PCB component type selection method and system based on knowledge graph and intelligent retrieval

Abstract

The invention relates to a PCB component type selection method and system based on a knowledge graph and intelligent retrieval. Relates to the technical field of PCB design, and the method integrates knowledge in multiple aspects of performance parameters, packaging forms, supplier information and the like of components by constructing a PCB component knowledge graph, and establishes semantic association among the knowledge. And performing intelligent retrieval in the knowledge graph by using an information retrieval algorithm based on design requirements and constraint conditions input by a user, and quickly and accurately screening out components meeting requirements. Meanwhile, the multi-mode large model is combined to carry out comprehensive evaluation on the components, factors such as cost and supply stability are considered, optimal component type selection suggestions are provided for users, the efficiency and accuracy of component type selection in PCB design are improved, the cost is reduced, and intelligent development in the field of PCB design of electronic products is promoted.

Description

Technical Field

[0001] This invention relates to the field of PCB design technology, specifically to a method and system for selecting PCB components based on knowledge graphs and intelligent retrieval. Background Technology

[0002] With the rapid development of electronic technology, PCB (Printed Circuit Board) design plays a crucial role in electronic products. PCB design is not merely about connecting various components; it also requires considering factors such as component performance parameters, package types, and supplier information. In traditional PCB design, engineers typically rely on experience and manual research to select suitable components. This method is not only time-consuming and labor-intensive but also prone to errors. As the complexity of electronic products continues to increase, higher demands are placed on the accuracy and efficiency of component selection. Furthermore, in modern electronic product design, component selection must not only meet functional requirements but also consider factors such as cost, reliability, power consumption, and size. Therefore, developing an automated and intelligent system for PCB component selection is particularly important.

[0003] However, existing PCB component selection tools have several shortcomings. First, most tools use simple keyword matching algorithms, which fail to fully understand the complex relationship between user needs and components, resulting in low accuracy and applicability of the recommendations. Second, these tools lack intelligent recommendation mechanisms and cannot provide personalized selection suggestions based on specific design requirements and constraints, often requiring users to perform multiple manual screenings and verifications, increasing design complexity and time costs. Furthermore, existing selection tools often overlook various factors in practical applications, such as cost, supply stability, and compatibility, making it difficult to provide comprehensive selection advice. Therefore, there is an urgent need for a PCB component selection system capable of intelligent recommendation, accurate matching, and comprehensive evaluation to improve selection efficiency and accuracy, reduce costs, enhance supply chain stability, and promote the intelligent development of the electronic product PCB design field. Summary of the Invention

[0004] The purpose of this invention is to solve the problems of low efficiency, incomplete information, and difficulty in meeting personalized needs in the traditional PCB component selection process. This invention provides a PCB component selection method and system based on knowledge graphs and intelligent retrieval. The selection solution of this method can realize intelligent retrieval and evaluation of components, thereby improving the efficiency and accuracy of selection.

[0005] The technical solution adopted by this invention to solve its technical problem is: In a first aspect, the present invention provides a PCB component selection system based on knowledge graphs and intelligent retrieval, comprising: The knowledge graph construction module is used to collect and organize various information about PCB components, build a component knowledge graph, and establish semantic relationships between components and between components and related attributes. The information retrieval module is used to search the knowledge graph based on natural language processing technology and large model-based retrieval enhancement generation technology, and to select the components that meet the requirements according to the design requirements and constraints input by the user. The selection and evaluation module is used to comprehensively consider the influencing factors of components using a multimodal large model, evaluate and rank the selected components, and provide users with the best component selection suggestions based on the evaluation and ranking results. The data update module is used to update component information, market data and technical standards in real time or periodically, and feed them back to the knowledge graph construction module and the selection evaluation module.

[0006] Furthermore, the knowledge graph construction module includes: The data acquisition unit is used to collect data from the data source, perform data cleaning and standardization on the data collected from the data source, and build an initial dataset. The entity and relation processing unit is used to extract entities from the initial dataset using named entity recognition technology; extract the relationships between entities from the initial dataset using relation extraction technology; link identical entities and relations to remove redundant information; and combine entities and relations to generate a triple (h,r,t), where hi represents the i-th entity and ri represents the i-th relation. The knowledge graph construction unit is used to integrate entities, relations and triples to ultimately form a knowledge graph G=(E,R,T), where E represents the set of entities after removing redundant information, R represents the set of relations after removing redundant information, and T represents the set of triples composed of all triples. Among them, the named entity recognition technology uses a pre-trained named entity recognition model to recognize named entities; the relation extraction technology uses a pre-trained relation extraction model to extract relations; and the linking of entities and relations uses a graph database to link entities and model relations.

[0007] Furthermore, the information retrieval module includes: The user requirement parsing unit is used to parse the design requirements and constraints input by the user and transform them into structured queries. The intelligent retrieval unit is used to perform intelligent retrieval in the knowledge graph based on the parsed user requirements, and to filter out the components that meet the requirements.

[0008] Furthermore, the user requirement parsing unit includes: The text preprocessing subunit is used to remove stop words and punctuation marks from user input, perform word segmentation, stemming, and lemmatization, and generate preprocessed text. The requirements parsing subunit is used to identify key entities in user input using named entity recognition technology and extract relationships from user input.

[0009] Furthermore, the intelligent retrieval unit includes: The TF-IDF retrieval subunit is used to calculate the importance of user-input keywords in the knowledge graph and retrieve candidate components related to user needs based on the TF-IDF value. The Word2Vec retrieval subunit is used to convert user-input keywords and component names and their attributes in the knowledge graph into word vectors, calculate similarity, and filter out candidate components related to user needs. The BERT retrieval subunit is used to encode the user-input sentence and the component descriptions in the knowledge graph into vectors, calculate similarity, and filter out candidate components that are relevant to the user's needs. The large model retrieval enhancement generation subunit is used to generate more complex queries to cover more potential matching components and retrieve more candidate components in the knowledge graph based on the generated queries.

[0010] Furthermore, the intelligent retrieval unit also includes a result sorting and filtering subunit and a candidate component filtering subunit; The result sorting and filtering subunit is used to calculate the comprehensive score of each candidate component based on the query results, sort and filter the candidate components, and finally output the recommended results that meet the user's needs. The candidate component filtering subunit is used to filter out candidate components related to user needs from the knowledge graph based on similarity scores. By setting a similarity threshold, components below the threshold will be excluded. The method for setting the threshold is determined based on the actual situation and experimental results. The final output list of candidate components is selected by calculating the similarity between the user's input sentence and the component description in the knowledge graph using the BERT model.

[0011] Furthermore, the selection evaluation module includes: The data preparation module is used to extract the performance, cost, and supply data of each component from the knowledge graph, and to clean and standardize the data to ensure its accuracy and consistency. The feature extraction and standardization module is used to extract features from the extracted performance, cost, and supply data, and to use the Z-score standardization method to convert different types of features into dimensionless standard scores, eliminating dimensional differences. The multimodal large model evaluation module is used to combine standardized feature vectors, text descriptions and image information into multimodal input data, input it into a pre-trained and fine-tuned multimodal large model, perform forward propagation, calculate the evaluation score of each component, and comprehensively reflect the overall performance of the component in terms of performance, cost and availability. The results sorting and filtering module is used to sort candidate components in descending order based on evaluation scores, and to filter them according to the user's personalized needs and constraints, generating a final recommendation list.

[0012] Furthermore, the selection evaluation module also includes: The entity alignment submodule is used to identify and merge identical entities by comparing their attributes and context information, thus removing redundant information. It also includes a dynamic update submodule, which is used to periodically or in real-time retrieve the latest information from the data source and perform incremental or full updates.

[0013] Furthermore, the data update module is specifically used for: Manage and maintain multiple data sources, using web scraping techniques and API interfaces to obtain component information, market data, and technical standards from these sources; collect the latest component information, market data, and technical standards from multiple data sources, using Python's Scrapy or BeautifulSoup libraries for web scraping and Python's requests library for API requests; clean and integrate the collected data, using the Pandas library for data deduplication, missing value imputation, and outlier handling, and merge data from different sources into a unified dataset; The cleaned and integrated data is stored in the knowledge graph to ensure data persistence and accessibility. Nodes and edges for components, market data and technical standards are defined. The knowledge graph is constructed using RDF technology and stored in a relational database or graph database. Develop real-time or periodic data update strategies to trigger real-time updates immediately when new data is detected in the data source; use Webhooks or long polling techniques, while setting fixed update intervals; use scheduled task providers such as Cron or APScheduler to verify the quality and integrity of the updated data, ensuring that all necessary fields have been updated and that the new and old data are consistent; and use data validation libraries such as Pandas or SQL queries for verification.

[0014] Secondly, this invention provides a PCB component selection method based on knowledge graphs and intelligent retrieval, including: Collect and organize various information about PCB components, construct a component knowledge graph, and establish semantic relationships between components and between components and related attributes; Based on natural language processing technology and large model-based retrieval enhancement generation technology, the system searches the knowledge graph according to the user's input design requirements and constraints, and selects the components that meet the requirements. By using a multimodal large model to comprehensively consider the influencing factors of components, the selected components are evaluated and ranked, and the evaluation and ranking results are used to provide users with the best component selection suggestions. Component information, market data, and technical standards are updated in real time or periodically, and the knowledge graph and multimodal large model are also updated.

[0015] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a PCB component selection method based on knowledge graphs and intelligent retrieval.

[0016] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a PCB component selection method based on knowledge graphs and intelligent retrieval.

[0017] Fifthly, the present invention provides a computer program product, which includes computer instructions that instruct a computer to execute a PCB component selection method based on knowledge graphs and intelligent retrieval.

[0018] The beneficial effects of this invention are as follows: This invention relates to a PCB component selection system based on knowledge graphs and intelligent retrieval. The aim is to achieve efficient and accurate selection of PCB components by constructing and utilizing a knowledge graph, combined with advanced natural language processing technology and a multimodal large-scale model. Existing PCB component selection methods typically rely on manual searching and comparison, which is inefficient and prone to errors. The market offers a wide variety of components with complex performance parameters, and supplier information is scattered, lacking effective integration and management methods. Therefore, the purpose of this invention is to provide a PCB component selection system based on knowledge graphs and intelligent retrieval. By constructing a component knowledge graph and utilizing advanced natural language processing technology and a multimodal large-scale model, intelligent retrieval and evaluation of components can be achieved, thereby improving the efficiency and accuracy of component selection. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following description is provided with accompanying drawings of the relevant technical solutions in the embodiments of the present invention or the prior art. It should be understood that the accompanying drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This invention provides the basic framework for a PCB component selection system based on knowledge graphs and intelligent retrieval. Figure 2 A flowchart illustrating the PCB component selection method based on knowledge graphs and intelligent retrieval provided by this invention; Figure 3 This is a schematic diagram of an electronic device provided by the present invention. Detailed Implementation

[0021] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] Example 1 The first objective of this embodiment is to provide a PCB component selection system based on knowledge graphs and intelligent retrieval. The system includes: a knowledge graph construction module, an information retrieval module, a selection evaluation module, and a data update module. The knowledge graph construction module collects and organizes various information about PCB components, including performance parameters, packaging forms, and supplier information, to construct a component knowledge graph and establish semantic relationships between components and between components and related attributes. The information retrieval module, based on natural language processing technologies such as TF-IDF, Word2Vec, and BERT, and large-model-based retrieval enhancement generation technology, intelligently retrieves components from the knowledge graph based on user-input design requirements and constraints, filtering out components that meet the requirements. The selection evaluation module utilizes a multimodal large-model to comprehensively consider factors such as component performance, cost, and supply, evaluating and ranking the retrieved components to provide users with optimal component selection suggestions. The data update module updates component information, market data, and technical standards in real time or periodically to ensure the accuracy and timeliness of the knowledge graph and selection evaluation. The following provides a detailed description of each module: The knowledge graph construction module comprises a data acquisition unit, an entity and relation processing unit, and a knowledge graph construction unit. Specifically, the data acquisition unit collects data from data sources (such as supplier websites, e-commerce platforms, and technical standards websites), cleans and standardizes the data, and constructs an initial dataset. The entity and relation processing unit uses named entity recognition technologies (such as spaCy and Stanford CoreNLP) to extract entities from the initial dataset, and uses relation extraction technologies (such as BERT and RoBERTa) to extract relationships between entities. It links identical entities and relations, removes redundant information, and generates triples (h, r, t), where hi represents the i-th entity and ri represents the i-th relation. The knowledge graph construction unit integrates entities, relations, and triples to ultimately form a knowledge graph G=(E, R, T), where E represents the set of entities after removing redundant information, R represents the set of relations after removing redundant information, and T represents the set of triples composed of all triples. A graph database (such as Neo4j) is used for entity linking and relation modeling.

[0023] Specifically, the data acquisition unit is used to collect data from data sources, perform data cleaning and standardization on the collected data, and construct an initial dataset; the entity and relation processing unit is used to extract entities from the initial dataset using named entity recognition technology; extract relationships between entities from the initial dataset using relation extraction technology; link identical entities and relations to remove redundant information; and combine entities and relations to generate triples (h, r, t), where hi represents the i-th entity and ri represents the i-th relation; the knowledge graph construction unit is used to integrate entities, relations, and triples. The final result is a knowledge graph G=(E,R,T), where E represents the set of entities after removing redundant information, R represents the set of relations after removing redundant information, and T represents the set of triples composed of all triples. Named entity recognition technology uses pre-trained named entity recognition models (e.g., spaCy, Stanford CoreNLP) for named entity recognition. Relation extraction technology uses pre-trained relation extraction models (e.g., BERT, RoBERTa) for relation extraction. Entity and relation links are linked using graph databases (e.g., Neo4j) for entity linking and relation modeling.

[0024] More specifically, the goal of the knowledge graph construction module is to build a comprehensive, dynamic, and accurate PCB component knowledge graph, integrating various aspects of knowledge such as component performance parameters, package types, and supplier information, and establishing semantic relationships between components and their related attributes. First, the data acquisition unit extracts useful information from various structured and unstructured data sources, which may include supplier websites, e-commerce platforms, and technical manuals. During the data acquisition phase, the system uses web crawling techniques and API interfaces to automatically acquire data. This data is often disorganized and inconsistently formatted, making preprocessing crucial. Preprocessing includes data cleaning, format standardization, noise reduction, and missing value handling; these steps ensure data quality and provide a reliable foundation for subsequent knowledge graph construction.

[0025] After preprocessing, the system needs to construct an initial dataset containing all the information extracted and cleaned from the data sources. Next, the entity and relation processing unit is the core of the entire knowledge graph construction process. The system uses named entity recognition (NAME) technology to identify various entities from the initial dataset, such as component names, performance parameters, packaging forms, and supplier information. NAME is a natural language processing technique that extracts this information by recognizing specific entities in text. After identifying entities, the system uses relation extraction technology to identify relationships between entities in the initial dataset, such as "capacitor connected to resistor." Relation extraction technology identifies the associations between entities by analyzing their contextual information in the text. The entity and relation processing unit also needs to link identical entities and relations to remove redundant information. For example, different data sources may use different names or descriptions to refer to the same entity; the system needs to unify these duplicate entities into a standard entity. To achieve this, the system uses entity alignment technology, comparing entity attributes and contextual information to identify and merge identical entities and relations. This process not only reduces redundant information but also improves the accuracy and consistency of the knowledge graph. Next, the entity and relation processing unit combines entities and relations to generate triples. A triple takes the form (hi, ri, ti), where hi and ti represent entities, and ri represents the relationship between entities. In this way, the system structurally represents all entities and relations, forming a set of triples containing rich semantic information. The triple generation process needs to consider the logical relationships and contextual information between entities and relations to ensure that the generated triples accurately reflect the structure and semantics of the knowledge graph.

[0026] Preferably, the task of the knowledge graph construction unit is to integrate the extracted entities, relations, and generated triples to form a complete knowledge graph. Here, E represents the set of entities after removing redundant information, R represents the set of relations after removing redundant information, and T represents the set of triples composed of all triples. During the knowledge graph construction process, the system integrates and optimizes all entities, relations, and triples to ensure the knowledge graph has a reasonable structure and complete information. The knowledge graph construction unit also needs to handle the dynamic updates of the knowledge graph. Since the information from the data source is constantly changing, the system needs to update the knowledge graph periodically or in real time to ensure the timeliness and accuracy of its content. This process includes both incremental updates and full updates. Incremental updates only update the changed parts, while full updates rebuild the entire knowledge graph. Through this dynamic update mechanism, the system can maintain the latest state of the knowledge graph, providing a reliable data foundation for subsequent intelligent retrieval and selection evaluation.

[0027] The information retrieval module includes a user requirement parsing unit and an intelligent retrieval unit. The user requirement parsing unit is used to parse the design requirements and constraints input by the user and transform them into structured queries. The intelligent retrieval unit is used to perform intelligent retrieval in the knowledge graph based on the parsed user requirements and select the components that meet the requirements.

[0028] Furthermore, the information retrieval module, based on natural language processing technologies such as TF-IDF, Word2Vec, and BERT, as well as large-model-based retrieval enhancement and generation technologies, intelligently searches the knowledge graph based on the user's input design requirements and constraints to filter out components that meet the requirements. The user requirement parsing unit parses the user's input design requirements and constraints, transforming them into structured queries. This includes a text preprocessing subunit and a requirement parsing subunit. The text preprocessing subunit removes stop words and punctuation marks from the user input, performs word segmentation, stemming, and lexical reconstruction, generating preprocessed text. The requirement parsing subunit uses named entity recognition technology to identify key entities in the user input, such as component type, performance parameters, and suppliers, and extracts relationships from the user input, such as "withstand voltage is 50V" and "the supplier is XYZ company." Based on the parsed user requirements, the intelligent retrieval unit performs intelligent retrieval in the knowledge graph to filter out components that meet the requirements. The intelligent retrieval unit includes TF-IDF retrieval subunits, Word2Vec retrieval subunits, BERT retrieval subunits, and large-model-based retrieval enhancement and generation subunits. The TF-IDF retrieval subunit calculates the importance of the keywords entered by the user in the knowledge graph and retrieves candidate components related to the user's needs based on the TF-IDF value.

[0029] The Word2Vec retrieval subunit converts user-input keywords and component names and attributes from the knowledge graph into word vectors, calculates similarity, and filters candidate components relevant to the user's needs. The BERT retrieval subunit encodes user-input sentences and component descriptions from the knowledge graph into vectors, calculates similarity, and filters candidate components relevant to the user's needs. The large-model retrieval enhancement generation subunit generates more complex queries to cover more potential matching components and retrieves more candidate components from the knowledge graph based on the generated queries. The subunits can calculate a comprehensive score for each candidate component based on TF-IDF, Word2Vec, BERT, and query results generated by the large model, and sort and filter the candidate components, ultimately outputting recommendations that meet the user's needs. Specifically, the comprehensive score is first calculated based on the query results generated by TF-IDF, Word2Vec, BERT, and the large model, using the following formula: composite_score=α·tfidf_score+β·word2vec_score+γ·bert_score+δ·large_model_score Among them, α, β, γ and δ are weighting coefficients used to balance the influence of different scoring methods.

[0030] Next, the candidate components are sorted according to their comprehensive scores, using the following formula: sorted_components=sort(candidate_components,composite_score) The candidate components are sorted from highest to lowest based on their overall scores. Then, the final results are filtered. Based on user needs and constraints, the final recommended results are selected from the sorted candidate components using the following formula: final_recommendations=filter(sorted_components,user_constraints) As a specific solution, the user requirement parsing unit includes: a text preprocessing subunit and a requirement parsing subunit; the text preprocessing subunit is used to remove stop words and punctuation marks from the user input, perform word segmentation, stemming, and lemmatization, and generate preprocessed text; the requirement parsing subunit is used to use named entity recognition technology to identify key entities in the user input, such as component type, performance parameters, suppliers, etc., and extract relationships in the user input, such as "withstand voltage is 50V" and "the supplier is XYZ Company"; Specifically, the intelligent retrieval unit includes: a TF-IDF retrieval subunit, a Word2Vec retrieval subunit, a BERT retrieval subunit, and a large-model retrieval enhancement generation subunit. The TF-IDF retrieval subunit calculates the importance of user-input keywords in the knowledge graph and retrieves candidate components related to user needs based on TF-IDF values. The Word2Vec retrieval subunit converts user-input keywords and component names and attributes in the knowledge graph into word vectors, calculates similarity, and filters candidate components related to user needs. The BERT retrieval subunit encodes user-input sentences and component descriptions in the knowledge graph into vectors, calculates similarity, and filters candidate components related to user needs. The large-model retrieval enhancement generation subunit generates more complex queries to cover more potential matching components and retrieves more candidate components from the knowledge graph based on the generated queries.

[0031] More specifically, the goal of the intelligent retrieval unit is to perform intelligent searches within the knowledge graph based on the parsed user requirements, filtering out components that meet the criteria. To achieve this goal, the intelligent retrieval unit employs various retrieval technologies, including TF-IDF retrieval, Word2Vec retrieval, BERT retrieval, and large model retrieval enhancement generation technology.

[0032] First, the TF-IDF retrieval subunit calculates the importance of the user-input keywords within the knowledge graph and retrieves candidate components relevant to the user's needs based on the TF-IDF values. The TF-IDF value reflects the frequency of a keyword within a document and its inverse document frequency across the entire document set, effectively determining the importance of the keywords. Formula: tfidf_values=tfidf(parsed_demand,knowledge_graph)candidate_components=retrieve_candidates(tfidf_values) For example, the TF-IDF values ​​of "capacitor," "50V," and "XYZ Company" in the knowledge graph can be calculated to retrieve relevant candidate components. Using this method, the TF-IDF retrieval sub-unit can quickly filter out components highly relevant to user needs.

[0033] Next, the Word2Vec retrieval subunit converts the user-input keywords and component names and their attributes from the knowledge graph into word vectors, calculates similarity, and filters out candidate components relevant to the user's needs. Word vector representation captures the semantic relationships between words, and by calculating similarity, it can more accurately match user needs. Formula: word_vectors=word2vec(parsed_demand,knowledge_graph) similarity_scores=cosine_similarity(word_vectors) candidate_components=filter_candidates(similarity_scores) For example, by converting "capacitor," "50V," and "XYZ Company" into word vectors, the similarity to components in the knowledge graph is calculated, and relevant candidate components are selected. Through word vector representation, Word2Vec's retrieval subunits can better understand and match user-input keywords, improving the accuracy and efficiency of retrieval.

[0034] More specifically, the BERT retrieval subunit encodes the user-input sentence and component descriptions from the knowledge graph into vectors, calculates similarity, and filters out candidate components relevant to the user's needs. Through a pre-trained language model, the BERT model can understand contextual information within sentences, providing more accurate matching results. The formula is as follows: sentence_vectors=bert(parsed_demand,knowledge_graph) similarity_scores=cosine_similarity(sentence_vectors) candidate_components=filter_candidates(similarity_scores) For example, the system encodes the user-input sentence "I need a capacitor with a withstand voltage of 50V, and the supplier is XYZ Company" and the component descriptions in the knowledge graph into vectors, calculates the similarity, and filters out relevant candidate components. Through the BERT model, the system can gain a deeper understanding of the user-input sentences and capture the complex semantic relationships within them, thereby improving the accuracy and relevance of the retrieval.

[0035] The large-scale model enhances the generation of sub-units to generate more complex queries, covering more potential matching components, and retrieves more candidate components from the knowledge graph based on the generated queries. Through extensive training data, the large-scale model learns deeper semantic and relational patterns, enabling it to generate more complex queries and improve retrieval accuracy and coverage. Formula: generated_queries=generate_queries(parsed_demand) candidate_components=retrieve_candidates(generated_queries) For example, using a large model to generate the query "Find capacitors with a withstand voltage of 50V from XYZ Company" can retrieve more candidate components from the knowledge graph. By generating more complex queries, the large model retrieval enhancement sub-unit can discover more potential matching components, further improving the comprehensiveness and accuracy of the retrieval.

[0036] These retrieval technologies collectively constitute the core functionality of the intelligent retrieval unit. The TF-IDF retrieval subunit quickly filters candidate components relevant to user needs by calculating keyword importance. The Word2Vec retrieval subunit captures semantic relationships between words through word vector representation, improving matching accuracy. The BERT retrieval subunit provides deeper matching results by understanding sentence context. The large-model retrieval enhancement generation subunit generates complex queries, covering more potential matching components. Through these multi-dimensional retrieval technologies, the intelligent retrieval unit can comprehensively and accurately filter components that meet user needs, improving overall system performance and user experience.

[0037] In practical applications, the combined use of these retrieval technologies can significantly improve the accuracy and efficiency of intelligent retrieval.

[0038] For example, when a user enters "I need a capacitor with a withstand voltage of 50V, and the supplier is XYZ Company", The TF-IDF retrieval subunit calculates the importance of keywords such as "capacitor," "50V," and "XYZ Company," and retrieves relevant candidate components based on their TF-IDF values. The Word2Vec retrieval subunit converts these keywords into word vectors, calculates their similarity to components in the knowledge graph, and further filters for relevant candidate components. The BERT retrieval subunit encodes the user-input sentence and component descriptions from the knowledge graph into vectors, calculates similarity, and filters for more accurate candidate components. Finally, the large-model retrieval enhancement generation subunit generates more complex queries, covering more potential matching components, and retrieves more candidate components from the knowledge graph. Through these steps, the intelligent retrieval unit can provide users with the most suitable component recommendations, ensuring the accuracy and practicality of the recommendations.

[0039] More specifically, the intelligent retrieval unit also includes a result sorting and filtering subunit. This subunit sorts and filters the retrieved candidate components to ensure the accuracy and practicality of the recommendations. The result sorting and filtering subunit ensures the accuracy and practicality of the recommendations by comprehensively scoring, sorting, and filtering the retrieved candidate components. The comprehensive scoring subunit calculates a comprehensive score for each candidate component by combining the results of multiple scoring methods, ensuring the comprehensiveness and accuracy of the scoring. The result sorting subunit sorts the candidate components according to the comprehensive score, ensuring that the component with the highest score is listed first. The result filtering subunit further filters the final recommendation results based on user needs and constraints, ensuring the high relevance and practicality of the recommendations. Through these steps, the system can provide users with the most suitable component recommendations, improving user satisfaction and the overall performance of the system.

[0040] Specifically, the goal of the result ranking and filtering subunit is to rank and filter the retrieved candidate components to ensure the accuracy and usability of the final recommendation results. This subunit calculates a comprehensive score for each candidate component based on query results generated by TF-IDF, Word2Vec, BERT, and large models, and then ranks and filters the candidate components, ultimately outputting recommendations that meet user needs. First, the comprehensive scoring subunit is responsible for calculating a comprehensive score for each candidate component based on the query results generated by TF-IDF, Word2Vec, BERT, and large models. The comprehensive score is determined by combining the results of multiple scoring methods to ensure the comprehensiveness and accuracy of the score. Specifically, the comprehensive scoring formula is: composite_score=α·tfidf_score+β·word2vec_score+γ·bert_score+δ·large_model_score; α, β, γ, and δ are weighting coefficients used to balance the influence of different scoring methods. By adjusting these weighting coefficients, the system can optimize the scoring results according to the actual situation, ensuring that the comprehensive score accurately reflects the degree of matching between the candidate components and user needs.

[0041] Next, the results sorting subunit sorts the candidate components based on their comprehensive scores. The sorting process aims to rank the candidate components from highest to lowest comprehensive score, ensuring that the highest-scoring component appears first. Specifically, the sorting formula is: sorted_components = sort(candidate_components, composite_score). In this way, the system can quickly determine which candidate components best meet the user's needs, thereby improving the accuracy and usability of the recommendation results. The sorting process not only improves the quality of the recommendation results but also provides an orderly foundation for subsequent filtering.

[0042] Finally, the result filtering subunit selects the final recommended results from the ranked candidate components based on the user's needs and constraints. The filtering process aims to further narrow down the range of candidate components according to the user's specific needs and constraints, ensuring the high relevance and practicality of the recommended results. Specifically, the filtering formula is: final_recommendations=filter(sorted_components,user_constraints).

[0043] For example, if a user requires a capacitor with a withstand voltage of 50V from XYZ Company, the system will select suitable components from the ranked candidate components as the final recommendation. Through this process, the system ensures that the recommendations not only meet the user's explicit needs but also satisfy their implicit needs and preferences.

[0044] Preferably, the result sorting and filtering subunit is used to sort and filter the retrieved candidate components to ensure the accuracy and practicality of the recommended results. The result sorting and filtering subunit ensures the accuracy and practicality of the recommended results by comprehensively scoring, sorting, and filtering the retrieved candidate components. The comprehensive scoring subunit calculates a comprehensive score for each candidate component by combining the results of multiple scoring methods, ensuring the comprehensiveness and accuracy of the scoring. The result sorting subunit sorts the candidate components according to the comprehensive score, ensuring that the component with the highest score is placed first, providing an orderly basis for subsequent filtering. The result filtering subunit further filters the final recommended results based on the user's needs and constraints, ensuring the high relevance and practicality of the recommended results. Through these steps, the system can provide users with the most suitable component recommendations, improving user satisfaction and the overall performance of the system. For example, if a user inputs "I need a capacitor with a withstand voltage of 50V, and the supplier is XYZ Company," the system will calculate the comprehensive score of each candidate component, sort them from high to low scores, and finally filter out a capacitor with a withstand voltage of 50V and supplied by XYZ Company as the final recommended result. In this way, the system not only ensures the accuracy and practicality of the recommendations, but also improves the user experience.

[0045] Specifically, the goal of the result ranking and filtering subunit is to rank and filter the retrieved candidate components to ensure the accuracy and usability of the final recommendation results. First, the comprehensive scoring subunit is responsible for calculating the comprehensive score for each candidate component based on the results of multiple retrieval methods. The comprehensive score is determined by combining TF-IDF scores, Word2Vec scores, BERT scores, and large model scores to ensure the comprehensiveness and accuracy of the scoring. Specifically, the comprehensive scoring formula is: `composite_score` = α·tfidf_score + β·word2vec_score + γ·bert_score + δ·large_model_score; where α, β, γ, and δ are weighting coefficients used to balance the influence of different scoring methods. By adjusting these weighting coefficients, the system can optimize the scoring results according to the actual situation, ensuring that the comprehensive score accurately reflects the degree of matching between candidate components and user needs. For example, the system can obtain the TF-IDF score for each candidate component from the TF-IDF retrieval module, the Word2Vec score from the Word2Vec retrieval module, the BERT score from the BERT retrieval module, and the large model score from the large model retrieval enhancement generation module. These scores respectively reflect the evaluation results of different methods on candidate components.

[0046] Next, the sorting subunit sorts the candidate components according to the comprehensive score. The sorting process aims to arrange the candidate components from high to low according to the comprehensive score, ensuring that the component with the highest score is placed first. Specifically, the sorting formula is: sorted_components=sort(score_dict,key=lambdax:x[1],reverse=True).

[0047] In this way, the system can quickly determine which candidate components best meet the user's needs, thereby improving the accuracy and usability of the recommendations. The sorting process not only improves the quality of the recommendations but also provides an ordered foundation for subsequent filtering. For example, if the user's requirement is a 50V capacitor supplied by XYZ Company, the system will sort the candidate components based on a comprehensive score, ensuring that the component with the highest score is displayed first. The sorting process involves creating a dictionary with the candidate component's ID as the key and the comprehensive score as the value, and then using a sorting function to sort the dictionary in descending order.

[0048] Finally, the result filtering subunit selects the final recommendation from the ranked candidate components based on the user's needs and constraints. The filtering process aims to further narrow down the range of candidate components according to the user's specific needs and constraints, ensuring the high relevance and practicality of the recommended results. Specifically, the filtering formula is final_recommendations={component_id:component_info|∀condition_key,component_info[condition_key]=user_constraints[condition_key]}. For example, the system will select a capacitor with a withstand voltage of 50V and supplied by XYZ Company as the final recommendation from the ranked candidate components. Through this process, the system ensures that the recommended results not only meet the user's explicit needs but also satisfy the user's implicit needs and preferences, thereby improving user satisfaction and the overall performance of the system.

[0049] The intelligent retrieval unit also includes a candidate component filtering subunit. This subunit filters candidate components relevant to the user's needs from the knowledge graph based on similarity scores. Specifically, a similarity threshold is set, and components below this threshold are excluded. The method for setting the threshold is determined based on actual conditions and experimental results. The final output list of candidate components is generated by calculating the similarity between the user's input sentence and the component description in the knowledge graph using the BERT model.

[0050] The selection and evaluation module utilizes a multimodal large model to comprehensively consider factors such as component performance, cost, and availability, evaluating and ranking the retrieved components to provide users with optimal selection recommendations. This module includes: a data preparation module, used to extract performance, cost, and availability data for each component from the knowledge graph, and perform cleaning and standardization to ensure data accuracy and consistency; a feature extraction and standardization module, used to extract features from the extracted performance, cost, and availability data, and use the Z-score standardization method to convert different types of features into dimensionless standard scores, eliminating dimensional differences; a multimodal large model evaluation module, used to combine the standardized feature vectors, text descriptions, and image information into multimodal input data, inputting it into a pre-trained and fine-tuned multimodal large model for forward propagation, calculating the evaluation score for each component, comprehensively reflecting the overall performance, cost, and availability of the component; and a result ranking and filtering module, used to sort candidate components in descending order based on evaluation scores, and filter them according to the user's personalized needs and constraints, generating a final recommendation list.

[0051] In addition, the knowledge graph construction module also includes an entity alignment submodule, which is used to identify and merge identical entities by comparing entity attributes and context information, removing redundant information, and ensuring the accuracy and consistency of the knowledge graph; and a dynamic update submodule, which is used to periodically or in real time obtain the latest information from the data source and perform incremental or full updates to maintain the timeliness and accuracy of the knowledge graph.

[0052] This system utilizes a multimodal large model to comprehensively consider factors such as component performance, cost, and availability, evaluating and ranking the selected components to provide users with optimal selection recommendations. Specifically, the data preparation module is responsible for extracting performance, cost, and availability data for each component from the knowledge graph, and then cleaning and standardizing the data to ensure accuracy and consistency.

[0053] Specifically, during data preparation, the system extracts performance data such as voltage, current, and power of all capacitors from the knowledge graph, cost data such as price and discounts, and supply data such as inventory, delivery time, and supplier reputation. These parameters reflect the technical performance of the components in practical applications. For standardization, the system needs to calculate the overall mean (µ) and standard deviation (σ) for each performance parameter. The mean represents the average level of the parameter, and the standard deviation represents the degree of dispersion of the parameter. Specifically, for each performance parameter (such as voltage, current, and power), this embodiment uses the following formula to calculate its mean and standard deviation:

[0054]

[0055] Where N is the number of samples, and xi is the performance parameter value of the i-th sample. Assuming this embodiment has voltage data for 100 capacitors, the calculated mean voltage is 60V and the standard deviation is 20V.

[0056] Then, the system uses the Z-score normalization method to convert the performance parameters of each component into dimensionless standard scores. Z-score normalization, by subtracting the mean and dividing by the standard deviation, makes the values ​​of different parameters comparable on the same scale. Specifically, for each component's performance parameter x, this embodiment uses the following formula to calculate its normalized value z:

[0057] The system combines the standardized performance parameters into a feature vector, which serves as one of the inputs to the multimodal large model.

[0058] Similarly, the system also extracts cost information for each component from the knowledge graph, such as unit price, bulk discounts, and shipping costs. This information helps users make reasonable choices within their budget. Next, the system calculates the overall mean (µ) and standard deviation (σ) for each cost parameter and converts them into dimensionless standard scores using the Z-score standardization method. For capacitor C1, whose price is 0.5 yuan, the standardized price value is:

[0059] Furthermore, the system generates a standardized cost feature vector, which serves as one of the inputs to the multimodal large model. In addition, the system extracts supply information for each component from the knowledge graph, such as inventory levels, delivery dates, and supplier reputation. This information ensures that the selected components can be supplied on time and have reliable supply chain support. Next, the system also calculates the population mean (µ) and standard deviation (σ) for each supply parameter and uses the Z-score standardization method to convert them into dimensionless standard scores. For example, for capacitor C1, with an inventory of 1000 units, the standardized inventory value is:

[0060] Finally, the system combines the standardized performance characteristics, cost characteristics, and supply characteristics into a multimodal input vector, which serves as the final input to the multimodal large model. Specifically, this embodiment concatenates the three feature vectors of each component to form a multidimensional feature vector. This vector contains comprehensive information about the component's performance, cost, and supply, fully reflecting the component's characteristics. For example, for capacitor C1, the standardized performance feature vector is [-0.5, 0.5, 0.8], the cost feature vector is [1.0, 1.4], and the supply feature vector is [0.67, -1.0, 1.0]. These three vectors are concatenated into a multimodal input vector: multimodal_input=[-0.5,0.5,0.8,1.0,1.4,0.67,-1.0,1.0] To further improve the accuracy of the evaluation results, the system allows users to adjust the weights of different features according to their needs and preferences. Specifically, this embodiment introduces weight coefficients α, β, and γ to represent the importance of performance, cost, and supply characteristics, respectively. Users can set these weights according to their needs, and the system will weight the feature vector accordingly. For example, if the user is more concerned with performance, they can set α=0.6, β=0.3, and γ=0.1. For capacitor C1, the weighted multimodal input vector is: vc′=[0.6×-0.5,0.6×0.5,0.6×0.8,0.3×1.0,0.3×1.4,0.1×0.67,0.1×-1.0,0.1×1.0] In this way, the system can flexibly adapt to the needs of different users and provide more personalized component selection and evaluation results.

[0061] In the multimodal large model evaluation section, this embodiment selects a pre-trained multimodal large model suitable for component selection and evaluation tasks. These models have been pre-trained on large-scale datasets and possess powerful feature extraction and representation capabilities.

[0062] Commonly used multimodal large models include CLIP (Contrastive Language-Image Pre-training), ViLT (Vision-and-Language Transformer), and VATT (Video-Audio-Text Transformer). To ensure the model can handle text descriptions, image information, and structured data, this embodiment selects the ViLT model, which is suitable for component selection and evaluation tasks. Using HuggingFace's Transformers library, this embodiment can easily load a pre-trained ViLT model. After loading the pre-trained model, to better adapt it to the component selection and evaluation task, this embodiment fine-tunes the model. Fine-tuning can be achieved by continuing to train the model on a domain-specific dataset, making it more focused on the relevant features of component selection. Therefore, this embodiment prepares a dataset containing information on component performance, cost, availability, and other aspects for model fine-tuning. The dataset should include text descriptions, image information, and structured features of the components. During fine-tuning, this embodiment uses the cross-entropy loss function and the Adam optimizer for training. The model learns the importance of components in different dimensions and adjusts its parameters to improve evaluation accuracy. The loss function used during fine-tuning is:

[0063] Where yi is the true label of the i-th sample, pi is the probability distribution predicted by the model, and θ is the model parameter. The fine-tuned ViLT model can better understand the multimodal characteristics of components, thus providing more accurate evaluation results.

[0064] Next, this embodiment prepares the input data for the multimodal large model. First, this embodiment combines the standardized performance characteristics, cost characteristics, and supply characteristics into a multimodal input vector, which serves as the final input to the multimodal large model. Specifically, this embodiment concatenates the three feature vectors of each component together to form a multidimensional feature vector. This vector contains comprehensive information about the component's performance, cost, and supply, fully reflecting the component's characteristics. For example, for capacitor C1, the standardized performance feature vector is [-0.5, 0.5, 0.8], the cost feature vector is [1.0, 1.4], and the supply feature vector is [0.67, -1.0, 1.0]. These three vectors are then concatenated into a multimodal input vector: For each component c, its multimodal input vector vc can be represented as: [-0.5, 0.5, 0.8, 1.0, 1.4, 0.67, -1.0, 1.0] vc=concatenate(normalized_performance(c),normalized_cost(c),normalized_supply(c)) In addition to structured features, this embodiment can also add text descriptions for each component, such as technical specifications, application cases, and user reviews. Text descriptions provide more contextual information, helping the model better understand the characteristics of the components. The text description for each component is extracted from the knowledge graph and input into the multimodal large model along with the feature vector. The text description can be in natural language or a preprocessed embedding representation (such as BERT embedding). For example, for capacitor C1, the system can extract the following text description from the knowledge graph: "C1 is a high-performance ceramic capacitor suitable for high-frequency circuits, featuring low loss and high stability." The text description t and the feature vector vc are input into the multimodal large model: input=(processor(t),vc). If the component has related image information (such as product images, circuit diagrams, etc.), this embodiment can also use the image information as additional input to further enrich the model's input data. Images of the components are obtained from the knowledge graph or supplier websites and input into the multimodal large model along with the text description and feature vector. The image information can be in RGB image format or a preprocessed feature vector (such as ResNet embedding). For example, for capacitor C1, the system can obtain product images from the supplier's website and input them into the model along with text descriptions and feature vectors. The image I, text description t, and feature vector vc are input into the multimodal large model: input=(processor(I,t),vc).

[0065] Furthermore, after preparing the input data, this embodiment inputs the multimodal input data into the multimodal large model for forward propagation, calculating the evaluation score for each component. The evaluation score comprehensively reflects the overall performance of the component in terms of performance, cost, and availability. Using the forward propagation function of the multimodal large model, the input data is passed to each layer of the model, ultimately obtaining the output result. During the forward propagation process, the model automatically calculates the evaluation score for each component based on the input multimodal data. For each component c, its evaluation score sc can be represented as sc = model(input). The evaluation score of each component is extracted from the model's output for subsequent sorting and filtering. The model's output is usually a probability distribution or continuous value, representing the evaluation result of each component. The system needs to extract the specific evaluation score from the output and store it in a dictionary or list. Assuming the model's output is a vector S of length N, where Si represents the evaluation score of the i-th component, the system can extract the evaluation score using the formula sc = S[i]. To help users understand the meaning of the evaluation scores, the system can provide a scoring explanation module to explain the specific meaning of each evaluation score. Based on the range of evaluation scores, the system can classify components into different levels (e.g., excellent, good, average, poor). Furthermore, the system can generate detailed rating reports explaining the specific performance, cost, and availability of each component. For example, for capacitor C1 with an evaluation score of 0.85, the system might generate the following explanation: "C1 performs excellently in terms of performance, cost, and availability, especially scoring highly in performance, making it suitable for high-frequency circuits." Furthermore, this embodiment sorts candidate components in descending order based on their evaluation scores, ensuring that the highest-scoring components are listed first, facilitating quick access to the optimal option for the user. A dictionary is created with the IDs of the candidate components as keys and their evaluation scores as values. This dictionary is then sorted in descending order using a sorting function. The sorting function can rank components based on their evaluation scores. Assuming a dictionary named `scores`, where `scores[c]` represents the evaluation score of component `c`, the system can use the following formula for sorting: sorted_components=sorted(scores.items(),key=λx:x[1],reverse=True); Based on user needs and constraints, the system filters and recommends final results from the sorted candidate components. It parses user input, extracts key conditions, and further narrows down the candidate components based on these conditions. The system iterates through the sorted list of candidate components, checking whether each component meets the user's filtering criteria. Assume the user inputs filtering criteria. For constraints, the system can use the following formula for filtering: final_recommendations={c|∀k∈constraints,component_info[k]=constraints[k]}.

[0066] Finally, the system generates a recommendation list of the screened components for users to view and select. The recommendation list includes not only basic information about the components but also detailed rating reports and reasons for recommendation. The system sorts the screened components according to their evaluation scores and generates a recommendation list containing all eligible components. The reasons for recommending each component can include its specific performance, cost, and availability. For example, the system might recommend capacitors. C1 is recommended because it not only has a high evaluation score but also fully meets the user's withstand voltage and supplier requirements. The recommendation list will detail all the indicators of C1 and the reasons for recommending it.

[0067] In the results sorting and filtering section, the system stores the evaluation score of each component in a dictionary for subsequent sorting operations. Specifically, the evaluation score of each component is extracted from the output of the multimodal large model and stored in a dictionary named `scores`, where the key is the component ID and the value is the evaluation score. Assuming the output of the multimodal large model is a vector S of length N, where Si represents the evaluation score of the i-th component, the system can create the scoring dictionary using the following formula: `scores={ci:Si|i=1,2,...,N}` Next, the system sorts the candidate components in descending order based on their evaluation scores, ensuring that the highest-scoring components are listed first, making it easier for users to quickly find the optimal option. To do this, the system uses Python's built-in sorting functions. The `sorted()` function sorts the score dictionary. The sorting function can rank components based on their evaluation scores and return a list containing the component IDs and evaluation scores. Assuming a score dictionary `scores`, the system can sort it using the following formula: sorted_components=sorted(scores.items(),key=λx:x[1],reverse=True).

[0068] After sorting, the system generates a recommendation list for users to view and select. The recommendation list includes not only basic information about the components but also detailed rating reports and reasons for recommendation. The system sorts the components according to their evaluation scores and generates a recommendation list containing all candidate components. The reasons for recommending each component can include its specific performance in terms of performance, cost, and availability. To further narrow down the candidate components, the system parses user input and extracts key filtering conditions, such as withstand voltage, supplier, and delivery date. These conditions ensure that the recommendations better meet the user's needs. Specifically, the system uses Natural Language Processing (NLP) technology or form parsing tools to extract keywords and values ​​from the user input. For example, if a user enters "withstand voltage 50V and supplier is XYZ Company" in a text box, the system will automatically parse out the conditions of withstand voltage 50V and supplier being XYZ Company. Assuming the user enters a string "user_input", the system can use regular expressions or Named Entity Recognition (NER) technology to extract filtering conditions. constraints=parse_conditions(user_input).

[0069] Next, the system filters out components that meet the extracted filtering criteria from the sorted candidate components. Specifically, the system iterates through the sorted list of candidate components, checking whether each component meets the user's filtering criteria. If a component meets all the criteria, it is added to the final recommendation result; otherwise, it is excluded. Assuming there is a dictionary `component_info` containing component information, the system can use the following formula for filtering: `final_recommendations={c|∀k∈constraints,component_info[c][k]=constraints[k]}`.

[0070] To enhance filtering flexibility, the system also supports fuzzy matching. Fuzzy matching allows users to input imprecise conditions, and the system matches them based on similarity or a threshold. Specifically, the system uses similarity calculation algorithms (such as Jaccard similarity and cosine similarity) or fuzzy matching algorithms (such as Levenshtein distance) to calculate the similarity between the user's input and the component's attributes. If the similarity exceeds a preset threshold, the component is considered to meet the user's filtering criteria. Assuming the user inputs a fuzzy condition (fuzzy_condition), the system can perform fuzzy matching using the following formula: similarity=calculate_similarity(component_info[c][k],fuzzy_condition); ifsimilarity>threshold:final_recommendations.append(c).

[0071] Finally, the system integrates the filtered components into a final recommendation list for users to view and select. The recommendation list not only includes basic component information but also detailed rating reports and reasons for recommendation. The system sorts the filtered components according to their evaluation scores and generates a recommendation list containing all eligible components. The reasons for recommending each component can include its specific performance in terms of performance, cost, and availability. To provide users with more choices, the system can add some alternative components to the final recommendation list. While these alternative components may not fully meet the user's filtering criteria, they perform well in certain aspects and are worth considering. Based on user preferences and needs, the system can select some highly-rated alternative components from the sorted candidate components and add them to the final recommendation list; the reasons for recommending these alternative components can explain their advantages in certain aspects.

[0072] The data update module is designed to update component information, market data, and technical standards in real time or periodically, ensuring the accuracy and timeliness of the knowledge graph and selection evaluation. Specifically, it first manages and maintains multiple data sources to ensure data diversity and reliability. These sources include supplier websites, e-commerce platforms, and technical standard websites. Web scraping and API interfaces are used to obtain component information, market data, and technical standards from these sources. Next, the latest component information, market data, and technical standards are collected from multiple data sources. Python's Scrapy or BeautifulSoup libraries are used for web scraping, and Python's requests library is used for API requests. Then, the collected data is cleaned and integrated to ensure data quality and consistency. The Pandas library is used for data deduplication, missing value imputation, and outlier handling, and data from different sources is merged into a unified dataset. Finally, the cleaned data is processed... The cleaned and integrated data is stored in a knowledge graph to ensure data persistence and accessibility. Nodes and edges for components, market data, and technical standards are defined. The knowledge graph is constructed using RDF technology and stored in a relational database or graph database. Furthermore, a real-time or periodic data update strategy is developed to ensure the accuracy and timeliness of the knowledge graph and selection evaluation. When new data is detected in the data source, a real-time update is immediately triggered using webhooks or long polling techniques. A fixed update interval is set, such as daily, weekly, or monthly, using a scheduled task timer such as Cron or APScheduler. Finally, the quality and integrity of the updated data are verified, and relevant personnel are notified. It is ensured that all necessary fields have been updated and that the new and old data are consistent, using data validation libraries such as Pandas or SQL queries. When the data update is successful or fails, a notification email or message is sent using the SMTP protocol or message queue. Through these steps, this invention effectively ensures the accuracy and timeliness of the knowledge graph and selection evaluation, providing more accurate and timely data support.

[0073] The data update module can update component information, market data, and technical standards in real time or periodically, ensuring the accuracy and timeliness of the knowledge graph and selection evaluation. Specifically, this module first manages and maintains multiple data sources to ensure data diversity and reliability. These data sources include supplier websites, e-commerce platforms, and technical standard websites. The system uses web scraping technology and API interfaces to obtain component information, market data, and technical standards from these data sources. Next, it collects the latest component information, market data, and technical standards from multiple data sources, using Python's Scrapy or BeautifulSoup libraries for web scraping and Python's requests library for API requests. Then, the collected data is cleaned and integrated to ensure data quality and consistency. The Pandas library is used for data deduplication, missing value imputation, and outlier handling, and data from different sources is merged into a unified dataset. Finally, the cleaned and integrated data is stored in the knowledge graph to ensure data persistence and accessibility. Nodes and edges for components, market data, and technical standards are defined, and the knowledge graph is constructed using RDF technology and stored in a relational database or graph database.

[0074] Specifically, to ensure the accuracy and timeliness of the knowledge graph and selection evaluation, the system has established real-time or periodic data update strategies. For real-time updates, the system triggers an update immediately when new data is detected in the data source. For example, if new component information is published on a supplier's website, the system will automatically detect and trigger the update process using webhooks or long polling technology. The specific trigger condition is checking if new data is available in the data source; once new data is confirmed, the system will immediately perform the update operation. This real-time update mechanism ensures that the data in the knowledge graph is always up-to-date, promptly reflecting market changes and updates to technical standards. For periodic updates, the system sets fixed update intervals, such as daily, weekly, or monthly.

[0075] To achieve this goal, the system uses scheduled task schedulers, such as Cron or APScheduler, to arrange periodic update tasks. Periodic updates not only ensure that the data in the knowledge graph remains up-to-date but also provide users with a stable data update cycle, avoiding the resource waste caused by frequent updates. Furthermore, periodic updates can be combined with real-time update mechanisms to form a hybrid update strategy, ensuring data immediacy while improving system stability and efficiency. To ensure the quality and integrity of updated data, the system also has a strict data verification and notification mechanism. Regarding data verification, the system performs comprehensive integrity and consistency checks on the updated data. Integrity checks ensure that all necessary fields have been updated and no key information is missing; consistency checks ensure consistency between the old and new data, avoiding data conflicts or inconsistencies. Specific verification methods can use data verification libraries, such as Pandas or SQL queries. To promptly notify relevant personnel, the system also has a robust notification mechanism. When a data update is successful or fails, the system sends notification emails or messages to relevant personnel, ensuring they are informed of the update status immediately. Specifically, the system can use the SMTP protocol or message queues to send notifications. This not only improves system transparency but also enhances collaboration and communication among teams. In this way, relevant personnel can take timely measures to address potential problems, ensuring the system's smooth operation and data accuracy. Ultimately, the updated knowledge graph contains the latest component information, market data, and technical standards, providing users with more accurate and timely data support. Through this approach, the system can not only efficiently manage and update large amounts of component data but also ensure the accuracy and timeliness of the knowledge graph, providing users with scientific and reliable selection advice. The entire data update process is interconnected, from data collection, cleaning, and integration to updating and verification; each step is tightly linked, ensuring the system's efficiency and accuracy, and providing users with a reliable data support platform.

[0076] This invention proposes a PCB component selection system based on knowledge graphs and intelligent retrieval. It constructs a knowledge graph containing component performance parameters, packaging types, and supplier information, and utilizes information retrieval algorithms to intelligently filter components based on user design requirements and constraints, quickly and accurately recommending suitable components. Simultaneously, it combines a multimodal large-scale model to comprehensively evaluate cost and supply stability, providing users with optimal selection suggestions, improving PCB design efficiency and accuracy, reducing costs, and enhancing supply chain stability.

[0077] Example 2 like Figure 2 As shown, the present invention also provides a PCB component selection method based on knowledge graphs and intelligent retrieval, comprising the following steps: S1: Collect and organize various information about PCB components, construct a component knowledge graph, and establish semantic relationships between components and between components and related attributes; S2, based on natural language processing technology and large model-based retrieval enhancement generation technology, searches the knowledge graph according to the user's input design requirements and constraints, and selects the components that meet the requirements. S3 utilizes a multimodal large model to comprehensively consider the influencing factors of components, evaluates and ranks the selected components, and provides users with the best component selection suggestions based on the evaluation and ranking results. S4 updates component information, market data, and technical standards in real time or periodically, and also updates the knowledge graph and multimodal large model.

[0078] Example 3 like Figure 3 As shown, a third objective of this invention is to provide an electronic device, including a memory 701, a processor 702, and a computer program stored in the memory 701 and executable on the processor. When the processor executes the computer program, it implements the aforementioned PCB component selection method based on knowledge graphs and intelligent retrieval. The device also includes a communication interface 703 and a bus 704.

[0079] When the processor executes the computer program, it implements the PCB component selection method based on knowledge graph and intelligent retrieval as described in Example 2.

[0080] Example 4 The fourth objective of this invention is to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the PCB component selection method based on knowledge graphs and intelligent retrieval as described in Embodiment 2.

[0081] In this embodiment, the computer-readable storage medium is a non-volatile storage medium, specifically a ROM, RAM, disk, or optical disk. In this embodiment, the computer-readable storage medium is a 512GB NVMe SSD using 3D NAND flash memory technology, which features high-speed read / write performance and long lifespan.

[0082] When the computer program is executed by the processor, it implements the PCB component selection method based on knowledge graph and intelligent retrieval as described in Example 2.

[0083] Example 5 The fifth objective of this invention is to provide a computer program product, which includes computer instructions that instruct a computer to execute the PCB component selection method based on knowledge graphs and intelligent retrieval, as described in Embodiment 2.

[0084] In this embodiment, the computer program product is a software package, including an installer, main program, configuration files, help documentation, and sample data. The software package is distributed as an ISO image file and can be installed on the target computer via CD or USB flash drive.

[0085] The computer instructions in computer program products are written in a variety of programming languages, including C++ (core algorithms and hardware interfaces), Python (data analysis and visualization), and JavaScript (web interface).

[0086] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0087] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0088] This invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, readable storage media, optical storage, etc.) containing computer-usable program code.

[0089] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0090] Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A PCB component selection system based on knowledge graphs and intelligent retrieval, characterized in that, include: The knowledge graph construction module is used to collect and organize various information about PCB components, build a component knowledge graph, and establish semantic relationships between components and between components and related attributes. The information retrieval module is used to search the knowledge graph based on natural language processing technology and large model-based retrieval enhancement generation technology, and to select the components that meet the requirements according to the design requirements and constraints input by the user. The selection and evaluation module is used to comprehensively consider the influencing factors of components using a multimodal large model, evaluate and rank the selected components, and provide users with the best component selection suggestions based on the evaluation and ranking results. The data update module is used to update component information, market data and technical standards in real time or periodically, and feed them back to the knowledge graph construction module and the selection evaluation module.

2. The PCB component selection system based on knowledge graphs and intelligent retrieval according to claim 1, characterized in that, The knowledge graph construction module includes: The data acquisition unit is used to collect data from the data source, perform data cleaning and standardization on the data collected from the data source, and build an initial dataset. The entity and relation processing unit is used to extract entities from the initial dataset using named entity recognition technology; extract the relationships between entities from the initial dataset using relation extraction technology; link identical entities and relations to remove redundant information; and combine entities and relations to generate a triple (h,r,t), where hi represents the i-th entity and ri represents the i-th relation. The knowledge graph construction unit is used to integrate entities, relations and triples to ultimately form a knowledge graph G=(E,R,T), where E represents the set of entities after removing redundant information, R represents the set of relations after removing redundant information, and T represents the set of triples composed of all triples. Among them, the named entity recognition technology uses a pre-trained named entity recognition model to recognize named entities; the relation extraction technology uses a pre-trained relation extraction model to extract relations; and the linking of entities and relations uses a graph database to link entities and model relations.

3. The PCB component selection system based on knowledge graphs and intelligent retrieval according to claim 1, characterized in that, The information retrieval module includes: The user requirement parsing unit is used to parse the design requirements and constraints input by the user and transform them into structured queries. The intelligent retrieval unit is used to perform intelligent retrieval in the knowledge graph based on the parsed user requirements, and to filter out the components that meet the requirements.

4. The PCB component selection system based on knowledge graphs and intelligent retrieval according to claim 3, characterized in that, The user requirement analysis unit includes: The text preprocessing subunit is used to remove stop words and punctuation marks from user input, perform word segmentation, stemming, and lemmatization, and generate preprocessed text. The requirements parsing subunit is used to identify key entities in user input using named entity recognition technology and extract relationships from user input.

5. The PCB component selection system based on knowledge graphs and intelligent retrieval according to claim 4, characterized in that, The intelligent retrieval unit includes: The TF-IDF retrieval subunit is used to calculate the importance of user-input keywords in the knowledge graph and retrieve candidate components related to user needs based on the TF-IDF value. The Word2Vec retrieval subunit is used to convert user-input keywords and component names and their attributes in the knowledge graph into word vectors, calculate similarity, and filter out candidate components related to user needs. The BERT retrieval subunit is used to encode the user-input sentence and the component descriptions in the knowledge graph into vectors, calculate similarity, and filter out candidate components that are relevant to the user's needs. The large model retrieval enhancement generation subunit is used to generate more complex queries to cover more potential matching components and retrieve more candidate components in the knowledge graph based on the generated queries.

6. The PCB component selection system based on knowledge graphs and intelligent retrieval according to claim 5, characterized in that, The intelligent retrieval unit also includes a result sorting and filtering subunit and a candidate component filtering subunit; The result sorting and filtering subunit is used to calculate the comprehensive score of each candidate component based on the query results, sort and filter the candidate components, and finally output the recommended results that meet the user's needs. The candidate component filtering subunit is used to filter out candidate components related to user needs from the knowledge graph based on similarity scores. By setting a similarity threshold, components below the threshold will be excluded. The method for setting the threshold is determined based on the actual situation and experimental results. The final output list of candidate components is selected by calculating the similarity between the user's input sentence and the component description in the knowledge graph using the BERT model.

7. The PCB component selection system based on knowledge graphs and intelligent retrieval according to claim 1, characterized in that, The selection and evaluation module includes: The data preparation module is used to extract the performance, cost, and supply data of each component from the knowledge graph, and to clean and standardize the data to ensure its accuracy and consistency. The feature extraction and standardization module is used to extract features from the extracted performance, cost, and supply data, and to use the Z-score standardization method to convert different types of features into dimensionless standard scores, eliminating dimensional differences. The multimodal large model evaluation module is used to combine standardized feature vectors, text descriptions and image information into multimodal input data, input it into a pre-trained and fine-tuned multimodal large model, perform forward propagation, calculate the evaluation score of each component, and comprehensively reflect the overall performance of the component in terms of performance, cost and availability. The results sorting and filtering module is used to sort candidate components in descending order based on evaluation scores, and to filter them according to the user's personalized needs and constraints, generating a final recommendation list.

8. The PCB component selection system based on knowledge graphs and intelligent retrieval according to claim 7, characterized in that, The selection evaluation module also includes: The entity alignment submodule is used to identify and merge identical entities by comparing their attributes and context information, thus removing redundant information. It also includes a dynamic update submodule, which is used to periodically or in real-time retrieve the latest information from the data source and perform incremental or full updates.

9. The PCB component selection system based on knowledge graphs and intelligent retrieval according to claim 1, characterized in that, The data update module is specifically used for: Manage and maintain multiple data sources, using web scraping techniques and API interfaces to obtain component information, market data, and technical standards from these sources; collect the latest component information, market data, and technical standards from multiple data sources, using Python's Scrapy or BeautifulSoup libraries for web scraping and Python's requests library for API requests; clean and integrate the collected data, using the Pandas library for data deduplication, missing value imputation, and outlier handling, and merge data from different sources into a unified dataset; The cleaned and integrated data is stored in the knowledge graph to ensure data persistence and accessibility. Nodes and edges for components, market data and technical standards are defined. The knowledge graph is constructed using RDF technology and stored in a relational database or graph database. Develop real-time or periodic data update strategies to trigger real-time updates immediately when new data is detected in the data source; use Webhooks or long polling techniques, while setting fixed update intervals; use scheduled task providers such as Cron or APScheduler to verify the quality and integrity of the updated data, ensuring that all necessary fields have been updated and that the new and old data are consistent; and use data validation libraries such as Pandas or SQL queries for verification.

10. A PCB component selection method based on knowledge graphs and intelligent retrieval, characterized in that, include: Collect and organize various information about PCB components, construct a component knowledge graph, and establish semantic relationships between components and between components and related attributes; Based on natural language processing technology and large model-based retrieval enhancement generation technology, the system searches the knowledge graph according to the user's input design requirements and constraints, and selects the components that meet the requirements. By using a multimodal large model to comprehensively consider the influencing factors of components, the selected components are evaluated and ranked, and the evaluation and ranking results are used to provide users with the best component selection suggestions. Component information, market data, and technical standards are updated in real time or periodically, and the knowledge graph and multimodal large model are also updated.

Images