Dynamic intelligent retrieval closed-loop feedback method based on multi-source aging knowledge base

Abstract

The invention provides a dynamic intelligent retrieval closed-loop feedback method based on a multi-source aging knowledge base, and relates to the technical field of data processing, and the method comprises the steps: carrying out the semantic analysis of a completion service request, and positioning and calling a target aging knowledge base; acquiring the completion service request, performing answer text retrieval and union solution in the target aging knowledge base, and outputting a basic answer text; and performing structured semantic formatting processing on the basic answer text by adopting the standardized answer template to generate a real-time request reply. The technical problem that user consultation response is distorted and delayed due to the fact that an existing campus intelligent service platform is lack of a knowledge base dynamic fresh-keeping mechanism, insufficient in semantic retrieval coverage and lack of closed-loop self-adaptive capacity is solved. Accurate preservation and intelligent retrieval closed loop of the dynamic knowledge base are realized, and the adaptive ability of an intelligent service platform is improved while the response accuracy and timeliness of campus intelligent consultation service are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base. Background Art

[0002] Existing campus smart service platforms generally adopt a static knowledge base architecture, and their data updates rely on manual scheduled maintenance, making it difficult to adapt to high-frequency changes in campus policies, course schedules, and other scenarios.

[0003] At the same time, due to the lack of an automated management mechanism for time-limited tags, expired information (such as previous course selection notices and expired event announcements) has been accumulated in the knowledge base of the campus smart service platform for a long time, resulting in teachers and students' search results being mixed with historical data and the latest policies, and consultation responses are seriously distorted.

[0004] Furthermore, traditional keyword matching models in the search process struggle to capture users' colloquial, multimodal queries. For example, the query "dormitory air conditioning repair report" may not be fully retrieved due to lack of synonyms like "hot water repair" or "logistics report." Furthermore, heterogeneous data from multiple sources (such as academic documents and club notifications) is not containerized and isolated, further amplifying search noise.

[0005] At the same time, user satisfaction feedback and high-frequency questions cannot drive the self-optimization of the software system of the campus smart service platform. Manual intervention is required to screen high-quality questions and answers to update the knowledge base, resulting in significant service response delays and high operation and maintenance costs.

[0006] In summary, the existing campus smart service platform has technical problems such as the lack of a dynamic knowledge base preservation mechanism, insufficient semantic retrieval coverage, and lack of closed-loop adaptive capabilities, which lead to distorted and delayed responses to user inquiries. Summary of the Invention

[0007] This application provides a dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base, which is used to solve the technical problems of the existing campus smart service platform, such as the lack of a dynamic knowledge base preservation mechanism, insufficient semantic retrieval coverage and lack of closed-loop adaptive capabilities, which lead to distorted and delayed user consultation responses.

[0008] In view of the above problems, this application provides a dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base, the method comprising:

[0009] After receiving a real-time service request from a user, NLU technology is used to perform semantic text completion processing on the real-time service request to obtain a completed service request; the completed service request is matched and called with a predefined standardized response template; semantic analysis is performed on the completed service request to locate the knowledge base association vector; the knowledge base association vector is used to traverse the container mapping table of the multi-source time-sensitive knowledge base to locate and call the target time-sensitive knowledge base; keyword expansion is performed on the completed service request to construct an extended search formula group; the target independent container is used to run the extended search formula group, and the answer text retrieval and union solution are performed in the target time-sensitive knowledge base to output a basic answer text; the standardized response template is used to perform structured semantic formatting on the basic answer text to generate a real-time request reply, wherein the real-time request reply carries an information source path identifier.

[0010] In one embodiment, the following processing is also performed:

[0011] Receive the user's satisfaction feedback coefficient for the real-time request reply; use the knowledge base association vector and the basic answer text as dual retrieval keys, trace back the historical access log, and obtain the request retrieval frequency characteristics and the historical satisfaction sequence; time-weight the satisfaction feedback coefficient and the historical satisfaction sequence, and output the reply satisfaction characteristics; based on the preset weight rule, fuse the reply satisfaction characteristics and the retrieval frequency characteristics, and output the temporary storage demand quantification value; if the temporary storage demand quantification value exceeds the preset threshold, store the completion service request and the basic answer text as a formatted question and answer pair in the temporary storage area of ​​the target time-sensitive knowledge base.

[0012] In one embodiment, the following processing is also performed:

[0013] Based on the consultation monitoring cycle, the calling frequency of the formatted question and answer pair is tracked; if the calling frequency of the formatted question and answer pair is set to 0 in Q consecutive consultation monitoring cycles, the formatted question and answer pair is overwritten and deleted in the temporary storage area.

[0014] In one embodiment, keyword expansion is performed on the completion service request to construct an expanded search formula group, and the following processing is further performed:

[0015] The completion service request is segmented to extract an initial keyword set; the initial keyword set is filtered using a preset low-frequency vocabulary to obtain a purified keyword set; the purified keyword set is semantically expanded in multiple dimensions to obtain multiple keyword clusters; the multiple keyword clusters are enumerated in multi-dimensional logical association combinations between clusters to obtain the expanded search formula group.

[0016] In one embodiment, the initial keyword set consists of a core noun subset and a verb phrase subset.

[0017] In one embodiment, the target independent container is used to run the expanded search formula group, the answer text search and union solution are performed in the target time-sensitive knowledge base, the basic answer text is output, and the following processing is performed:

[0018] Independent threads are allocated to the extended search formula group to form a parallel search task group; a multi-threaded concurrent mechanism is enabled in the target independent container to execute the parallel search task group and output a plurality of candidate answer text sets; a union is performed on the plurality of candidate answer text sets to output a plurality of union answer texts; weighted association screening is performed based on a plurality of text timeliness tags and a plurality of call frequency tags of the plurality of union answer texts to output the basic answer text.

[0019] In one embodiment, the following processing is also performed:

[0020] Based on a preset knowledge base update cycle, a crawler is driven to collect multi-source public information to obtain multi-source document resources; semantic analysis is performed on P updated document resources in the multi-source document resources to output P resource association vectors; the P resource association vectors are used to traverse the container mapping table of the multi-source time-sensitive knowledge base to obtain P resource attribution guidelines; and the mapping of the P updated document resources is archived and updated to the multi-source time-sensitive knowledge base based on the P resource attributions.

[0021] In one embodiment, the following processing is also performed:

[0022] After extracting time-sensitive feature keywords from the P updated document resources, P text time-sensitive labels are analyzed and output; after text formatting is performed on the P updated document resources, the P text time-sensitive labels are mapped and bound to the P updated document resources.

[0023] In one embodiment, the following processing is also performed:

[0024] The independent aging rule engine traverses the text aging tag fields of the documents in the target aging knowledge base based on a preset scanning cycle, wherein the independent aging rule engine is associated with the target aging knowledge base; if expired data is identified, metadata formatting and stripping are performed and then moved into the archive storage area; if expired threshold data is identified, overwriting deletion is performed during the archive update process of the target aging knowledge base.

[0025] The technical solution provided in this application has at least the following technical effects or advantages:

[0026] After receiving a user's real-time service request, the real-time service request is semantically completed using NLU technology to obtain a completed service request; the completed service request is matched and called with a predefined standardized response template; the knowledge base association vector is located by performing semantic analysis on the completed service request; the container mapping table of the multi-source time-sensitive knowledge base is traversed using the knowledge base association vector to locate and call the target time-sensitive knowledge base; the completed service request is keyword expanded to construct an extended search formula group; the extended search formula group is run using a target independent container, and the answer text is retrieved and solved in the target time-sensitive knowledge base to output a basic answer text; the basic answer text is structured and formatted using the standardized response template to generate a real-time request reply, wherein the real-time request reply carries an information source path identifier. This achieves the precise preservation of the dynamic knowledge base and the closed loop of intelligent retrieval, significantly improving the response accuracy and timeliness of campus intelligent consulting services while enhancing the technical effect of the adaptive ability of the smart service platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0028] Figure 1 The following is a flow chart of a dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base provided by the present application;

[0029] Figure 2 A schematic diagram of the process of constructing an extended search formula group in the dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base provided by the present application is shown. DETAILED DESCRIPTION

[0030] This application provides a dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base, which is used to solve the technical problems of the existing campus smart service platform, such as the lack of a dynamic knowledge base preservation mechanism, insufficient semantic retrieval coverage and lack of closed-loop adaptive capabilities, which lead to distorted and delayed user consultation responses.

[0031] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings, but it should be understood that the protection scope of the present invention is not limited by the specific embodiments.

[0032] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like to indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0033] Unless expressly stated otherwise, throughout the specification and claims, the term "comprise" or variations such as "include" or "comprising", etc., will be understood to include the stated elements or components but not to exclude other elements or other components.

[0034] Example, a flowchart of a dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base provided by an embodiment of the present invention, see Figure 1 , the method comprising:

[0035] Step A100: After receiving the user's real-time service request, the real-time service request is semantically completed using NLU technology to obtain a completed service request.

[0036] Specifically, in the smart service platform, when teachers and students submit open-ended or semantically ambiguous real-time service requests (such as "How do I apply for a scholarship?") through the interactive interface, the smart service platform uses natural language understanding (NLU) technology to complete the context and eliminate ambiguity of the original request to obtain the completed service request.

[0037] For example, the colloquial expression "How to apply for a scholarship" is completed with the structured semantics "Undergraduate National Scholarship Application Process Guide", thereby accurately capturing user intent and adapting to the retrieval logic of the knowledge base.

[0038] Step A200: Use the completion service request to match and call a predefined standardized response template.

[0039] The smart service platform matches predefined standardized response templates to the requested service type (e.g., academic consultation, life services). For example, for a "scholarship application" request, the platform automatically uses a three-part template framework: "Policy Basis - Processing Steps - Contact Information." This ensures that responses comply with campus regulations and standardizes the response format to improve information delivery efficiency.

[0040] Step A300: Locate the knowledge base association vector by performing semantic analysis on the completion service request.

[0041] Deep semantic analysis of the completed service request generates vectorized coordinates representing the user's intent, which serve as location markers for knowledge base navigation. For example, when a user queries "gymnasium opening hours," semantic analysis maps this query into an association vector pointing to the "campus facilities library," accurately targeting a dedicated container storing the latest venue information and providing target guidance for subsequent time-sensitive searches. This step intelligently connects user needs with dynamic knowledge base resources, ensuring that search paths are directly connected to real-time data sources within the campus.

[0042] Step A400: Use the knowledge base association vector to traverse the container mapping table of the multi-source time-sensitive knowledge base to locate and call the target time-sensitive knowledge base.

[0043] The knowledge base association vector is used as a navigation identifier to match the container mapping relationship table of the multi-source time-sensitive knowledge base, accurately locate and call the target time-sensitive knowledge base, and investigate the exclusive container instance of the target time-sensitive knowledge base.

[0044] For example, when a user queries for "elective course withdrawal time", the platform hits the "course management library" container (target time-sensitive knowledge base) in the container mapping table based on the semantic vector (knowledge base association vector), and simultaneously triggers the time-sensitive engine of the time-sensitive knowledge base to filter expired data, ensuring that the search results always point to the effective policies of the current semester, thereby achieving accurate scheduling and timeliness guarantee of dynamic knowledge base resources.

[0045] Step A500: Expand the keywords of the completion service request and construct an extended search formula group.

[0046] In one implementation, Figure 2 The method A500 of the present invention includes performing keyword expansion on the completion service request and constructing an extended search formula group. The method includes:

[0047] Step A510: Perform word segmentation processing on the completion service request to extract an initial keyword set.

[0048] Step A520: Using a preset low-frequency word library to filter the initial keyword set to obtain a purified keyword set.

[0049] Step A530: Perform multi-dimensional semantic expansion on the purified keyword set to obtain multiple keyword clusters.

[0050] Step A540: Enumerate the multi-dimensional logical association combinations among the multiple keyword clusters to obtain the expanded search formula group.

[0051] In one implementation, the initial keyword set consists of a core noun subset and a verb phrase subset.

[0052] Specifically, this embodiment performs word segmentation analysis on the completed service request, extracts core nouns (such as "scholarship") and verb phrases (such as "application process") to form an initial keyword set, focusing on the essential information elements of the user request.

[0053] Use a preset low-frequency vocabulary (such as a stop word list) to filter out redundant words (such as "how" and "of") in the initial keywords, generate a high-value purified keyword set, and eliminate noise interference.

[0054] Multi-dimensional semantic expansion is performed on multiple purified keywords in the purified keyword set one by one, for example, through synonym replacement ("dormitory" → "apartment"), hyponym association ("gymnasium" → "badminton court"), and finally multiple keyword clusters with semantic associations are generated to cover the diverse expressions of users' potential intentions.

[0055] Dynamically combine multiple keyword clusters according to logical relationships: keywords within a cluster are associated with "OR" (such as "opening hours OR business hours"), and keywords between clusters are connected in series with "AND" (such as "(gymnasium OR swimming pool) AND (opening hours OR business hours)") to build an extended search formula group with high recall rate.

[0056] This embodiment converts user brief queries into multi-dimensional, high-coverage search expressions through a layered semantic expansion and logical combination mechanism, significantly improving the recall rate in complex scenarios while achieving the technical effect of ensuring both search accuracy and fault tolerance.

[0057] Step A600: Use the target independent container to run the extended search formula group, perform answer text retrieval and set solution in the target time-sensitive knowledge base, and output the basic answer text.

[0058] In one implementation, the target independent container is used to run the expanded search formula group, answer text retrieval and union are performed in the target time-sensitive knowledge base, and a basic answer text is output. Step A600 of the method provided by the present invention includes:

[0059] Step A610: Allocate independent threads to the expanded search formula group to form a parallel search task group.

[0060] Step A620: Enable a multi-threaded concurrent mechanism in the target independent container to execute the parallel search task group and output multiple candidate answer text sets.

[0061] Step A630: performing a union solution on the multiple candidate answer text sets and outputting multiple union answer texts.

[0062] Step A640: Perform weighted association screening based on the multiple text timeliness tags and multiple call frequency tags of the multiple union answer texts, and output the basic answer text.

[0063] Specifically, this embodiment distributes the extended search formula group to parallel threads according to logical independence. For example, "gymnasium OR swimming pool" and "opening hours OR business hours" are split into independent search tasks to avoid execution conflicts and form a parallel search task group.

[0064] A multi-threaded concurrent mechanism is enabled in the target independent container to execute the parallel search task group, output a plurality of candidate answer text sets, perform a union solution on the plurality of candidate answer text sets, and output a plurality of union answer texts.

[0065] For example, merge "Gymnasium Opening Hours Description" and "Swimming Pool Business Notice" to form a complete answer pool covering multiple scenarios.

[0066] Based on the multiple text timeliness tags (such as the proximity of the deadline) and multiple call frequency tags (such as historical search popularity) of the multiple union answer texts, weighted association screening is performed to filter out high-timeliness and high-frequency core texts (such as the latest version of the venue guide) to generate the final basic answer text.

[0067] This embodiment uses parallel union retrieval and spatiotemporal weight screening to dynamically output accurate answers with high timeliness and value while ensuring the integrity of the results, achieving the technical effect of significantly improving the response quality of complex queries and user experience.

[0068] Step A700: Use the standardized response template to perform structured semantic formatting processing on the basic answer text to generate a real-time request reply, wherein the real-time request reply carries an information source path identifier.

[0069] This embodiment restructures the retrieved basic answer text according to a predefined template. For example, it integrates scattered scholarship application materials and process instructions into a standardized three-paragraph reply, and automatically marks the information source path (such as "Excerpted from Article 3.2 of the 2024 Funding Guidelines of the Academic Affairs Office") to ensure that the reply content is authoritative, standardized and traceable, thereby improving information credibility and service efficiency.

[0070] This embodiment realizes the precise preservation of dynamic knowledge base and intelligent retrieval closed loop, significantly improving the response accuracy and timeliness of campus intelligent consulting services while enhancing the adaptive capabilities of the smart service platform.

[0071] In one implementation, the method further includes:

[0072] Step A710: Receive the user's satisfaction feedback coefficient for the real-time request response.

[0073] Step A720: Use the knowledge base association vector and the basic answer text as dual search keys, trace back historical access logs, and obtain request search frequency characteristics and historical satisfaction sequences.

[0074] Step A730: Time-sequentially weight the satisfaction feedback coefficient and the historical satisfaction sequence, and output the response satisfaction characteristics.

[0075] Step A740: Based on the preset weight rule, the response satisfaction feature and the retrieval frequency feature are integrated to output a quantitative value of temporary storage demand.

[0076] Step A750: If the temporary storage demand quantification value exceeds a preset threshold, the completion service request and the basic answer text are stored as a formatted question-answer pair in a temporary storage area of ​​the target time-sensitive knowledge base.

[0077] In one implementation, the method further includes:

[0078] Step A760: Based on the consultation monitoring cycle, the calling frequency of the formatted question and answer pair is tracked.

[0079] Step A770: If the calling frequency of the formatted question and answer pair is set to 0 in Q consecutive consultation monitoring cycles, the formatted question and answer pair is overwritten and deleted in the temporary storage area.

[0080] Specifically, the smart management platform receives user satisfaction feedback (such as five-star ratings or emoticons) on the real-time request responses, and quantifies them into numerical coefficients (satisfaction feedback coefficients) to measure service effectiveness.

[0081] The knowledge base association vector and the basic answer text are used as dual search keys, and historical access logs are traced back to obtain the request search frequency characteristics and historical satisfaction sequence for extracting the same question.

[0082] The current satisfaction feedback coefficient is dynamically weighted and fused with the historical satisfaction sequence (e.g., recent feedback has a higher weight), and a comprehensive response satisfaction feature is output to reflect the evolution trend of the quality of question answering.

[0083] According to preset rules (such as satisfaction weight 70% + frequency weight 30%), the comprehensive satisfaction and retrieval frequency characteristics are integrated to generate a temporary storage demand quantitative value. The temporary storage demand quantitative value is used to judge the necessity of sedimentation of high-frequency and high-value questions and answers.

[0084] If the temporary storage demand quantification value exceeds a preset threshold, the completion service request and the basic answer text are stored as a formatted question-answer pair in a temporary storage area of ​​the target time-sensitive knowledge base for priority retrieval by subsequent similar queries.

[0085] Based on the consultation monitoring cycle, the calling frequency of the formatted question and answer pair is tracked to track the changes in its service popularity. If the calling frequency of the formatted question and answer pair is set to 0 for Q consecutive consultation monitoring cycles (such as Q = 3 monitoring weeks), it is determined that its value has declined, and the formatted question and answer pair is overwritten and deleted in the temporary storage area to release storage resources.

[0086] At the same time, building on the satisfaction-driven automatic accumulation mechanism of question-and-answer pairs implemented on the smart management platform, a human-machine collaborative optimization approach is simultaneously integrated: teachers receive student inquiries through online communication mailboxes (such as the principal's mailbox and the academic affairs message board). Using large models, they cluster and analyze the massive amount of questions, screening out high-frequency and common topics (such as "dormitory air conditioning repair time limit" and "cross-disciplinary course selection process"). After manual verification and supplementary answers, these are organized into structured question-and-answer pairs (such as a standardized "question-answer-policy basis" template) and manually entered into the knowledge base's temporary storage area. This process forms a dual-track complement with the system's automatic accumulation mechanism: the automatic mechanism captures immediate and high-frequency needs, while manual entry covers complex policy interpretations, jointly building a dynamically optimized knowledge supply ecosystem.

[0087] This embodiment uses a dynamic storage mechanism driven by the dual dimensions of satisfaction and frequency to automatically deposit high-frequency and high-value question and answer pairs into the knowledge base, and cleans them up in a timely manner after their value decays, thereby achieving adaptive iterative optimization of knowledge resources and continuously improving service accuracy and response efficiency.

[0088] In one implementation, the method further includes:

[0089] Step A400 - Step A: Based on the preset knowledge base update cycle, drive the crawler to collect multi-source public information to obtain multi-source document resources.

[0090] Step A400-b: Perform semantic analysis on the P updated document resources in the multi-source document resources, and output P resource association vectors.

[0091] Step A400-c: Use the P resource association vectors to traverse the container mapping table of the multi-source time-sensitive knowledge base to obtain P resource ownership guidelines.

[0092] Step A400-d: updating the P updated document resource mapping archives to the multi-source time-sensitive knowledge base based on the P resource affiliations.

[0093] In one implementation, the method further includes:

[0094] Step A400-d1: After extracting time-sensitive feature keywords from the P updated document resources, analyze and output P text time-sensitive labels.

[0095] Step A400-d2: After performing text formatting processing on the P update document resources, the P text timeliness label mappings are bound to the P update document resources.

[0096] Specifically, the smart service platform automatically schedules crawlers based on a preset knowledge base update cycle (e.g., daily at midnight) to collect the latest documents (such as academic affairs notices and event announcements) from public channels such as the official website, WeChat official accounts, and campus news accounts. This multi-source document resource pool is generated. This process strictly adheres to privacy protection principles and only crawls publicly available data sources, ensuring the real-time and regulatory compliance of the dynamic knowledge base.

[0097] Perform deep semantic analysis on the P updated document resources contained in the multi-source document resources, extract features such as the title and core content of the text, generate high-dimensional semantic vectors through a vectorization model (such as Sentence-BERT), and output P resource association vectors.

[0098] The P resource association vectors are used to traverse the container mapping table of the multi-source time-sensitive knowledge base (e.g., {"Course Library": Container 1, "Activity Library": Container 2}) to determine the knowledge base type to which the document belongs. For example, if the vector for "Elective Course Supplementary List" hits the container mapping interval of the "Course Management Library," the "Course Library" guidance tag is obtained, ultimately resulting in P resource affiliation guidance.

[0099] By parsing the P updated document resources and identifying timeliness keywords (such as "deadline 2024-09-30" and "valid until the end of this semester"), P text timeliness tags are analyzed and output. For example, "deadline 2024-10-15" is extracted from the updated document resource of "Scholarship Application Guide" to generate a text timeliness tag.

[0100] After text formatting (e.g., segmentation and HTML tag removal) is performed on the P updated document resources, the P text aging tags are mapped and bound to the P updated document resources in the form of metadata, so that the container can automatically manage the document life cycle based on this tag.

[0101] The P updated document resource mappings are archived and updated to the multi-source timeliness knowledge base according to the P resource affiliations, wherein the multi-source timeliness knowledge base stores a plurality of timeliness knowledge bases.

[0102] This embodiment achieves the technical effect of automatically classifying and archiving public data and binding it with time-sensitive tags, ensuring real-time and accurate retrieval and reducing operation and maintenance costs.

[0103] In one implementation, the method further includes:

[0104] Step A810: The independent aging rule engine traverses the text aging tag fields of the documents in the target aging knowledge base based on a preset scanning cycle, wherein the independent aging rule engine is associated with the target aging knowledge base.

[0105] Step A820: If expired data is identified, metadata formatting and stripping are performed and the data is moved into the archive storage area.

[0106] Step A830: If expiration threshold data is identified, overwriting deletion is performed during the target time-sensitive knowledge base archive update process.

[0107] It should be understood that the method for updating data in each knowledge base in the multi-source time-sensitive knowledge base is consistent in this embodiment. This embodiment takes updating data in the target time-sensitive knowledge base as an example to elaborate on the technical solution in detail.

[0108] In addition to being configured with a target independent container, the target aging knowledge base is also configured with an independent aging rule engine to ensure that the document resource management aging strategy is executed independently (such as the course library is retained for one semester and the activity library is retained for one month).

[0109] The independent aging rule engine is run regularly to scan the aging tag field of each document resource stored in the target knowledge base document (for example, at 7:00 p.m. every day). When the independent aging rule engine detects expired data (the aging tag is earlier than the current time), it automatically strips its vector index and other metadata, and moves the original text into the archive area for future reference, freeing up the main library storage resources.

[0110] For data exceeding the expiration threshold (the timeliness label is within the preset expiration window), an overwrite deletion is triggered during the knowledge base maintenance window (such as the off-peak service period) to completely remove the data to avoid redundancy.

[0111] This embodiment uses the automated scanning and policy execution of the aging engine to strip expired data from the index and archive it, and to clear nearly-expiring data at a targeted location, significantly improving retrieval accuracy and reducing storage load, thereby ensuring the technical effect of dynamically preserving the knowledge base.

[0112] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base, characterized by: include: After receiving the user's real-time service request, the real-time service request is semantically completed using NLU technology to obtain a completed service request; Using the completion service request to match and call a predefined standardized response template; locating a knowledge base association vector by performing semantic analysis on the completion service request; Using the knowledge base association vector to traverse the container mapping table of the multi-source time-effect knowledge base, locate and call the target time-effect knowledge base; Perform keyword expansion on the completion service request to construct an extended search formula group; Using a target independent container to run the expanded search formula group, performing answer text retrieval and set solution in the target time-sensitive knowledge base, and outputting a basic answer text; The standardized response template is used to perform structured semantic formatting processing on the basic answer text to generate a real-time request reply, wherein the real-time request reply carries an information source path identifier.

2. The dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base according to claim 1 is characterized in that: Also includes: receiving a user's satisfaction feedback coefficient for a response to the real-time request; Using the knowledge base association vector and the basic answer text as dual search keys, reviewing historical access logs to obtain request search frequency characteristics and historical satisfaction sequences; Time-sequentially weighting the satisfaction feedback coefficient and the historical satisfaction sequence to output a response satisfaction feature; Based on the preset weight rules, the response satisfaction feature and the retrieval frequency feature are integrated to output a quantitative value of the temporary storage demand; If the temporary storage demand quantified value exceeds a preset threshold, the completion service request and the basic answer text are stored as a formatted question-answer pair in a temporary storage area of ​​the target time-sensitive knowledge base.

3. The dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base according to claim 2 is characterized in that: Also includes: Based on the consultation monitoring cycle, the call frequency of the formatted question and answer pair is tracked; If the calling frequency of the formatted question-answer pair is set to 0 in Q consecutive consultation monitoring cycles, the formatted question-answer pair is overwritten and deleted in the temporary storage area.

4. The dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base according to claim 1 is characterized in that: Perform keyword expansion on the completion service request to construct an extended search formula group, including: Performing word segmentation processing on the completion service request to extract an initial keyword set; Using a preset low-frequency word library to filter the initial keyword set to obtain a purified keyword set; Performing multi-dimensional semantic expansion on the purified keyword set to obtain multiple keyword clusters; The plurality of keyword clusters are enumerated based on multi-dimensional logical association combinations among the clusters to obtain the expanded search formula group.

5. The dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base according to claim 4 is characterized in that: The initial keyword set consists of a core noun subset and a verb phrase subset.

6. The dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base according to claim 4 is characterized in that: The target independent container is used to run the expanded search formula group, answer text retrieval and set solution are performed in the target time-sensitive knowledge base, and the basic answer text is output, including: Performing independent thread allocation on the extended search formula group to form a parallel search task group; Enabling a multi-threaded concurrent mechanism in the target independent container to execute the parallel search task group and output a plurality of candidate answer text sets; performing a union solution on the multiple candidate answer text sets and outputting multiple union answer texts; According to the multiple text timeliness tags and the multiple call frequency tags of the multiple union answer texts, weighted association screening is performed to output the basic answer text.

7. The dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base according to claim 1 is characterized in that: Also includes: Based on the preset knowledge base update cycle, the crawler is driven to collect multi-source public information and obtain multi-source document resources; Performing semantic analysis on P updated document resources in the multi-source document resources, and outputting P resource association vectors; Using the P resource association vectors to traverse the container mapping table of the multi-source time-sensitive knowledge base, to obtain P resource ownership guidelines; The P updated document resource mapping archives are updated to the multi-source time-effectiveness knowledge base according to the P resource affiliations.

8. The dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base according to claim 7 is characterized in that: Also includes: After extracting time-sensitive feature keywords from the P updated document resources, analyze and output P text time-sensitive labels; After text formatting is performed on the P update document resources, the P text timeliness labels are mapped and bound to the P update document resources.

9. The dynamic intelligent retrieval closed-loop feedback method based on a multi-source time-sensitive knowledge base according to claim 7, characterized in that: Also includes: The independent aging rule engine traverses the text aging tag fields of the documents in the target aging knowledge base based on a preset scanning period, wherein the independent aging rule engine is associated with the target aging knowledge base; If expired data is identified, metadata formatting and stripping are performed and the data is moved into the archive storage area; If expired threshold data is identified, overwriting deletion is performed during the target time-sensitive knowledge base archive update process.

Images