A knowledge retrieval system architecture method and platform

Abstract

The present invention relates to the field of knowledge retrieval technology, and discloses a knowledge retrieval system architecture method and platform. The method first obtains equipment manuals to construct an explicit knowledge layer, and obtains implicit knowledge to construct a latent knowledge layer; then cross-layer association is performed through three progressive processing layers: bottom-level term alignment mapping, middle-level association matrix construction, and high-level personalized weight dynamic adjustment; finally, a fusion retrieval mechanism is established based on cross-layer association; the present invention solves the problems of separate storage of explicit and implicit knowledge and difficulty in accurate matching of implicit knowledge in traditional retrieval platforms, realizes layered fusion retrieval of the two, allows technical personnel to simultaneously obtain standardized operations and practical experience, and improves retrieval efficiency.

Description

Technical Field

[0001] The present invention relates to the field of knowledge retrieval technology, and more specifically, to a knowledge retrieval system architecture method and platform. Background Art

[0002] With the continuous development and technological innovation of the manufacturing industry, technicians face problems such as difficulty in acquiring knowledge and low retrieval efficiency when performing work order tasks such as equipment maintenance and process debugging.

[0003] The Chinese patent with the authorization announcement number CN112307772B proposes a method for constructing a knowledge base of Cantonese porcelain based on semantic ontology. By constructing a knowledge model and a retrieval model through semantic analysis of Cantonese porcelain patterns, the practicality and usability of the Cantonese porcelain knowledge base are improved. However, this method focuses on the field of Cantonese porcelain. The Cantonese porcelain knowledge base construction method cannot effectively split the three-level structure and assign labels to the manufacturing equipment knowledge. It is difficult to convert the knowledge in the equipment manual into structured retrieval units and cannot quickly locate the required knowledge. When processing work orders, this method cannot solve the problem of integrated retrieval of explicit knowledge and implicit knowledge. It is difficult for technical personnel to use it to obtain relevant knowledge on equipment maintenance and process debugging, and the retrieval efficiency is low.

[0004] The Chinese patent with authorization announcement number CN118797012B discloses a visual question-answering method, system, and device for crop diseases that introduces external knowledge. It uses image information to participate in knowledge retrieval and fuses different modal features to generate answers, achieving good results in the field of crop diseases. However, this technology has shortcomings in the knowledge acquisition scenario of the manufacturing industry. Manufacturing technicians are mainly faced with equipment operation and maintenance tasks, not image recognition and question answering. Moreover, the implicit knowledge of the manufacturing industry mostly exists in the form of unstructured work order descriptions and practical experience, which is different from the implicit knowledge form in the field of crop diseases. This existing technology is unable to perform block parsing and standardized mapping of work order texts, making it difficult to explore the connection between implicit knowledge and explicit knowledge, and unable to achieve hierarchical fusion retrieval of the two. It cannot meet the needs of technicians for rapid retrieval and effective application of knowledge when processing work orders.

[0005] The Chinese patent application with publication number CN117112760A proposes a large model of intelligent education based on a knowledge base. Through the collaborative work of multiple modules, it provides intelligent reasoning and responses based on a self-developed knowledge base, which to a certain extent solves the security and privacy issues in intelligent education. However, this model has limitations in manufacturing knowledge retrieval. Its focus is on intelligent dialogue and security and privacy, and it is not designed for the characteristics of manufacturing equipment knowledge. When processing equipment manual knowledge and work order experience knowledge, it is impossible to effectively split and index the equipment knowledge, nor can it perform in-depth analysis and standardization of the work order text. It is difficult to meet the needs of hierarchical fusion retrieval of explicit knowledge and implicit knowledge when processing work orders.

[0006] Existing technologies are difficult to solve the problem of hierarchical fusion retrieval of explicit knowledge and implicit knowledge. Traditional knowledge retrieval systems store the two types of knowledge separately, making retrieval operations cumbersome and inefficient. Summary of the Invention

[0007] In order to overcome the above-mentioned shortcomings of the prior art, the present invention provides a knowledge retrieval system architecture method and platform, which constructs an explicit knowledge layer and an implicit knowledge layer, breaking the storage barrier between the two types of knowledge in traditional systems. Through three progressive processing layers: bottom-level term alignment mapping, middle-level association matrix construction, and high-level personalized weight dynamic adjustment, cross-layer association between explicit knowledge and implicit knowledge is achieved. The fusion retrieval mechanism established on this basis can accurately recommend knowledge based on the technician's retrieval request and contextual information. The present invention realizes the hierarchical fusion retrieval of implicit experience and explicit knowledge, which helps to improve retrieval efficiency.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A knowledge retrieval system architecture method, comprising:

[0010] Obtain the equipment manual and use it as the data source to split the equipment knowledge into knowledge units with hierarchical labels. Create a fault code triple index and store the knowledge units with hierarchical labels and the fault code triple index in the explicit knowledge base, using the explicit knowledge base as the explicit knowledge layer.

[0011] Acquire tacit knowledge and build a tacit knowledge layer based on the tacit knowledge;

[0012] Three progressive processing layers are used to cross-link the explicit knowledge layer with the implicit knowledge layer;

[0013] Based on the cross-layer association between explicit knowledge layer and implicit knowledge layer, a context-sensitive fusion retrieval mechanism is established for real-time fusion recommendation of explicit knowledge and implicit experience.

[0014] Furthermore, the method of splitting the device knowledge to obtain knowledge units with hierarchical labels includes:

[0015] Using the equipment manual as the data source, the equipment knowledge is split into three levels to obtain three knowledge units. Each knowledge unit is assigned a unique hierarchical label to obtain knowledge units with hierarchical labels.

[0016] Furthermore, the first level of the three-level structure is the device type, the second level is the system module, and the third level is the knowledge category.

[0017] Furthermore, the tacit knowledge includes at least work order text information;

[0018] The method for constructing the implicit knowledge layer includes: parsing the work order text in blocks to obtain structured semantic elements; performing standardized mapping of the structured semantic elements into colloquial expressions to obtain standardized semantic elements; generating a structured semantic description of the work order experience based on the standardized semantic elements, and storing the structured semantic description of the work order experience in an implicit experience library.

[0019] Furthermore, the block parsing method includes dividing the work order text into multiple information blocks and extracting the core semantics of each information block.

[0020] Furthermore, the three progressive processing layers are bottom layer, middle layer and top layer respectively;

[0021] The method of using three progressive processing layers to perform cross-layer association between the explicit knowledge layer and the implicit knowledge layer includes: performing term alignment mapping at the bottom layer; constructing an association matrix at the middle layer with explicit knowledge units as rows and implicit experience features as columns; the explicit knowledge units refer to knowledge units with hierarchical labels, and the implicit experience features refer to standardized semantic elements; and performing dynamic adjustment of personalized weights at the high layer.

[0022] Furthermore, the method for dynamically adjusting the personalized weight at the high level includes: recording an operation trace log in real time, and obtaining the personalized weight based on the operation trace log.

[0023] Furthermore, the operation trace log includes explicit knowledge access features, implicit knowledge application features and operation step sequences;

[0024] Calculate the basic interest weight based on the explicit knowledge access characteristics and implicit knowledge application characteristics;

[0025] Based on the sequence of operation steps, error correction weights are obtained;

[0026] The personalized weight is obtained based on the basic interest weight and the error correction weight.

[0027] Furthermore, the method for obtaining the error correction weight based on the sequence of operation steps includes:

[0028] Mark the key node operation results in the operation step sequence, the operation results are success or failure, and the key node with failure result is defined as the failure point; the failure includes explicit failure and implicit failure;

[0029] The number of explicit errors is counted and defined as the number of explicit errors; the number of implicit errors is counted and defined as the number of implicit errors. The error correction weight is calculated based on the number of explicit errors and the number of implicit errors.

[0030] A knowledge retrieval system architecture platform, which is used to implement the above-mentioned knowledge retrieval system architecture method, the system includes:

[0031] The explicit knowledge layer construction module is used to obtain equipment manuals, use the equipment manuals as the data source, split the equipment knowledge into knowledge units with hierarchical labels, establish a fault code triple index, and store the knowledge units with hierarchical labels and the fault code triple index in the explicit knowledge base, which serves as the explicit knowledge layer.

[0032] Tacit knowledge layer construction module: used to obtain tacit knowledge and build the tacit knowledge layer based on the tacit knowledge;

[0033] Knowledge layer association module: uses three progressive processing layers to perform cross-layer association between explicit knowledge layer and implicit knowledge layer;

[0034] Fusion retrieval module: Based on the cross-layer association between the explicit knowledge layer and the implicit knowledge layer, a context-sensitive fusion retrieval mechanism is established for real-time fusion recommendation of explicit knowledge and implicit experience.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] The present invention optimizes the knowledge retrieval mode of the knowledge retrieval system for manufacturing technology applications as a whole. By constructing explicit knowledge layers and implicit knowledge layers, the storage barriers between explicit knowledge and implicit knowledge in traditional knowledge retrieval systems are broken, and structured texts such as equipment manuals are integrated with the knowledge in unstructured work order descriptions. Three progressive processing layers are used for cross-layer association, so that different types of knowledge can be interconnected. The fusion retrieval mechanism established on this basis can accurately recommend knowledge based on the technicians' retrieval requests and contextual information, allowing technicians to obtain standardized operating guidelines and actual operating experience at the same time when processing work orders. It effectively solves the problem that traditional knowledge retrieval systems can only provide surface knowledge but cannot associate deep principles and practical experience, improves the retrieval efficiency and maintenance efficiency of technicians, reduces the work order processing time, and improves the one-time repair rate of faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] 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.

[0038] Figure 1 This is a principle flow chart of a knowledge retrieval system architecture method of the present invention;

[0039] Figure 2 This is a flow chart of the method for splitting device knowledge according to the present invention;

[0040] Figure 3 This is a flow chart of a method for obtaining personalized weights based on operation trace logs according to the present invention;

[0041] Figure 4 Schematic diagram of the integration and association of the explicit knowledge layer and the implicit knowledge layer of the present invention;

[0042] Figure 5 A flow chart of the method for establishing a fusion search mechanism for the present invention;

[0043] Figure 6 This is a functional module diagram of a knowledge retrieval system architecture platform of the present invention. DETAILED DESCRIPTION

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0045] Example 1

[0046] See also Figure 1 As shown, this embodiment provides a knowledge retrieval system architecture method, including:

[0047] Step S1000: Obtain the equipment manual, use the equipment manual as a data source, split the equipment knowledge into knowledge units with hierarchical labels, establish a fault code triple index, store the knowledge units with hierarchical labels and the fault code triple index into an explicit knowledge base, and use the explicit knowledge base as the explicit knowledge layer.

[0048] Furthermore, step S1000 includes:

[0049] Step S1100, splitting device knowledge;

[0050] Further, if Figure 2 As shown, step S1100 includes:

[0051] Step S1110: Obtain the device manual. Using the device manual as a data source, perform a three-level structure on the device knowledge to obtain three knowledge units. The first level of the three-level structure is the device type, the second level is the system module, and the third level is the knowledge category. The three knowledge units are a device type knowledge unit, a module knowledge unit, and a knowledge category knowledge unit.

[0052] Step S1120 , assigning a unique hierarchical label to each knowledge unit obtained by decomposing the device knowledge, thereby obtaining a knowledge unit with a hierarchical label; the format of the hierarchical label is [D: device type code - M: module code - K: knowledge category code];

[0053] Step S1130: storing the knowledge units with hierarchical labels into the explicit knowledge base as a structured retrieval entry.

[0054] Specifically, on traditional knowledge retrieval platforms, equipment manuals typically exist in the form of unstructured text, with complex content and a lack of effective organization and classification. Retrieving the required knowledge requires technicians to spend a considerable amount of time searching and filtering, making it difficult to quickly locate knowledge relevant to specific work order operations, resulting in low maintenance efficiency. Equipment manuals are standardized technical documents that contain a wide range of specialized knowledge about the equipment. The first level of the three-level structure is the equipment type, such as common CNC machine tools and injection molding machines. This is used to distinguish different types of equipment. Because the knowledge systems of different types of equipment vary significantly, using this as a first-level classification helps quickly identify the equipment category to which knowledge belongs. The second level is the system module, such as the feed system of a CNC machine tool or the hydraulic system of an injection molding machine. System modules are components of the equipment, each with its own unique functions and related knowledge. Categorizing these modules further refines the knowledge structure. The third level is the knowledge category, such as principle descriptions, operating procedures, and fault codes. This level clearly defines the specific knowledge type, making it easier for technicians to access specific knowledge. Taking the "CNC Machine Tool Operation Manual" as an example, the content is divided into "CNC Machine Tool - Feed System - Operation Steps - Screw Lubrication Process". Through this three-level split, the originally unstructured and complicated manual content is transformed into clear, hierarchical knowledge units, which is convenient for subsequent processing and retrieval.

[0055] Each knowledge unit derived from device knowledge is assigned a unique hierarchical label in the format [D:Device Type Code - M:Module Code - K:Knowledge Category Code] to facilitate accurate retrieval. For example, the knowledge unit "CNC Machine Tool - Feed System - Operation Procedure - Screw Lubrication Procedure" is assigned the label [D:001-M:012-K:003], where D represents the device type code, with 001 corresponding to the CNC machine tool; M represents the module code, with 012 corresponding to the feed system; and K represents the knowledge category code, with 003 corresponding to the knowledge related to the screw lubrication procedure within the operation procedure. This encoding method transforms complex knowledge content into a label format that is easily recognized and processed by machines, improving retrieval accuracy and efficiency.

[0056] Knowledge units with hierarchical labels are stored in the explicit knowledge base as a structured search entry. The explicit knowledge base is where these structured knowledge units are stored, providing basic support for retrieval. For example, when a technician needs to find all the knowledge about a CNC machine tool feed system, they can search by "D:001-M:012". This will quickly locate and return the relevant knowledge units from the explicit knowledge base, significantly reducing search time and improving knowledge acquisition efficiency.

[0057] Step S1100, through three-level structure splitting and label assignment, addresses the issue of equipment manuals' disorganized knowledge structure and difficulty in rapid retrieval. Traditional equipment manuals often consist of long paragraphs of text, making it time-consuming and laborious for technicians to find specific knowledge. However, this three-level splitting breaks down knowledge into units at different levels, each with its own label, making knowledge retrieval more structured and organized. For example, when a technician needs to find the operating procedures for an injection molding machine's hydraulic system, they can use the device type code (D) to find the corresponding code for the injection molding machine, then use the module code (M) to find the hydraulic system code, and finally use the knowledge category code (K) to find the operating procedure code. This allows them to quickly locate the required knowledge, saving significant search time, improving knowledge acquisition efficiency, and reducing the time cost of acquiring standardized knowledge. Step S1100 lays a solid foundation for the subsequent construction of the explicit knowledge layer and is a key step in the structured representation of explicit knowledge. It enables the knowledge in the equipment manual to be stored in a structured form within the explicit knowledge base, working in conjunction with subsequent steps such as the fault code triple index and cross-layer association with the implicit knowledge layer. The structured knowledge units provided in step S1100 provide a positioning basis for the fault code triple index, enabling fault codes to be quickly associated with corresponding knowledge units. Hierarchical labeling facilitates term alignment mapping between explicit and implicit knowledge layers, helping to uncover implicit connections between explicit knowledge and implicit experience.

[0058] Step S1200, establishing a fault code triple index;

[0059] Furthermore, step S1200 includes:

[0060] Step S1210: extract the fault code list from the equipment manual, create a fault code triple index for each fault code in the fault code list, and store the fault code triple index in the explicit knowledge base;

[0061] Step S1220: Associate the fault code triple index with the corresponding module knowledge unit.

[0062] Specifically, the equipment manual contains a lot of fault code information, and the fault code triple index is a "phenomenon-cause-treatment" triple. Such an index is constructed for each fault code, which can record in detail the fault phenomenon, cause and treatment method corresponding to the fault code. For example, for fault code E007, the corresponding phenomenon is abnormal fluctuation of the spindle speed, which is caused by insufficient torque of the coupling bolt. The corresponding module is M:012, and the treatment method is to refer to the operation steps K:003-002. Step S1210 integrates the scattered fault diagnosis information in the equipment manual by detailed disassembly and index construction of the fault code, which is convenient for subsequent quick search and use. The implementation process is to first filter out all fault codes from the equipment manual to form a fault code list, and then for each fault code, determine its corresponding phenomenon, cause and treatment measures based on the relevant content in the manual, and then construct a triple index. Finally, these indexes are stored in the explicit knowledge base to ensure orderly storage and convenient access of data.

[0063] In step S1220, the fault code triple index is associated with the corresponding module knowledge unit. In step S1100, the device knowledge has been split into knowledge units with hierarchical labels, such as [D: device type code - M: module code - K: knowledge category code]. In this step, the fault code triple index is associated with the corresponding module knowledge unit, so that when searching by fault code, the relevant module knowledge can be quickly located. For example, the cause of the fault code E007 mentions the corresponding module M:012. Then, through this association, when the E007 fault code is retrieved, the system can quickly associate it with the relevant knowledge under the M:012 module, such as the structural principle, operating specifications, etc. of the module, providing more comprehensive knowledge support for fault diagnosis. This step realizes rapid navigation from fault codes to specific knowledge by establishing an association between indexes and knowledge units, thereby improving the efficiency and accuracy of knowledge retrieval. The implementation process is to match and associate the module information involved in the fault code triple index with the module knowledge unit generated in step S1100, establish a connection between the two in the explicit knowledge base, and ensure the consistency and traceability of the information.

[0064] Step S1200 solves the problem of scattered fault diagnosis information and difficulty in quickly locating it in traditional systems. In traditional knowledge retrieval systems, when faced with a fault, technicians need to search for the cause and treatment method of the fault in a large amount of equipment manuals, which consumes a lot of time and energy. However, by establishing a fault code triple index in step S1210, the fault-related information is centralized and integrated, and by establishing an association with the module knowledge unit in step S1220, technicians can quickly obtain relevant cause analysis and treatment suggestions directly through the fault code when encountering a fault, greatly reducing the time for finding information and improving the efficiency of fault diagnosis. When the technician conducts a search, the fault code index in step S1200 can quickly provide relevant explicit knowledge, which is combined with the experience of the implicit knowledge layer to form a more comprehensive search result. Without step S1200, during the fusion search, it is impossible to quickly and accurately obtain fault-related knowledge from the explicit knowledge layer, which will lead to incomplete retrieval results, reduced fault diagnosis efficiency, and difficulty in achieving efficient fusion retrieval of implicit experience and explicit knowledge.

[0065] Step S2000: Acquire implicit knowledge and construct an implicit knowledge layer based on the implicit knowledge; the implicit knowledge at least includes work order text information;

[0066] Furthermore, step S2000 includes:

[0067] Step S2100: Parse the work order text into blocks to obtain structured semantic elements;

[0068] The block parsing method includes dividing the work order text into multiple information blocks and extracting the core semantics of each information block; the information blocks include fault phenomenon information blocks, processing action information blocks and experience summary information blocks;

[0069] Extracting the core semantics of each information block includes extracting fault keywords from the fault phenomenon information block, extracting action verbs from the processing action information block, and extracting scene labels from the experience summary information block;

[0070] Specifically, tacit knowledge primarily stems from technicians' practical experience. In addition to work order text, it also encompasses technicians' real-time observations and operational summaries during equipment maintenance and process commissioning. For example, through long-term maintenance work, technicians can roughly judge the degree of wear on equipment components based on their experience. This experience, difficult to precisely express in words, also constitutes tacit knowledge. This tacit knowledge is important to acquire because it contains key information for solving practical problems. However, traditional methods have made it difficult to effectively utilize it, necessitating subsequent processing.

[0071] Chunk parsing is a processing technology for natural language text. Because work order texts are written by technicians in natural language, they are complex and lack a standardized format, making them difficult to use directly. Chunk parsing breaks them down into distinct semantic components, converting them into structured semantic elements for easier analysis and utilization. In practice, natural language processing algorithms are used to segment text based on its semantic logic. For example, a work order text may describe something like, "The equipment made a strange noise while operating. I checked all the connections and found a loose screw. Furthermore, when repairing this type of equipment at night, it's best to first check key transmission components." The algorithm can identify that "the equipment made a strange noise while operating" is related to the fault phenomenon, "checked all the connections" is the action, and "when repairing this type of equipment at night, it's best to first check key transmission components" is a summary of experience. Consequently, the text is segmented into corresponding information chunks, achieving initial structuring. "The equipment made a strange noise while operating" is classified as the fault phenomenon chunk, "checked all the connections" is classified as the action chunk, and "when repairing this type of equipment at night, it's best to first check key transmission components" is classified as the summary of experience. Through this classification, information of different nature in the work order text can be processed separately, improving the pertinence and efficiency of information processing.

[0072] Extracting fault keywords from the fault phenomenon information block can quickly locate the fault type. For example, extracting "strange sound" from "the equipment makes a strange sound during operation" as a fault keyword facilitates subsequent retrieval of similar fault cases. Extracting action verbs from the processing action information block can clarify the specific operation method for handling the fault. For example, extracting the action verb "check" from "checked each connection point" provides an operation reference for other technicians. Extracting scenario labels from the experience summary information block can mark the specific scenarios where the experience is applicable. For example, extracting "night" as a scenario label from "When repairing this type of equipment at night, it is best to check the key transmission components first" makes the application of experience more targeted. This is achieved by using a pre-built keyword library, verb library, and scenario label library, and extracting them through a text matching algorithm. When a word in the work order text matches a word in the library, it is extracted as the corresponding core semantics.

[0073] Step S2100 effectively addresses the disordered natural language expression of work orders by technicians. Technicians often use freeform, disorganized language when recording work orders. S2100 processes the work order text into chunks, extracting core semantics from aspects such as the fault phenomenon, handling actions, and experience summaries. This transforms the disordered natural language into structured semantic elements, laying a solid foundation for building an implicit knowledge layer. This improves the accuracy and efficiency of implicit knowledge retrieval, creates favorable conditions for the integration of explicit and implicit knowledge, and helps enhance both maintenance and learning efficiency for technicians.

[0074] Step S2200 , performing a standardized mapping of colloquial expressions on the structured semantic elements to obtain standardized semantic elements; and generating a structured semantic description of the work order experience based on the standardized semantic elements.

[0075] Furthermore, step S2200 includes:

[0076] Step S2210: Construct a work order spoken language-professional terminology comparison table to form a multi-level mapping rule; the multi-level mapping rule includes a first-level equivalence mapping, a second-level feature mapping, and a third-level scenario mapping;

[0077] Step S2220 , standardizing the structured semantic elements using multi-level mapping rules, replacing spoken expressions with professional terms to obtain standardized semantic elements;

[0078] Step S2230: Generate a structured semantic description of the work order experience based on the standardized semantic elements, and store the structured semantic description of the work order experience in an implicit experience database.

[0079] Specifically, technicians are accustomed to using colloquial expressions when recording work orders, which can vary between technicians and hinder unified knowledge management and retrieval. For example, "bearing noise" and "bearing noise" should both be standardized as "abnormal bearing vibration" in professional fields. Equivalence mapping converts these synonymous colloquial expressions into standard professional terms to ensure consistency. Feature mapping infers and maps possible fault causes based on the fault characteristics expressed in the colloquial expression. For example, when an oil leak occurs, based on extensive historical maintenance data and empirical analysis, it is likely caused by seal failure. Therefore, the oil leak is mapped to seal failure, with the probability calculated based on historical data. Traditional work orders only record the phenomenon, such as oil leak, without explicitly specifying the specific cause. Feature mapping transforms the phenomenon into a structured cause label with diagnostic value, providing an inference basis for subsequent fault diagnosis. Scenario mapping identifies specific operating conditions in the colloquial expression and maps them to corresponding scenario labels. Scenario mapping addresses the scenario-specific nature of implicit experience. Technicians' experience is often related to specific working conditions, such as quick troubleshooting techniques in low-light conditions during night shifts. Through scenario mapping, these experiences can be marked as retrievable scenario tags, making it easier to make accurate recommendations based on the context of the current work order, such as time and equipment status. Building comparison tables and multi-level mapping rules can eliminate the understanding barriers caused by differences in language expression, allowing the experience of different technicians to be processed according to unified standards.

[0080] Based on the constructed multi-level mapping rules, the structured semantic elements extracted in step S2100 are processed. First, equivalence mapping is applied to replace the colloquial keywords obtained by block parsing with standard terms; secondly, feature mapping is performed to supplement the potential causes of the fault phenomenon; finally, scenario labels are added to the processing actions and experience summaries through scenario mapping; for example, if the fault keyword is bearing noise, it is replaced with abnormal bearing vibration according to the equivalence mapping rules; the colloquial action description in the processing action information block is also converted into professional expressions according to the corresponding rules; the scenario-related spoken language in the experience summary information block is converted into standard scenario labels according to the scenario mapping rules. Through this step, the work order experience recorded by different technicians is unified in the use of terminology, which facilitates subsequent storage, retrieval and analysis. The specific implementation process is to traverse each part of the structured semantic element, compare the colloquial expression with the work order spoken language-professional terminology comparison table, and replace it according to the mapping rules to obtain standardized semantic elements.

[0081] The standardized semantic elements are combined in the form of "standardized fault keywords + standardized action verbs + standardized scenario labels" to form a structured semantic description of work order experience. For example, after standardization, the structured description of the work order mentioned earlier might be [Abnormal bearing vibration - Check connection parts - Nighttime]. These structured descriptions are then stored in the implicit experience database, a database specifically designed to store technicians' implicit knowledge. This transforms the technicians' personalized experience into searchable structured data, supporting queries based on the "phenomenon (fault) - action - scenario" dimension. For example, when a technician encounters a similar equipment failure, they can quickly retrieve similar handling experiences from the implicit experience database by entering the relevant fault phenomenon, action, or scenario information. This step enables the structured storage of implicit knowledge, enabling technicians to quickly access relevant experience when encountering problems, thereby improving problem-solving efficiency.

[0082] Step S2200 addresses the arbitrariness and ambiguity of natural language expression. In traditional technician work order records, descriptions of the same fault or operation vary due to differences in technicians' language habits and expressive abilities, making it difficult to retrieve and utilize this implicit knowledge. Step S2210 constructs a comparison table and multi-level mapping rules to convert colloquial expressions into standard terms; step S2220 performs standardized replacement; and step S2230 generates and stores structured descriptions. This allows technicians' implicit knowledge to be presented in a unified, standardized format, improving the accuracy and efficiency of knowledge retrieval.

[0083] Through the multi-level mapping of S2200, the non-standard terms of implicit knowledge are converted into standardized labels that can be recognized by the explicit knowledge layer, such as being associated with [M:012-K:005], thus achieving the semantic connection of "implicit term → standard term → explicit knowledge unit". This provides the necessary semantic foundation for the cross-layer association in the subsequent step S3000, especially the alignment of bottom-level terms and the construction of the middle-level feature matrix, so that implicit experience can be associated with explicit knowledge in the same semantic space. Block parsing and standardization processing convert fragmented experience into structured triples, namely phenomenon-action-scene, solving the problem of "key experience being difficult to accurately match" caused by the ambiguity of natural language. For example, when a technician enters "abnormal noise", the system can simultaneously retrieve the principle explanation in the explicit knowledge and the "night shift abnormal noise troubleshooting skills" in the implicit experience through the standardized "abnormal bearing vibration", avoiding missed detection or false detection caused by inconsistent terminology in traditional systems. By extracting scene labels, such as "night shift" and "high-speed machining", the optimization experience of technicians under specific working conditions is converted into retrievable structured data, solving the technical problem in traditional systems where experience relies solely on personal memory storage and cannot be effectively shared. In the association matrix constructed in step S3200, the implicit experience features are directly derived from the standardized semantic elements generated by S2200. Without the processing of S2200, the implicit experience features will remain in a colloquial and non-standardized state, and will not be able to establish a quantitative association with the explicit knowledge units, resulting in the association matrix being unable to accurately reflect the semantic relationship between the two, thereby affecting the accuracy of the fusion retrieval.

[0084] Step S3000: cross-layer association between the explicit knowledge layer and the implicit knowledge layer using three progressive processing layers; the three progressive processing layers are bottom layer, middle layer, and top layer respectively;

[0085] Furthermore, step S3000 includes:

[0086] Step S3100, performing term alignment mapping at the bottom layer;

[0087] Specifically, the work order colloquialism-professional terminology comparison table constructed in step S2200 achieves a many-to-many mapping between implicit knowledge-layer terminology and explicit knowledge-layer terminology. Technicians often use colloquial terms to describe equipment failures, troubleshooting procedures, and other issues. These colloquial expressions differ from the professional terminology found in explicit knowledge, such as equipment manuals, making knowledge integration and retrieval difficult. For example, when describing a problem with a component, a technician might express it as "the screw is loose," while the corresponding professional term in explicit knowledge is "the bolt torque is insufficient." The work order colloquialism-professional terminology comparison table is constructed to address this issue. Using this comparison table, the system accurately maps the implicit knowledge-layer colloquial terminology used by technicians to the explicit knowledge-layer professional terminology, achieving a many-to-many mapping relationship. For example, "the screw is loose" can be mapped to "the bolt torque is insufficient" and further associated with the explicit knowledge tag [M:012-K:005], which points to a specific knowledge unit, such as bolt installation specifications. The process of achieving this mapping is that after receiving the search terms entered by the technician or the colloquial terms in the work order record, the system automatically queries the work order colloquial language-professional terminology comparison table, finds the corresponding professional terminology, and locates the corresponding knowledge unit label in the explicit knowledge layer based on the pre-set association rules.

[0088] A mapping dictionary and word weight table are established to support fuzzy matching of long-tail colloquial terms to standard terminology. Weights are set based on word frequency statistics. In real-world applications, colloquial expressions are rich and diverse, and a large number of long-tail colloquial terms exist, making it difficult to accurately match all of them to professional terminology. To address this issue, a mapping dictionary is established to store the correspondence between colloquial terms and standard terminology, while a word weight table is established to measure the reliability of the matching. Word frequency statistics are used as the basis for weighting. High-frequency colloquial terms appear frequently in real-world use, and their corresponding mappings are more reliable, thus receiving higher weights. Low-frequency terms appear less frequently and receive relatively lower weights. For example, "burning machine" is often used to describe a spindle motor burnout and is a high-frequency colloquial term. Its mapping to "spindle motor burnout" is assigned a weight greater than 0.9. Meanwhile, "clicking sound" is relatively uncommon and a low-frequency term. Its mapping to "bearing wear" is assigned a weight of 0.6. During implementation, the system analyzes a large amount of historical work order text, counting the frequency of each colloquial term, assigning corresponding weights based on frequency, and storing this information in the mapping dictionary and word weight table. During the retrieval process, when encountering long-tail colloquial words, the system performs fuzzy matching based on the mapping dictionary and combines the word weight table to evaluate the credibility of the matching results.

[0089] Step S3100 achieves term alignment mapping between the explicit and implicit knowledge layers. This eliminates language differences and establishes a cross-layer terminology bridge. This allows technicians to accurately perform associative searches between the explicit and implicit knowledge layers, regardless of whether they use colloquial expressions or specialized terminology. This seamless transition from implicit vocabulary to explicit content provides foundational support for subsequent convergent searches.

[0090] Step S3200: In the middle layer, a correlation matrix is ​​constructed with explicit knowledge units as rows and implicit experience features as columns; explicit knowledge units refer to knowledge units with hierarchical labels, and implicit experience features refer to standardized semantic elements;

[0091] Specifically, step S3200 aims to explore the implicit connections between explicit knowledge and implicit experience, construct a knowledge network, and further improve the efficiency of knowledge retrieval and utilization. Explicit knowledge units are those that have been decomposed and labeled hierarchically in step S1100, such as [D:001-M:012-K:003] representing the operational steps for lubricating the screw of a CNC machine tool feed system. Implicit experience features are semantic elements that have been standardized in step S2200, such as abnormal noise, inspection, and night shift.

[0092] An association matrix is ​​constructed with explicit knowledge units as rows and implicit experience features as columns. The purpose of constructing this matrix is ​​to quantify the correlation between each knowledge unit and various fault characteristics. In practice, the knowledge retrieval system generates this association matrix by analyzing and statistically analyzing a large number of historical diagnostic cases. For example, among numerous cases involving abnormal noise faults, statistics show that 80% of them reference the torque calculation formula. Therefore, in the association matrix, the cell value corresponding to the explicit knowledge unit (the torque calculation formula) and the implicit experience feature (the abnormal noise) is set to 0.8. This quantifies the connection between explicit knowledge and implicit experience. The association matrix can mine the semantic relationships between different knowledge units and discover the implicit connections between explicit knowledge and implicit experience. For example, the matrix reveals that the correlation between abnormal noise and anti-loosening gasket selection is as high as 0.9, although there may not be a direct link between the two in the explicit knowledge base. This association matrix reveals the inherent connections between them, providing technicians with more comprehensive knowledge association information. Moreover, the association matrix is ​​not static; the knowledge retrieval system continuously optimizes the association strength values ​​based on the latest diagnostic cases. When a new case appears, the system will re-count the relevant data and adjust the values ​​of the corresponding cells in the matrix, so that the knowledge retrieval system can adapt to the actual operating behavior of technicians and newly emerging fault conditions, and continuously improve the accuracy and practicality of knowledge associations.

[0093] Step S3200 solves the problem that it is difficult to find the implicit connection between explicit knowledge and implicit experience. In traditional knowledge retrieval systems, it is difficult for technicians to find the connection between seemingly unrelated explicit knowledge and implicit experience from a large amount of knowledge, resulting in the inability to fully utilize existing knowledge resources when solving practical problems. By constructing an association matrix to make these implicit connections explicit, technicians can obtain more comprehensive knowledge and improve their problem-solving capabilities. For example, when a technician encounters an abnormal noise fault, the association matrix can be used to find the close connection between the selection of anti-loosening gaskets and the fault, so that this factor can be taken into account during the maintenance process to avoid missing possible causes of the fault. When the technician enters the search term for a fusion search, the knowledge retrieval system can not only directly search the explicit knowledge base and the implicit experience base based on the search term, but also use the association matrix to mine the implicit connections related to the search term to generate a more comprehensive set of candidate answers. For example, when searching for abnormal noise faults, the system uses the association matrix to discover the connection between abnormal noise and the selection of anti-loosening washers. It then simultaneously searches the implicit experience library for scenario-based processing techniques, including "anti-loosening washers - inspection - high-speed machining." Combined with the explicit knowledge library's explanation of the principles of abnormal bearing vibration, it provides technicians with richer, more targeted search results, improving the quality and efficiency of retrieval. Without step S3200, the implicit connection between explicit knowledge and implicit experience cannot be discovered, and the fusion search can only be based on superficial keyword matching, failing to achieve deep knowledge fusion. The knowledge acquired by technicians will not be comprehensive and in-depth, making it difficult to effectively solve complex practical problems.

[0094] Step S3300: Dynamically adjust the personalized weight at the high level.

[0095] Furthermore, step S3300 includes:

[0096] Step S3310: Recording an operation trace log in real time, wherein the operation trace log includes explicit knowledge access features, implicit knowledge application features, and an operation step sequence;

[0097] Specifically, the purpose of step S3310 is to comprehensively collect behavioral data of technicians in the process of handling work orders, so as to gain a deeper understanding of their knowledge needs and mastery. The explicit knowledge access feature is used to record the interaction between technicians and explicit knowledge, including the hierarchical label of the knowledge unit accessed, the duration of a single visit, and the frequency of visits. The hierarchical label can clearly identify the specific knowledge content accessed by the technician. For example, [D:001-M:012-K:003] indicates that the technician accessed the knowledge related to the operating steps of the screw lubrication of the CNC machine tool feed system; the duration of a single visit reflects the technician's attention to the knowledge or the difficulty of understanding it. If the technician stays on the [K:004] page for a long time, it may mean that it is difficult to understand or that the knowledge is considered important; the frequency of visits reflects the technician's demand for specific knowledge. For example, frequent visits to [M:012-K:005] bolt installation specifications indicate that this part of knowledge is more critical to their work. The implicit knowledge application feature focuses on the technician's application of implicit knowledge, including the standardized semantic elements adopted in work order processing and their residence time. Standardized semantic elements record the implicit experience used by technicians in actual operations. Elements with high frequency of adoption, such as "abnormal noise-inspection-night shift", appear many times, indicating that technicians often rely on this type of experience in practice; the dwell time reflects the importance technicians attach to specific implicit experience or the difficulty of applying it. If the technicians stay on the page of abnormal bearing vibration handling skills for a long time, it may mean that there is a certain degree of difficulty in applying the experience or that they think it is very important. The sequence of operation steps records the key nodes of the work order processing process, such as fault code query → component disassembly → torque detection → reset debugging, showing the process of technicians handling work orders and providing a basis for analyzing the rationality of their operations. The implementation method is that when technicians use the knowledge retrieval platform to process work orders, this information is automatically recorded through the background program and stored in the operation trajectory log database for subsequent analysis.

[0098] Step S3320: Mark the key node operation results in the operation step sequence. The operation results are success or failure. The key node with a failure result is defined as a failure point. The failure includes explicit failure and implicit failure.

[0099] Specifically, step S3320 is a further analysis and processing of the sequence of operation steps. By marking the operation results, it is possible to discover the problems encountered by technicians in handling work orders. Explicit errors refer to violations of the standardized process of the explicit knowledge layer, such as directly resetting the bolt torque detection step without following the [M:012-K:005] operation. This indicates that the technicians have insufficient grasp and application of explicit knowledge. Implicit errors refer to failure to adopt effective experience in the implicit knowledge layer, such as ignoring the "prioritize high-speed wear parts" suggestion with the [S: Night Shift] label, which leads to the recurrence of the fault, indicating that the technicians do not pay enough attention to and apply implicit experience. During implementation, the knowledge retrieval system judges each key node in the sequence of operation steps based on the preset explicit knowledge standardized process and implicit knowledge effective experience rules. If the operation does not comply with the rules, it is marked as an error, and the error type and related information are recorded. In this way, the technicians' error points can be clearly identified, providing a basis for subsequent weight adjustments.

[0100] Step S3330: Obtaining personalized weights based on the operation track log;

[0101] Further, if Figure 3 As shown, step S3330 includes:

[0102] Step S3331: Calculate the basic interest weight W based on the explicit knowledge access characteristics and implicit knowledge application characteristics. base ;

[0103] Step S3332: Count the number of explicit errors, which is defined as the explicit error number; count the number of implicit errors, which is defined as the implicit error number. Based on the explicit error number and the implicit error number, calculate the error correction weight W. err ;

[0104] Step S3333: For the knowledge unit involving the error point, the basic interest weight and the error correction weight are superimposed to obtain the personalized weight W final .

[0105] Specifically, step S3330 calculates personalized weights based on the data collected and analyzed in steps S3310-S3320 to achieve accurate sorting of knowledge items to meet the personalized needs of technicians.

[0106] First, the explicit knowledge access features, including access frequency and single access duration, and the implicit knowledge application features, including the frequency of adoption of standardized semantic elements and residence time, are normalized. Normalization is a data processing method that aims to eliminate the dimensional differences of different feature data and make them comparable. For example, the numerical range of access frequency may be 0-100 times, while the range of single access duration may be 0-60 minutes. Through normalization, these data of different magnitudes are converted to the same value range, such as 0-1. The normalized access frequency is then defined as the explicit knowledge access frequency, the normalized single access duration is defined as the explicit knowledge access duration, the normalized standardized semantic element adoption frequency is defined as the implicit knowledge adoption frequency, and the normalized residence time is defined as the implicit knowledge residence time. These normalized features are then weighted and fused to calculate the basic interest weight W. base The weighted fusion method is to assign a certain weight to each feature according to its importance in reflecting the knowledge needs of technicians. For example, if it is believed that the access frequency and the frequency of adoption of standardized semantic elements are more important in reflecting the knowledge needs of technicians, then they can be assigned relatively high weights, and then W can be obtained according to certain calculation rules. base The calculation rule can be to multiply each feature value by its corresponding weight and then add them together. Different operational behavior characteristics have different importance in determining the technician's knowledge interest. Weighted fusion can comprehensively consider these factors and more accurately reflect the technician's interest in different knowledge.

[0107] The purpose of counting the number of explicit and implicit errors is to quantify the errors made by technicians during operation. For example, during multiple work orders, the number of times a reset was performed without following the bolt torque test step [M:012-K:005]; for implicit errors, the number of times a fault recurred due to ignoring the "prioritize high-speed wear parts" suggestion with the [S: Night Shift] label. Calculate W err When the number of mistakes is greater, it means that the technician's knowledge or experience in this area is weaker, and the weight of relevant knowledge needs to be increased to a greater extent to guide the technician to strengthen the learning and application of this knowledge. For example, different levels can be divided according to the number of mistakes, with fewer mistakes corresponding to lower W err Adjust the range, the more mistakes you make, the higher the W err Adjust the amplitude to achieve targeted handling of error situations.

[0108] Step S3333 is to convert the previously calculated W base and W err Combined, for the knowledge unit involving the error point, the two weights are added together to obtain the final personalized weight Wfinal This superposition method can comprehensively consider the interests and mistakes of technicians, determine a weight for each knowledge unit that better meets their actual needs, and sort the knowledge items according to this weight during retrieval, giving priority to displaying knowledge content that is more important and more in need of learning for technicians.

[0109] Step S3330 addresses the lack of personalization in knowledge push. In traditional knowledge retrieval systems, knowledge retrieval and push are often performed in a unified fashion, failing to meet the individual needs of different technicians. Step S3330 calculates personalized weights based on the technician's explicit knowledge access characteristics, implicit knowledge application characteristics, and error patterns, making knowledge push more tailored to each technician's specific circumstances. For example, for technicians who frequently access a certain type of knowledge and make a high number of errors in related operations, the knowledge retrieval system will increase the weight of this type of knowledge and prioritize relevant content, helping the technician strengthen their learning and mastery of this knowledge, thereby improving learning outcomes. Step S3330 provides core support for precise knowledge push. The personalized weights generated in step S3330 directly determine the priority of the search results in step S4300. During a fusion search, the knowledge retrieval system ranks candidate answer sets containing both explicit knowledge and implicit experience based on the personalized weights, ensuring that search results better meet the technician's actual needs. Without step S3330, search results would fail to reflect the individual differences of technicians, potentially leading to a mismatch between the knowledge acquired and their specific needs, reducing the practicality of the knowledge and learning efficiency. For example, technician A frequently misses checking "bolt torque". Without personalized weight adjustment, important knowledge such as "bolt torque detection steps" and "night shift bolt rapid detection skills" may not be displayed first when retrieving relevant knowledge. Technician A may continue to ignore these key contents, which is not conducive to his skill improvement.

[0110] Step S3300 solves the problem that technicians have difficulty focusing on key points during the knowledge acquisition process. By recording the operation trajectory log in real time, understanding the technicians' knowledge access and application in detail, marking the points of error, and calculating personalized weights, technicians can more clearly understand their own knowledge weaknesses and key needs. Step S3300 plays a role in optimizing knowledge retrieval and learning paths in the entire solution. It works in conjunction with steps S3100 and S3200 to further improve the association mechanism between explicit knowledge and implicit knowledge. Figure 4As shown, the explicit knowledge layer and the implicit knowledge layer are associated through the bottom layer, middle layer and high layer to form a progressive cross-layer semantic association system; the bottom layer term alignment mapping eliminates language differences and establishes a cross-layer term bridge, and the middle layer association matrix mines implicit logical associations and constructs a knowledge network; the bottom layer term alignment mapping and the middle layer association matrix construction realize the connection of knowledge, and step S3300, on this basis, screens and sorts the knowledge according to the individual differences of technical personnel, making knowledge retrieval more targeted and realizing accurate knowledge push.

[0111] Step S3300 provides a personalized ranking basis for fusion retrieval, making the search results more aligned with the technician's actual needs and achieving a shift from "searching for knowledge" to "searching for knowledge tailored to the individual." Without step S3300, the entire knowledge retrieval system would be able to achieve fusion retrieval, but it would not be optimized to meet the technician's personalized needs, resulting in low practicality and efficiency of the search results.

[0112] Step S4000 : Based on the cross-layer association between the explicit knowledge layer and the implicit knowledge layer, a context-sensitive fusion retrieval mechanism is established for real-time fusion recommendation of explicit knowledge and implicit experience.

[0113] Further, if Figure 5 As shown, step S4000 includes:

[0114] Step S4100: Receive a natural language search request from a technician, decompose the search terms in the search request, and construct structured queries for the explicit knowledge base and the implicit experience base respectively; score the search results of the two knowledge bases according to their relevance to obtain a set of candidate answers;

[0115] Step S4200: Obtain context information of the technician's current task and filter the candidate answer set based on the context information;

[0116] Step S4300: Generate a fusion retrieval report containing explicit knowledge and implicit experience based on the filtered candidate answer set.

[0117] Specifically, step S4000 leverages the previously constructed explicit knowledge layer, implicit knowledge layer, and cross-layer associations to establish a fusion retrieval mechanism. This allows for real-time fusion and recommendation of explicit knowledge and implicit experience, addressing challenges faced by technicians in knowledge acquisition and application. A technician enters a natural language search term, such as "the screw is loose," into the search interface. The knowledge retrieval system first decomposes and converts the natural language search term using the work order colloquialism-professional terminology comparison table and mapping dictionary constructed in step S3100. "The screw is loose" is mapped to the standard term "insufficient bolt torque" using the comparison table. Based on this standard term, a structured query is then constructed in both the explicit knowledge base and the implicit experience base. In the explicit knowledge base, the hierarchical tags [M:012-K:005] are used to directly locate relevant knowledge such as bolt installation specifications. In the implicit experience base, a search is triggered for work order experience containing the keyword "bolt torque." After the search is complete, the knowledge retrieval system scores the results from both knowledge bases based on their relevance. The basis for judging the relevance includes the degree of match between the search term and the knowledge content, the degree of association between the knowledge unit and the current work order scenario, etc. For example, if the search term "insufficient bolt torque" appears multiple times in a piece of explicit knowledge and the knowledge is closely related to the current equipment maintenance scenario, then the relevance score of this explicit knowledge will be higher; similarly, for the work order experience in the implicit experience library, if "bolt torque" is mentioned and the processing scenario is similar to the current work order, its relevance score will also increase accordingly. In this way, a set of candidate answers containing multiple possible answers is obtained to prepare for subsequent screening and recommendation. Step S4100 uses natural language processing and knowledge retrieval technology to solve the problem that technicians have difficulty in simultaneously acquiring explicit and implicit knowledge due to scattered knowledge storage, realizes efficient retrieval of the two knowledge bases, and improves the comprehensiveness of knowledge acquisition.

[0118] Contextual information includes the equipment model and the time period of the fault, for the work order the technician is currently handling. After obtaining this information, the knowledge retrieval system filters the candidate answer set based on the correlation matrix of explicit knowledge units and implicit experience features constructed in step S3200. For example, when a technician searches for knowledge related to the "abnormal noise" fault, the knowledge retrieval system extracts the principle explanation of "abnormal bearing vibration" from the explicit knowledge base and relevant experience from the implicit experience base in step S4100. Using the correlation matrix, it finds that the correlation between "abnormal noise" and "locking gasket selection" is 0.9. If the technician's current work order involves a high-speed machining scenario, the knowledge retrieval system will prioritize the scenario-specific processing techniques from the implicit experience base that include "locking gasket - inspection - high-speed machining" and retain them in the candidate answer set, while excluding other answers that are irrelevant to the current scenario. Step S4200 utilizes contextual information and the correlation matrix to address the problem of the candidate answer set potentially containing a large amount of irrelevant information. This makes the retrieval results more targeted, improves the efficiency and accuracy of knowledge screening, and helps technicians more quickly find knowledge that meets their specific needs.

[0119] Comprehensive sorting is performed based on multiple dimensions, including relevance, contextual adaptability, and personalized weight. Relevance is based on the scoring results of step S4100, contextual adaptability is based on the screening of contextual information in step S4200, and personalized weight is based on the technician's personalized weight generated in step S3300. For example, technician A frequently misses "bolt torque" inspections, which is a case of error point marking. When retrieving relevant knowledge, the personalized weight generated in S3300 will increase the weight of the explicit knowledge of bolt torque inspection steps and the implicit experience of night shift bolt rapid inspection techniques. This knowledge will be presented before other knowledge items during the S4300 sorting. Finally, the knowledge retrieval system presents the search report using mixed text and image layout and highlight extraction. Mixed text and image layout can present complex knowledge content to technicians in a more intuitive way, such as using pictures to show the structure of equipment components and text to explain the operating steps. Highlight extraction highlights key information, such as the cause of the problem and the solution, allowing technicians to quickly grasp the key points. Step S4300 integrates explicit knowledge and implicit experience to form a one-stop diagnostic learning model that allows technicians to learn, check and use at the same time. This solves the problem of technicians having difficulty combining theoretical knowledge with practical experience, improves the intuitiveness and interactivity of knowledge absorption, and helps technicians better understand and apply knowledge, thereby improving learning and work efficiency.

[0120] Step S4000 addresses several key issues faced by technicians during knowledge retrieval and application. First, it solves the problem of incomplete knowledge acquisition. In traditional knowledge retrieval systems, technicians need to query different knowledge sources separately, and implicit knowledge is difficult to accurately match, resulting in incomplete knowledge acquisition. Step S4100, by simultaneously searching both the explicit knowledge base and the implicit experience base, expands the scope of knowledge acquisition, ensuring that technicians can acquire more comprehensive knowledge. Second, it solves the problem of difficult knowledge screening. Step S4200 uses contextual information and the association matrix to filter the candidate answer set, avoiding the interference of a large amount of irrelevant knowledge and enabling technicians to quickly find knowledge relevant to the current task. Finally, it solves the problem of unfriendly knowledge presentation. Step S4300 presents the fusion search report using mixed text and image layout and highlight extraction, enhancing the intuitiveness and interactivity of knowledge absorption and facilitating technicians' understanding and application of knowledge. Step S4000 is a key step in achieving the goal of retrieval and fusion. In the entire solution, it plays a role in transforming the previously constructed and associated knowledge into practical and usable results. Step S3000 establishes a connection mechanism between explicit knowledge and implicit experience, and step S4000 uses this mechanism to accurately retrieve and recommend knowledge based on the technician's search request and contextual information, making the application of knowledge more targeted and efficient. S1000 and S2000 respectively construct the explicit knowledge layer and the implicit knowledge layer. S4000 integrates these two layers of knowledge, retrieves them, and presents them in a friendly manner, allowing technicians to obtain standardized operations and practical experience in one interface, avoiding the trouble of switching between multiple knowledge sources and improving the efficiency of knowledge acquisition and application. Step S4000, through collaboration with other steps, achieves the integrated intelligent retrieval goal of standardized knowledge + personalized experience, greatly improving the learning and work efficiency of technicians, shortening the work order processing time, and increasing the one-time repair rate of faults.

[0121] Example 2

[0122] This embodiment provides a knowledge retrieval system architecture platform based on embodiment 1, such as Figure 6 Shown, including:

[0123] The explicit knowledge layer construction module is used to obtain equipment manuals, use the equipment manuals as the data source, split the equipment knowledge into knowledge units with hierarchical labels, establish a fault code triple index, and store the knowledge units with hierarchical labels and the fault code triple index in the explicit knowledge base, which serves as the explicit knowledge layer.

[0124] Tacit knowledge layer construction module: used to obtain tacit knowledge and build the tacit knowledge layer based on the tacit knowledge;

[0125] Knowledge layer association module: uses three progressive processing layers to perform cross-layer association between explicit knowledge layer and implicit knowledge layer;

[0126] Fusion retrieval module: Based on the cross-layer association between the explicit knowledge layer and the implicit knowledge layer, a context-sensitive fusion retrieval mechanism is established for real-time fusion recommendation of explicit knowledge and implicit experience.

[0127] In the knowledge layer association module, the three progressive processing layers are bottom layer, middle layer and top layer respectively;

[0128] The method of using three progressive processing layers to perform cross-layer association between the explicit knowledge layer and the implicit knowledge layer includes:

[0129] Step S3100, performing term alignment mapping at the bottom layer;

[0130] Step S3200: In the middle layer, a correlation matrix is ​​constructed with explicit knowledge units as rows and implicit experience features as columns; explicit knowledge units refer to knowledge units with hierarchical labels, and implicit experience features refer to standardized semantic elements;

[0131] Step S3300: Dynamically adjust the personalized weight at the high level.

[0132] In the fusion retrieval module, the method for establishing a context-sensitive fusion retrieval mechanism for real-time fusion recommendation of explicit knowledge and implicit experience includes:

[0133] Step S4100: Receive a natural language search request from a technician, decompose the search terms in the search request, and construct structured queries for the explicit knowledge base and the implicit experience base respectively; score the search results of the two knowledge bases according to their relevance to obtain a set of candidate answers;

[0134] Step S4200: Obtain context information of the technician's current task and filter the candidate answer set based on the context information;

[0135] Step S4300: Generate a fusion retrieval report containing explicit knowledge and implicit experience based on the filtered candidate answer set.

[0136] The methods and systems of the present application may be implemented in many ways. For example, the methods and systems of the present application may be implemented using software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps used in the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above unless otherwise specified.

[0137] In addition, the parts of the above technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0138] The above-described specific embodiments further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is merely a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A knowledge retrieval system architecture method, characterized in that: The method comprises: Obtain the equipment manual and use it as the data source to split the equipment knowledge into knowledge units with hierarchical labels. Create a fault code triple index and store the knowledge units with hierarchical labels and the fault code triple index in the explicit knowledge base, using the explicit knowledge base as the explicit knowledge layer. Acquire tacit knowledge and build a tacit knowledge layer based on the tacit knowledge; Three progressive processing layers are used to perform cross-layer association between the explicit knowledge layer and the implicit knowledge layer; the three progressive processing layers are respectively the bottom layer, the middle layer and the top layer; The method for cross-layer association of explicit knowledge layers and implicit knowledge layers using three progressive processing layers includes: performing term alignment mapping at the bottom layer; constructing an association matrix at the middle layer with explicit knowledge units as rows and implicit experience features as columns; the explicit knowledge units are knowledge units with hierarchical labels, and the implicit experience features are standardized semantic elements; and performing dynamic adjustment of personalized weights at the top layer; Based on the cross-layer association between explicit knowledge layer and implicit knowledge layer, a context-sensitive fusion retrieval mechanism is established for real-time fusion recommendation of explicit knowledge and implicit experience.

2. A knowledge retrieval system architecture method according to claim 1, characterized in that: The method for splitting device knowledge to obtain knowledge units with hierarchical labels includes: Using the equipment manual as the data source, the equipment knowledge is split into three levels to obtain three knowledge units. Each knowledge unit is assigned a unique hierarchical label to obtain knowledge units with hierarchical labels.

3. A knowledge retrieval system architecture method according to claim 2, characterized in that: The first level of the three-level structure is the device type, the second level is the system module, and the third level is the knowledge category.

4. A knowledge retrieval system architecture method according to claim 3, characterized in that: The tacit knowledge at least includes work order text information; The method for constructing the implicit knowledge layer includes: parsing the work order text in blocks to obtain structured semantic elements; performing standardized mapping of the structured semantic elements into colloquial expressions to obtain standardized semantic elements; generating a structured semantic description of the work order experience based on the standardized semantic elements, and storing the structured semantic description of the work order experience in an implicit experience library.

5. A knowledge retrieval system architecture method according to claim 4, characterized in that: The block parsing method includes dividing the work order text into multiple information blocks and extracting the core semantics of each information block.

6. A knowledge retrieval system architecture method according to claim 5, characterized in that: The method for dynamically adjusting the personalized weight at the high level includes: recording an operation trace log in real time, and obtaining the personalized weight based on the operation trace log.

7. A knowledge retrieval system architecture method according to claim 6, characterized in that: The operation trace log includes explicit knowledge access features, implicit knowledge application features and operation step sequences; Calculate the basic interest weight based on the explicit knowledge access characteristics and implicit knowledge application characteristics; Based on the sequence of operation steps, error correction weights are obtained; The personalized weight is obtained based on the basic interest weight and the error correction weight.

8. A knowledge retrieval system architecture method according to claim 7, characterized in that: The method for obtaining the error correction weight based on the sequence of operation steps includes: Mark the key node operation results in the operation step sequence, the operation results are success or failure, and the key node with failure result is defined as the failure point; the failure includes explicit failure and implicit failure; The number of explicit errors is counted and defined as the number of explicit errors; the number of implicit errors is counted and defined as the number of implicit errors. The error correction weight is calculated based on the number of explicit errors and the number of implicit errors.

9. A knowledge retrieval system architecture platform, used to implement a knowledge retrieval system architecture method according to any one of claims 1 to 8, characterized in that: The system comprises: The explicit knowledge layer construction module is used to obtain equipment manuals, use the equipment manuals as the data source, split the equipment knowledge into knowledge units with hierarchical labels, establish a fault code triple index, and store the knowledge units with hierarchical labels and the fault code triple index in the explicit knowledge base, which serves as the explicit knowledge layer. Tacit knowledge layer construction module: used to obtain tacit knowledge and build the tacit knowledge layer based on the tacit knowledge; Knowledge layer association module: uses three progressive processing layers to perform cross-layer association between the explicit knowledge layer and the implicit knowledge layer; uses three progressive processing layers to perform cross-layer association between the explicit knowledge layer and the implicit knowledge layer; the three progressive processing layers are respectively the bottom layer, the middle layer, and the top layer; the method of using three progressive processing layers to perform cross-layer association between the explicit knowledge layer and the implicit knowledge layer includes: performing term alignment mapping at the bottom layer; constructing an association matrix at the middle layer with explicit knowledge units as rows and implicit experience features as columns; the explicit knowledge units are knowledge units with hierarchical labels, and the implicit experience features are standardized semantic elements; and performing dynamic adjustment of personalized weights at the top layer; Fusion retrieval module: Based on the cross-layer association between the explicit knowledge layer and the implicit knowledge layer, a context-sensitive fusion retrieval mechanism is established for real-time fusion recommendation of explicit knowledge and implicit experience.

Images