracted; then, the pointer field of the corresponding fragment anchor identifier is updated in the index table; simultaneously, a new jump-back entry URL is generated (concatenated using jumpable parameters); the original pointers and entries are synchronously replaced in the reach link to ensure that the same reach node corresponds to the new landing point; the consistency after the update is verified by confirming no residual deviation through string comparison; if the update fails, it is rolled back and retried. After processing, the pointers and jump-back entries are summarized as corrected, supporting real-time synchronization on the user side and avoiding continuous deviations affecting service continuity.

[0082] Sub-step 5.5: Write the correction record to the correction record table.

[0083] Although the correction operation has been executed, details must be persisted to optimize future decisions in order to provide prior signals and tracking. The correction reason (description of inconsistent components) and the points of contact before and after correction (comparison of old and new pointers) are written to the correction record table, which will be used as a prior signal in subsequent decision-making stages. An insert statement is created to insert the correction reason into the "Reason Description" column and the points of contact before and after correction as a JSON pair into the "Point of Contact Comparison" column; then, the fragment anchor identifier and reference pattern (the pattern extracted from the interpretation closure judgment) are correlated; after the insert is executed, the commit is confirmed; when the same fragment anchor identifier and the same reference pattern appear, the table record will preferentially trigger the roll-up or backtracking strategy (applied through query matching); simultaneously, an update notification is pushed to the reach side to synchronize the reach content and the backtracking entry. After processing, it is concluded that the correction record table has been updated, serving as a prior signal to reduce the probability of further deviation and improve the adaptive convergence capability of the evidence chain.

[0084] The jumpback verification module retrieves jumpback points and checks for consistency. If inconsistencies are found, it triggers a re-retrieval and re-adjudication, correcting the errors and writing the correction record to the correction log table, thus achieving a closed-loop feedback system. This process reuses previous steps, ensuring the unified entry point function of the segment anchor identifier. The prior mechanism of the correction log table enhances adaptability, optimizes consistency of basis for user interaction scenarios, and improves the reliability and effectiveness of the legal service chain.

[0085] It should be noted that the system of the present invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting a variety of hardware environments and usage requirements.

[0086] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0087] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely to distinguish one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0088] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A legal service intelligent search and push system based on user needs, characterized in that, Comprising: Semantic solidification module: read the case keywords to form a search intent, write the case keywords to the semantic source table; Fragment anchor module: according to the search intent, perform structured positioning in the regulation and document library, extract to sentence segment granularity and generate fragment anchor identification, and jointly limit the pointing position with document identification, hierarchical path and sentence segment sequence position; Version decision module: query the evolution table with fragment anchor identification to complete version compatibility judgment and interpretation closure judgment, generate pointing decision conclusion as adoption or roll-up adoption or backtracking correction in the order of judging version first and interpretation second, and output corresponding effective fragment pointing and back jump entry and record the decision reason; Touch binding module: read the case progress touch node, bind the effective fragment pointing and the metadata of the touch node, generate touch content and embed the back jump entry, and keep the consistent mapping of fragment anchor identification in search, decision and touch; Back jump checking module: recycle back jump landing point to check the consistency with effective fragment pointing, if inconsistent, trigger re-search and re-decision with fragment anchor identification as the only entry, and modify the pointing and back jump entry, and write the modification record to the modification record table.

2. The intelligent search and push system for legal services based on user needs according to claim 1, characterized in that: The semantic solidification module reads the case keywords from the search bar, performs word segmentation and invalid word cleaning, merges synonyms according to the legal domain word table and the case word table to obtain the search intent, writes the case keywords and the search intent into the semantic source table synchronously, records the source time and the calling entry and calling link, and splits continuous character strings into independent word units through word segmentation operation, removes invalid words to refine the content, and maps similar or equivalent words to unified standard terms through synonym merging to form standardized expression.

3. The intelligent search and push system for legal services based on user needs according to claim 2, characterized in that: The semantic solidification module reads the case keywords from the search bar, uses the event-driven mechanism to listen to the submission signal of the search bar, extracts the complete character string as the case keywords, generates a word list through word segmentation operation, traverses and deletes the matched items by matching the invalid word library, replaces the standard representative word according to the legal domain word table and the case word table, recombines the word list to form the standardized expression, integrates the standardized word set to build the search intent, inserts the case keywords and the search intent into the semantic source table as field values, and inserts the current timestamp into the source time column.

4. The intelligent search and push system for legal services based on user needs according to claim 3, characterized in that: The fragment anchoring module performs structured positioning in the regulation and document library according to the search intent, establishes a positioning path according to the hierarchical structure of chapters, clauses, items and segments of the regulation, drills down the hierarchical structure of the regulation and document library and positions to the segment, generates a fragment anchor identifier for the positioning result, the fragment anchor identifier is composed of a document identifier, a hierarchical path and a segment sequence, is used for uniquely pointing to a text location, and establishes a one-to-one mapping between the fragment anchor identifier and the search intent in an index table, records a positioning path, a corpus source and a back jump parameter, and is used as a unified entrance for subsequent evolution query, ruling and touch binding. The hierarchical drilling starts from the chapter level and gradually deepens to the segment, so as to ensure that the extracted segment directly corresponds to the case gist.

5. The user demand-based legal service intelligent search and push system according to claim 4, characterized in that: The fragment anchoring module reads the search intent from the semantic source table, loads the metadata directory of the regulation and document library, parses the hierarchical tree structure of each document, assigns a unique path identifier to each node, connects the node sequence numbers using a point separator, records the start and end offsets of the path, stores the path in a temporary path cache in association with the search intent, drills down the hierarchical structure of the regulation and document library, performs relevance sorting from the chapter level to select relevant chapters, filters relevant clauses and items in the selected chapters, and finally positions to the segment, generates a fragment anchor identifier for the positioned segment, splices the document identifier, the hierarchical path and the segment sequence to form the fragment anchor identifier, establishes a one-to-one mapping in the index table, records the positioning path, the corpus source and the back jump parameter, and serves as a unified entrance for evolution query, ruling and touch binding.

6. The user demand-based legal service intelligent search and push system according to claim 5, characterized in that: The version ruling module queries the evolution table with the fragment anchor identifier, the evolution table records the fields of effective state and effective time point, invalid time point, revision association and explanation association, obtains a version candidate set related to the fragment and establishes a time sequence mapping, first completes version consistency determination, judges according to the effective state of the version where the fragment is located and the continuity of the evolution chain, enters the next determination when consistent, generates a pointing ruling conclusion for backtracking modification when inconsistent, and backtracks to the latest effective version and generates a corresponding effective fragment pointing and back jump entrance, then completes the interpretation closure determination, parses the intra-clause references and explanation links within the fragment, constructs a single-layer reference parsing relationship, checks item by item whether it lands on a specific segment under the same effective state, records as closed when all landings, records as not closed when there are suspended or cross-version landings, generates a pointing ruling conclusion in the order of version determination first and interpretation determination second, outputs the consistent and closed one, outputs the consistent but not closed one after uprolling and promotes the pointing to the upper level fragment and generates a corresponding back jump entrance, and records the conclusion type, the adoption level, the non-closed reference position, the backtracking or uprolling reason in the ruling record table.

7. The user demand-based legal service intelligent search and push system according to claim 2, characterized in that: The version decision module queries the evolution table with the segment anchor identifier, constructs a SQL query statement to extract matching version records, collects field values to form a version candidate set, sorts the effective time point and the invalid time point to establish a time sequence mapping, checks the effective state and evolution chain continuity for the current version, marks and enters the interpretation closure determination when they are consistent, traverses the time sequence mapping in reverse to the nearest valid version to generate a valid segment pointing and back jump entry when they are inconsistent, uses regular expression matching in the segment to match intra-row references and interpretation links, constructs a directed graph as a single-layer reference resolution relationship, checks whether they are in the same effective state and land to a sentence segment, and marks them as closed or not closed, and generates a pointing decision conclusion in order.

8. The user demand-based legal service intelligent search and push system according to claim 3, characterized in that: The reach binding module reads the reach node from the case progress, the metadata of the reach node includes the node category, the reach condition, the display entry, and the relative order, binds the effective segment pointing and the metadata of the reach node, extracts the title, the core sentence segment, and the necessary prompt according to the node category, forms the reach content and embeds the back jump entry in the reach content, writes the segment anchor identifier, the effective segment pointing, and the summary of the decision record to the reach link, ensures that the same reach node and the same segment anchor identifier correspond to the same landing point, avoids generating new semantic sources and repeated basis, binds the operation to create an associated object, and takes the node category as the key and the effective segment pointing as the corresponding.

9. The user demand-based legal service intelligent search and push system according to claim 4, characterized in that: The reach binding module queries the progress state of the current active case from the case progress database, filters and extracts the reach node record using the case ID, parses the metadata field and assigns it to the node category, the reach condition, the display entry, and the relative order variable, verifies the reach condition, creates a binding object by selecting the matching part from the effective segment pointing, maps the metadata field and the pointing one by one, verifies the uniqueness of the binding using hashing, queries the predefined template according to the node category, fills in the template by parsing the title, the core sentence segment, and the necessary prompt from the effective segment pointing, forms the reach content string, generates the back jump entry URL and embeds it in the reach content, inserts the segment anchor identifier, the effective segment pointing, and the summary of the decision record into the reach link, and ensures that they correspond to the same landing point.

10. The user demand-based legal service intelligent search and push system according to claim 5, characterized in that: The back jump checking module recovers the back jump landing point after the user side generates a view or reply action, checks whether the landing point and the valid segment pointer are consistent according to the index table, writes the passing record when consistent, and deposits the reliable landing point sample, triggers re-retrieval and re-decision when inconsistent, reuses the segment anchor module and the version decision module to generate a new valid segment pointer and back jump entry, and writes the correction reason and the landing point before and after correction into the correction record table, which is used as a priori signal in the subsequent decision stage, and the uprolling or backtracking strategy is preferentially executed when the same segment anchor identifier and the same reference mode appear, so as to reduce the probability of deviation again, and the reach side is synchronously updated to maintain consistency according to the reach content and the back jump entry, and the recovered back jump landing point includes the actual positioning document identifier, hierarchical path and sentence segment sequence position.

Images