Large model-based demand information duplication judgment method and device, equipment, storage medium and product

Abstract

The embodiment of the invention relates to the technical field of intelligent text analysis, in particular to a demand information duplication judgment method and device based on a large model, equipment, a storage medium and a product, and the method comprises the steps: converting a demand text uploaded by a user into structured demand information through structured processing; screening similar historical demands from a historical demand library based on vectorization retrieval and semantic reordering; character-level and semantic-level two-dimensional similarity calculation is adopted, and a comprehensive score is obtained through weighted fusion; and automatically judging a demand repetition state according to a score threshold, and synchronously outputting an interpretability conclusion containing functional intention similarity description and field coincidence degree analysis. According to the scheme, through double verification of large model semantic understanding and structured rules, the accuracy and interpretability of demand duplicate judgment are remarkably improved, and meanwhile, continuous optimization and self-adaptive extension of a historical demand library are supported.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of intelligent text analysis, and particularly relate to a demand information duplication detection method and device based on a large model, equipment, a storage medium and a product. BACKGROUND

[0002] In the field of product development, demand management is a crucial link. With the acceleration of product iteration, the number of user-submitted requirements has increased dramatically. How to efficiently manage these requirements and avoid duplication of effort has become a major challenge for R&D teams. Traditional demand management methods mainly rely on manual review, comparing the textual content of demand descriptions to determine whether they are duplicated. However, this method has many drawbacks: first, the diversity of textual descriptions makes it difficult to ensure the accuracy of manual judgment. Even if two requirements are essentially the same, due to differences in description methods, word usage habits or expression angles, manual review may not accurately identify their duplication. Second, manual review is inefficient and difficult to meet the needs of rapid iteration. In the product development process, time is money, and any delay can delay product launch and affect market competitiveness. Manual review requires a lot of time and effort, making it difficult to handle large-scale, high-frequency demand submissions. In addition, manual review is also susceptible to subjective factors, leading to inconsistencies in the results. Different reviewers may have different understanding and judgment standards for the same requirement, resulting in disagreements and disputes.

[0003] To solve the above problems, the industry has begun to explore automated duplication detection technology. However, existing automated duplication detection methods are mostly based on simple text similarity calculations, such as keyword matching, TF-IDF algorithms, etc. These methods, while improving duplication detection efficiency to some extent, are unable to accurately understand the deep meaning and contextual relationships of requirements, and are not effective in handling complex semantic relationships. For example, when two requirements differ greatly in text but are highly similar in semantics, these methods often fail to accurately identify their duplication. SUMMARY

[0004] The main purpose of the embodiments of the present application is to provide a demand information duplication detection method, device, equipment, storage medium and product based on a large model, aiming to solve the technical problem of how to accurately identify duplicate improvement requirements in the intelligent manufacturing scenario and achieve efficient reuse of production optimization solutions in the prior art.

[0005] To achieve the above purpose, the embodiments of the present application provide a demand information duplication detection method based on a large model, which comprises the following steps: structurally processing the demand text uploaded by the user to obtain structured demand information containing demand entity annotations; based on the structured demand information, through a large language model integrated with retrieval enhancement generation technology, approximate nearest neighbor retrieval is performed in a similar historical demand vector library, and the relevance of the retrieval results is reordered based on context semantics to screen out several similar historical demands; The structured demand information is compared with each of the similar historical demands, character-level similarity scores and semantic-level similarity scores are calculated, and the character-level similarity scores and the semantic-level similarity scores are weighted and fused according to a preset weight configuration to obtain a comprehensive similarity score corresponding to each similar historical demand; According to the comprehensive similarity score, a demand repetition determination result is obtained.

[0006] Optionally, the user-uploaded demand text is structured to obtain structured demand information containing demand entity annotations, including: According to the demand text, a keyword set is extracted; According to the keyword set, a preset entity type is matched to obtain demand entity annotations; According to the demand entity annotations and a predefined association rule, logical combination is performed to obtain the structured demand information.

[0007] Optionally, based on the structured demand information, through a large language model integrated with retrieval enhancement generation technology, approximate nearest neighbor retrieval is performed in a similar historical demand vector library, and the relevance of the retrieval results is reordered based on context semantics to screen out several similar historical demands, including: According to the structured demand information, a query vector is generated to obtain a vectorized query request; According to the vectorized query request, approximate nearest neighbor retrieval is performed in a similar historical demand vector library to obtain a preliminary retrieval result set; Through the large language model, the relevance of the preliminary retrieval result set is reordered based on context semantics to obtain an optimized retrieval result; Based on the ranking score of the optimized retrieval result, a preset number of similar historical demands are screened out.

[0008] Optionally, the character-level similarity scores and the semantic-level similarity scores are calculated, and the character-level similarity scores and the semantic-level similarity scores are weighted and fused according to a preset weight configuration, including: According to the character sequence of the structured demand information and the character sequence of each similar historical demand, character-level similarity is calculated to obtain a character-level similarity score corresponding to each similar historical demand; Semantic similarity is calculated based on the semantic vector representation of the structured demand information and the semantic vector representation of each similar historical demand to obtain a semantic similarity score corresponding to each similar historical demand. Based on the preset weight configuration, the character-level similarity score and semantic-level similarity score corresponding to each similar historical requirement are weighted and fused to obtain the comprehensive similarity score.

[0009] Optionally, obtaining the demand duplication determination result based on the comprehensive similarity score includes: If the overall similarity score is less than a preset threshold, then the semantic feature vector required at the current time is obtained based on the semantic similarity score. Based on the character-level similarity score, the structured encoding of the required text is obtained; Based on the semantic feature vector and structured encoding, perform the database entry operation for similar historical requirements and send a database entry completion notification to the user.

[0010] Optionally, obtaining the demand duplication determination result based on the comprehensive similarity score further includes: If the overall similarity score is greater than a preset threshold, a duplicate demand determination conclusion is generated and fed back to the client terminal; Based on the semantic similarity score, a description of the functional intent similarity between the current requirement and similar historical requirements is obtained; Based on the character-level similarity score, an analysis of the field overlap of the required text is obtained; Based on the functional intent similarity description and field overlap analysis description, a request for repeated confirmation is generated and sent to the user terminal.

[0011] Furthermore, to achieve the above objectives, this application also proposes a demand information deduplication device based on a large model, the demand information deduplication device based on a large model comprising: The text processing module is used to process the user-uploaded requirement text into a structured form, resulting in structured requirement information containing requirement entity annotations. The demand retrieval module is used to perform near nearest neighbor retrieval in a similar historical demand vector library based on the structured demand information and a large language model with integrated retrieval enhancement generation technology, and to reorder the retrieval results based on the relevance of context semantics in order to filter out several similar historical demands. The similarity comparison module is used to compare the structured requirement information with each of the similar historical requirements. By calculating the character-level similarity score and the semantic-level similarity score, and by weighting and fusing the character-level similarity score and the semantic-level similarity score according to the preset weight configuration, a comprehensive similarity score corresponding to each similar historical requirement is obtained. The analysis conclusion feedback module is used to obtain the demand duplication determination result based on the similarity score, and to feed back the demand duplication determination result to the user.

[0012] Furthermore, to achieve the above objectives, this application also proposes a large-model-based demand information deduplication device, which includes: a memory, a processor, and a large-model-based demand information deduplication program stored in the memory and executable on the processor. The large-model-based demand information deduplication program is configured to implement the steps of the large-model-based demand information deduplication method described above.

[0013] Furthermore, to achieve the above objectives, this application also proposes a storage medium storing a large-model-based requirement information deduplication program. When the large-model-based requirement information deduplication program is executed by a processor, it implements the steps of the large-model-based requirement information deduplication method described above.

[0014] Furthermore, to achieve the above objectives, this application also proposes a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the large model-based demand information deduplication method described above.

[0015] One or more technical solutions proposed in this application have at least the following technical effects: The demand information deduplication method based on a large language model proposed in this invention significantly improves the ability to identify professional technical needs and its application value by integrating structured processing and semantic understanding technologies. This method employs structured processing, specifically by annotating and extracting key information from the requirement text. This makes the expression of requirement information more standardized and clear, helping subsequent retrieval and comparison processes to more accurately focus on the core content. Then, using a large language model with retrieval enhancement generation technology, the method identifies the requirement information record with the highest similarity to the structured requirement information in the historical requirement database. The similarity assessment includes character-level similarity scoring and semantic-level similarity scoring. Character-level similarity scoring reflects the similarity of requirement texts at the textual level, while semantic-level similarity scoring captures the similarity of requirements at a deeper semantic level. These two scores are weighted and fused according to a preset weight configuration. The calculation method of the comprehensive similarity score can be adjusted according to the different levels of emphasis placed on textual and semantic similarity in actual requirement management scenarios. Through this method of calculating the comprehensive similarity score, the similarity between requirements is precisely quantified in numerical form, providing a clear and quantifiable basis for subsequent requirement duplication determination. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the first embodiment of the requirement information deduplication method based on a large model according to this application. Figure 2 The flowchart of the deduplication method for the first embodiment of the demand information deduplication method based on a large model provided in this application is as follows: Figure 3 A schematic diagram of the model application of the first embodiment of the requirement information deduplication method based on a large model provided in this application; Figure 4 A flowchart illustrating the second embodiment of the requirement information deduplication method based on a large model provided in this application. Figure 5 A flowchart illustrating the third embodiment of the requirement information deduplication method based on a large model provided in this application. Figure 6 A structural block diagram of the first embodiment of the demand information deduplication device based on a large model provided in this application; Figure 7 This is a schematic diagram of the structure of a demand information deduplication device based on a large model in the hardware operating environment involved in the embodiment of this application. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of the embodiments of this application and are not intended to limit the embodiments of this application.

[0020] To better understand the technical solutions of the embodiments of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0021] Currently, in actual production operations, different roles such as workshop technicians and equipment engineers submit improvement requests based on their respective work perspectives. These requests often present the following typical problems: improvement requests for the same production problem are often submitted multiple times by different personnel, making it difficult for existing systems to effectively identify these semantically similar requests, leading to duplicate processing. Traditional requirement management methods rely on manual comparison or simple keyword matching, which cannot accurately identify semantically similar improvement requests, resulting in wasted resources. Therefore, how to accurately identify repetitive improvement requests in smart manufacturing scenarios and achieve efficient reuse of production optimization solutions is a pressing technical problem that needs to be solved.

[0022] To address the aforementioned issues, this application provides a method for deduplication of requirement information based on a large-scale model. This method transforms user-uploaded requirement text into structured requirement information through structured processing. Using a large language model, it employs both character-level and semantic-level similarity calculations, and obtains a comprehensive score through weighted fusion. Based on the score, similar historical requirements are filtered from a historical requirement database. Furthermore, the method automatically determines the requirement duplication status based on a score threshold, and simultaneously outputs interpretable conclusions including descriptions of functional intent similarity and field overlap analysis. This solution significantly improves the accuracy and interpretability of requirement deduplication through dual verification using large-scale model semantic understanding and structured rules, while also supporting continuous optimization and adaptive expansion of the historical requirement database.

[0023] The execution subject of this application embodiment can be a demand information deduplication device based on a large model, or a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or a thermal management device capable of implementing the above functions based on a demand information deduplication device based on a large model. This embodiment does not specifically limit it in this way. The following uses a demand information deduplication device based on a large model as the execution subject as an example to describe this embodiment and the following embodiments.

[0024] Figure 1 This is a flowchart illustrating the requirement information deduplication method based on a large model provided in an embodiment of this application.

[0025] Reference Figure 1 In this embodiment, the method for deduplication based on demand information of a large model includes steps S10 to S40: Step S10: Perform structured processing on the user-uploaded requirement text to obtain structured requirement information containing requirement entity annotations.

[0026] Structured processing of request text refers to transforming non-standardized text requests submitted by users into a machine-readable format with standardized fields using natural language processing techniques. Specifically, this process first performs word segmentation and part-of-speech tagging on the original text to identify key entities and action descriptions.

[0027] In practice, the above process segments the continuous text submitted by the user into meaningful units. For example, the requirement text "device malfunction" is segmented into "device / malfunction / failure," transforming the original natural language text into multiple isolated keywords. Each segmented word is then labeled with its part of speech (e.g., noun, verb, adjective). For instance, "device" is labeled as a noun, and "malfunction" as a verb. This process helps subsequent processing better understand the text's structure. Combined with a pre-defined domain knowledge framework, key information in the requirement text can be accurately extracted and mapped into a standardized data structure.

[0028] Understandably, this structured processing not only eliminates the interference of colloquial expressions but also transforms free text into standardized information containing clear semantic units. For example, users might use different expressions to describe the same problem, such as "Equipment A is broken," "Equipment A has malfunctioned," or "Equipment A has a problem." However, after structured processing, these expressions will be uniformly mapped to a standardized format like "Equipment Name: Equipment A, Fault Phenomenon: Fault." The core value of this structured processing lies in establishing a unified demand representation system. Improvement needs in the manufacturing field often involve multiple expressions for the same problem. For example, excessive temperature might be described as equipment overheating or poor heat dissipation. Structured processing, through the standardization of domain terminology, can normalize these near-synonymous expressions. In practice, it combines industry knowledge graphs to establish synonym mapping relationships for concepts such as equipment components and failure modes, ensuring that different people's descriptions of similar problems converge to the same semantic nodes. For example, in a factory, the demand for tool wear, whether it's milling cutter chipping or cutting edge dulling, will ultimately be categorized under the standard category of tool wear. It should be noted that the degree of structuring needs to be balanced with actual business operations; excessive standardization may result in the loss of key details.

[0029] Step S20: Based on structured demand information, a large language model with integrated retrieval enhancement generation technology is used to perform an approximate nearest neighbor search in the similar historical demand vector database, and the search results are reordered based on the relevance of contextual semantics to select several similar historical demands.

[0030] The similar historical demand vector library refers to a pre-established structured information database containing all previously processed demands. Each record stores key information such as equipment type, fault symptoms, and solutions in a standardized format. During a search, the system matches demands based on the various features of the structured demand entities, such as equipment category, fault location, and problem type. Different features yield different matching results. When matching results from multiple features intersect, the matching result within this intersection has the highest similarity to the aforementioned demand information. For example, for a demand regarding "spindle bearing overheating," the system will automatically search the historical database for demand records that also involve spindle bearings and have temperature-related issues.

[0031] Understandably, the purpose of this step is to analyze the semantic similarity between different requirements. For example, when searching for the problem of "tool wear," records in the historical database related to "insufficient milling cutter life" would also be identified as similar requirements, because both reflect the essential problem of declining tool performance. This search method fully considers the synonymy and relevance of manufacturing terminology, ensuring that requirements with different expressions but the same essence can be accurately identified.

[0032] In one feasible embodiment, a query vector is generated based on structured demand information to obtain a vectorized query request; an approximate nearest neighbor search is performed in a similar historical demand vector library based on the vectorized query request to obtain a preliminary search result set; the preliminary search result set is reordered based on the relevance of context semantics using a large language model to obtain optimized search results; and a preset number of similar historical demands are selected based on the ranking score of the optimized search results.

[0033] Vectorized query requests refer to the process of converting structured requirement information into numerical vector representations in a high-dimensional space using a deep learning model. Specifically, this process first encodes the structured requirement information using a pre-trained deep learning model (such as a Transformer architecture model). The model maps each requirement to a high-dimensional vector space based on the vocabulary, semantics, and contextual relationships within the requirement text. In this space, similar requirement vectors are geometrically closer, allowing the semantic similarity between requirements to be measured by calculating the similarity between vectors. This representation enables subsequent steps to quickly locate candidate sets of similar requirements within a massive vector database using optimized indexing algorithms and approximate nearest neighbor retrieval.

[0034] Understandably, in practical applications, even if the descriptions of requirements differ, such as "increased radial runout of the spindle" and "abnormal vibration of the bearing housing," their inherent correlation can be identified through vectorization and semantic analysis. The system uses a pre-trained domain semantic model to identify requirements with different descriptions but pointing to the same equipment problem as highly similar cases. This method is particularly suitable for complex fault description scenarios in manufacturing, significantly improving the efficiency of reusing historical experience.

[0035] It should be understood that, in this embodiment, the large model, through deep semantic analysis of the requirement text, can accurately identify the inherent connections between different technical terms, such as understanding the causal relationship between "spindle vibration" and "bearing abnormality" in mechanical equipment failure. This understanding capability allows structured information to go beyond surface word matching and uncover the essential technical characteristics behind the requirement. Secondly, when the large model transforms the structured requirement into a vector representation in a high-dimensional semantic space, it incorporates professional knowledge from the manufacturing field, ensuring that the transformed vector not only contains textual features but also implies the correlation of technical parameters. This provides high-quality input for subsequent approximate nearest neighbor retrieval. Finally, the large model's attention mechanism can analyze the deep-seated correlation between the retrieval results and the current requirement, and intelligently re-rank them by comprehensively considering background factors such as equipment service life and operating environment. This makes the final similar cases provided both technically relevant and practically valuable.

[0036] Step S30: Compare the structured requirement information with each similar historical requirement, calculate the character-level similarity score and the semantic-level similarity score, and perform weighted fusion of the character-level similarity score and the semantic-level similarity score according to the preset weight configuration to obtain the comprehensive similarity score corresponding to each similar historical requirement.

[0037] The similarity calculation process is achieved through multi-dimensional feature matching. This step first extracts the structured features of the current and historical requirements, and then performs a quantitative comparison.

[0038] Understandably, similarity calculation involves not only matching semantic surface features but also assessing the consistency of the technical connotations of the two requirements. For example, for problems like "overheating" and "abnormal temperature rise" on a CNC machine tool spindle—which have different expressions but are essentially the same—the system will use a domain knowledge model to identify their technical relevance, thus giving a high similarity score. Simultaneously, this method also analyzes the numerical differences in fault feature parameters and combines this with contextual information such as equipment model and operating conditions for a comprehensive judgment, enabling the system to accurately identify requirement records that, although expressed differently, actually reflect the same equipment problem.

[0039] Step S40: Obtain the result of demand duplication determination based on similarity score.

[0040] The generation of duplicate requirement determination results requires consideration of a preset duplicate threshold. This threshold is a critical value set after extensive case validation, used to distinguish whether requirements have sufficient similarity. If the similarity between the current requirement and historical cases exceeds this threshold, it is determined to be a duplicate requirement; otherwise, it is determined to be a new requirement.

[0041] Understandably, the timeliness of the demand is also considered during the judgment process. For example, frequently occurring problems may be given a higher matching priority. When providing feedback, the system will automatically associate the historical solutions with the highest matching degree and mark key matching factors. For example, when the spindle vibration problem of CNC machine tools is close to the score of a historical bearing wear case, the system will clearly indicate the similarity contribution of the two in the vibration spectrum characteristics.

[0042] It's important to note that after reaching a conclusion, the system must provide the user with a complete set of information supporting the decision-making process. In practice, the results of repeated requirement assessments are typically presented in a structured manner, such as displaying similarity scores, key matching parameter comparisons, summaries of historical solutions, and applicability assessments simultaneously on the interface. This design helps users quickly understand the matching logic. For example, for a requirement related to abnormal equipment temperature control, the system might simultaneously provide three historical records with different similarities, indicating their correlation with ambient temperature, load curves, or sensor data. Users can then decide whether to directly adopt the historical solution or further optimize and adjust it based on this information.

[0043] In this embodiment, user-uploaded free text requests are standardized and parsed to extract key entities and construct unified structured information. A domain-knowledge-enhanced semantic model is used to perform deep retrieval in the historical request database, comprehensively comparing technical parameters such as device features and fault phenomena to calculate a multi-dimensional similarity score. Finally, a preset threshold is used to determine the repetition of requests, and the user is fed back with a judgment result containing detailed matching criteria. This solution achieves intelligent request management through natural language processing and machine learning technologies.

[0044] In summary, this technical solution transforms unstructured requirement text into standardized semantic representations, eliminating matching difficulties caused by differences in human expression and improving the accuracy of requirement analysis. The deep semantic retrieval mechanism can identify the technical commonalities behind different expressions, ensuring that relevant historical experience is fully explored. Structured feedback information helps users quickly understand the basis for duplicate judgments, reducing decision-making blind spots when reusing experience, and overall optimizing the efficiency and quality of requirement processing.

[0045] Please refer to Figure 2 , Figure 2 This is the overall flowchart of the deduplication method in the first embodiment of this method.

[0046] As shown in the figure, the process first performs structured processing on the user-input requirement text. This includes labeling and extracting key information, transforming the requirement text into a standardized data format that can be processed by machines. Then, a semantic retrieval engine driven by a large language model extracts potential similar items from the historical requirement database, generating a candidate set after context-related optimization and sorting. Next, a comprehensive score is generated through a two-dimensional analysis of precise character sequence comparison and semantic vector similarity calculation, combined with preset weights. Finally, the system automatically performs duplicate detection or adds new items to the database based on the score threshold, and provides the user with interpretable conclusions including functional similarity and field overlap analysis. This method achieves high-precision duplicate detection and knowledge base self-optimization for manufacturing technology requirements through a dual mechanism of large-scale model semantic understanding and structured rule verification.

[0047] Please refer to Figure 3 , Figure 3 This is a schematic diagram illustrating the application of the model in the first embodiment of this method.

[0048] As is understood, the demand deduplication method of this invention starts with the user inputting the original demand text. First, it performs entity recognition and vectorization encoding through the structured parsing module, outputting structured data with entity annotations. Then, in the retrieval module, it first coarsely screens candidate demands from the historical database based on the Approximate Nearest Neighbor (ANN) algorithm, and then performs semantic re-ranking through a large RAG-enhanced model (such as the GPT-3.5 architecture) to obtain the Top-3 similar historical demands. Subsequently, it enters the similarity calculation module, which calculates character-level (Jaccard improved algorithm) and semantic-level (Sentence-BERT cosine similarity) scores respectively, and dynamically fuses them according to preset weights to generate a comprehensive similarity score. Finally, it combines hard / soft thresholds to output the results and provides feedback to the user through a visual report.

[0049] In one feasible embodiment, the user-uploaded requirement text is processed into structured requirement information containing requirement entity annotations, including: extracting keywords from the requirement text to obtain a keyword set; matching the keyword set with preset entity types to obtain requirement entity annotations; and logically combining the requirement entity annotations with predefined association rules to obtain structured requirement information.

[0050] Keyword extraction refers to using natural language processing technology to identify core words with business value from the original requirement text, such as equipment component names and fault phenomenon descriptions, forming a vocabulary set that reflects the essential characteristics of the requirement. Requirement entity annotation involves matching keywords with predefined equipment fault domain entity types. For example, "main shaft" is categorized as an equipment component entity, and "overheating" is categorized as a fault type entity. These entities are logically combined using predefined association rules to generate structured information containing dimensions such as the main equipment, problem description, and severity.

[0051] Understandably, this method effectively solves the semantic ambiguity problem caused by colloquial expressions. Diverse descriptions of the same problem from different people, such as "abnormal bearing noise" and "spindle noise," can all be mapped to the standard problem definition of abnormal bearing vibration after structured processing. This approach preserves the original semantics of the requirements while achieving standardized expression at the business level, providing a reliable data foundation for subsequent requirement deduplication and analysis.

[0052] Based on the first embodiment of the present invention, in the second embodiment of the present invention, the content that is the same as or similar to that in the first embodiment can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 4 In the method for deduplication of demand information based on a large model, step S30 includes steps S301 to S303: Step S301: Calculate the character-level similarity between the character sequence of the structured requirement information and the character sequences of each similar historical requirement to obtain the character-level similarity score corresponding to each similar historical requirement.

[0053] The core of character-level similarity calculation lies in capturing subtle differences in technical terms through character-by-character matching. For example, it can effectively identify correlations even when there is a mix of traditional and simplified Chinese characters or regional spelling variations in the expression. The calculation process considers the specificity of technical terms, giving special treatment to key fields such as equipment component names, ensuring that even technical terms with only partial character differences receive a reasonable evaluation. This method is particularly suitable for handling common homophones or abbreviations in user input, and can, to some extent, bridge the semantic gap in natural language processing.

[0054] Understandably, character-level matching aims to construct a similarity benchmark for surface-level text features. Its greatest advantage lies in identifying requirements with highly similar expressions. When engineers describe equipment malfunctions using similar terminology, this method maintains stable scores even with minor spelling differences. In practical applications, this method can discover similar expression patterns of the same technical problem by different people. For example, recording the same malfunction as "abnormal operating noise" and "abnormal sound during operation" may have slight differences in individual characters, but they essentially point to the same equipment problem. This matching method provides a foundational reference for subsequent deeper semantic analysis.

[0055] Step S302: Perform semantic similarity calculation based on the semantic vector representation of the structured demand information and the semantic vector representation of each similar historical demand to obtain a semantic similarity score corresponding to each similar historical demand.

[0056] Semantic vector representation refers to converting technical requirement text into numerical features in a high-dimensional space using deep learning models. Unlike character-level surface matching, semantic similarity calculation can identify content with the same technical connotation under different expressions. For example, although "motor over-temperature protection" and "motor overheating trip" are significantly different in wording, they express the same essential equipment fault. Semantic vector comparison can effectively reveal this correlation. This method relies on a pre-trained language model, fine-tuned on fault description text in the manufacturing field, enabling it to accurately understand the unique expressions related to various technical terms in this industry.

[0057] Understandably, semantic similarity scoring focuses more on the essence of technical issues, enabling the system to overcome the limitations of traditional keyword matching. In practice, equipment malfunctions often have multiple ways of being described; senior engineers may use technical terms, while on-site operators may use colloquial descriptions. Semantic vector methods can determine similarity through vector distance in the semantic space.

[0058] It should be understood that semantic similarity calculation still needs to be used in conjunction with other methods. In cases where certain technical terms are highly standardized, relying solely on semantic matching may be overgeneralized, in which case character-level matching needs to be combined to provide constraints. This multi-level, multi-angle similarity evaluation system ensures both the depth of semantic understanding and the rigor of technical judgment.

[0059] Step S303: Based on the preset weight configuration, the character-level similarity score and semantic-level similarity score corresponding to each similar historical requirement are weighted and fused to obtain a comprehensive similarity score.

[0060] Weight configuration refers to the system's method of adjusting the proportion of character-level and semantic-level scores according to business needs. Its purpose is to balance the relative importance of text form matching and deep semantic matching.

[0061] Understandably, this method, by dynamically adjusting the fusion ratio of the two scoring methods, retains the advantage of character-level comparison and recognition of precise matching of technical terms, while also leveraging the strength of semantic-level analysis in discovering potential relationships. For example, when dealing with highly standardized equipment parameter descriptions, the weight of character-level analysis can be appropriately increased to ensure accurate correspondence with technical specifications. When there are cases in historical examples where the terminology is not standardized but semantically similar to the current requirements, this method can arrive at an objective evaluation through reasonable weight allocation.

[0062] This embodiment combines character-level matching and semantic-level analysis to compare and calculate structured requirement information with historical requirements. After adjusting the two scores with preset weights, a comprehensive similarity score is generated and output in a sorted order. Character-level calculation focuses on the precise matching of technical terms and can identify variations in expression; semantic-level calculation relies on a deep learning model to capture the deep technical connotations of the text; the weight configuration mechanism flexibly balances the strength of the two to adapt to the requirements of different application scenarios.

[0063] In summary, this embodiment balances the accuracy of technical text with the flexibility of semantic understanding, effectively addressing the limitations of single matching methods when dealing with terminology variations or differences in expression. Secondly, through dynamic weight adjustment, the system can adapt to the dual needs of highly standardized technical parameters and non-standard colloquial descriptions, expanding the applicability of historical case matching. The resulting comprehensive scoring system ensures both strict correspondence of technical terms and the discovery of potentially related cases, providing technical personnel with more comprehensive decision-making support, thereby improving the accuracy and efficiency of fault diagnosis and experience reuse.

[0064] Please refer to Figure 5 In the demand information duplication method based on a large model, the demand duplication determination result is obtained according to the similarity score, and the demand duplication determination result is fed back to the user, including steps A11~A13: Step A11: If the similarity score is less than the preset threshold, the semantic feature vector required at the current time is obtained based on the semantic similarity score.

[0065] This step uses a preset threshold to automate the deduplication of requirements. When the score exceeds the predetermined judgment benchmark, the duplicate judgment process is triggered. After determining that the current requirement is not a duplicate, the semantic feature vector of the requirement needs to be further analyzed.

[0066] Step A12: Obtain the structured encoding of the required text based on the character-level similarity score.

[0067] Understandably, once it's confirmed that the user-submitted requirement information does not overlap with existing requirements in the database, the system still needs to add the requirement to the database. After confirming this, the system will activate a deep semantic analysis module to extract features from the requirement text at multiple levels. This process primarily relies on a pre-trained language model to convert unstructured information such as fault phenomena and equipment parameters described in natural language into vectorized representations in a high-dimensional space, thereby preserving the key semantic features of the original requirement.

[0068] Understandably, by training the large model with manufacturing terminology, equipment parameters, and failure modes, the vector can accurately reflect the technical features and potential relationships hidden in the requirements document.

[0069] Step A13: Based on the semantic feature vector and structured encoding, perform the database entry operation for similar historical requirements and send a database entry completion notification to the user.

[0070] Semantic feature vectors refer to the high-dimensional feature representation of requirement text extracted through deep learning models, which can reflect professional and technical semantics such as equipment failures and process parameters; structured coding is a standardized classification and identification of requirements according to the company's established standards. The combination of the two not only preserves the technical details of the requirements, but also meets the standardized requirements of database management, making subsequent requirement retrieval and analysis more accurate and efficient.

[0071] Understandably, storing new requirement information, after semantic feature analysis and structured coding, in a standardized format in the enterprise's historical requirement database ensures that subsequent new data can be intelligently linked to historical data. Once the data entry operation is complete, the system automatically triggers a notification mechanism, providing feedback on the processing result to the requirement submitter through a pre-defined message channel. This closed-loop design guarantees the traceability of the entire requirement process.

[0072] In one feasible embodiment, the step of obtaining the demand duplication determination result based on the similarity score also includes steps A21 to A24.

[0073] Step A21: If the similarity score is greater than the preset threshold, a duplicate demand determination conclusion is generated and fed back to the client terminal.

[0074] When the similarity score exceeds a pre-set threshold, the system automatically generates a report indicating duplicate requirements and sends it back to the client terminal. This report includes not only the basis for comparing duplicate requirements but also an index of relevant historical solutions. This information is accurately and promptly pushed to the user through the feedback interface provided by the client terminal. This process ensures rapid processing of requirements, improves solution reusability, and avoids redundant analysis and resource waste.

[0075] Step A22: Based on the semantic similarity score, obtain a description of the functional intent similarity between the current requirement and similar historical requirements.

[0076] This step, functional intent similarity description, refers to explaining the degree of alignment between current and historical requirements in terms of core objectives and technical solutions, based on semantic similarity scoring. Specifically, it utilizes natural language processing technology to perform deep semantic analysis on the requirement text, generating a similarity report between the current and historical requirements. This report is generated according to a specific template structure, including several pre-defined sections such as a requirement overview, core objectives, and solution comparison. This facilitates users' quick understanding and comparison of the similarities and differences between the two requirements.

[0077] Understandably, by using natural language processing technology to deeply analyze the technical issues and expected business effects behind the requirement text, it is possible to determine whether the user-uploaded requirement text and similar historical requirements from the preceding steps are essentially seeking the same solution. Therefore, this explanation of functional intent similarity clarifies to users the logic behind grouping these requirements together, providing an easily understandable basis for users uploading requirements to understand the judgment results and avoid redundant development.

[0078] Step A23: Based on the character-level similarity score, obtain the field overlap analysis description of the required text.

[0079] In addition to explanations of functional intent similarity, explanations of character-level similarity are also needed to form a complete set of judgment criteria. Character-level similarity, as a basic and direct text comparison method, focuses on quantifying the overlap ratio of two texts at the character sequence level, such as by comparing whether keywords or specific technical terms are used in a common and consistent manner.

[0080] Understandably, this field overlap analysis, as a supplement to the functional intent similarity analysis, focuses on the surface features of the text and can effectively capture significant fields to provide preliminary and objective textual evidence. This clearly shows users the similarity clues in the specific expression of the requirements document, helping them to intuitively understand the basis for the duplication judgment.

[0081] Step A24: Based on the functional intent similarity description and field overlap analysis description, generate a requirement duplication confirmation consultation message and send it to the user terminal.

[0082] Understandably, presenting the field overlap analysis and functional intent similarity explanations to users together creates a complete chain of argumentation, from form to substance, and from surface to depth. This provides both intuitive evidence based on textual surface features and a deeper explanation based on semantic understanding, making the final duplicate determination more comprehensive and facilitating accurate confirmation and decision-making for users.

[0083] It should be understood that "repeated confirmation of consultation information" refers to sending the same conclusions and supporting evidence as the user's uploaded requirements to the customer, and then soliciting and obtaining the user's feedback on the automated judgment result after the customer is aware of the conclusion.

[0084] In this embodiment, when the system determines that a new requirement does not overlap with a historical requirement, it automatically extracts semantic features and performs structured encoding, then stores the new requirement in the database. When it determines that a requirement is duplicated, it generates a duplicate confirmation consultation message containing explanations of functional intent similarity and field overlap analysis, and feeds this message, along with the determination conclusion, back to the user terminal to solicit the user's approval. This method achieves a comprehensive comparison from surface-level text features to deep semantics, forming a complete chain of judgment criteria.

[0085] In summary, this embodiment, by combining character-level and semantic-level similarity analysis, not only achieves accurate identification of duplicate requirements but also automatically standardizes and stores new requirements after they are determined to be non-duplicate. This process provides dual explanations of functional intent and field overlap, making the judgment criteria transparent and verifiable, significantly improving user understanding and trust in the automated judgment results. Simultaneously, the system proactively solicits user feedback after judgment, forming a closed-loop management process, thereby effectively assisting users in decision-making and improving the accuracy and efficiency of requirement management.

[0086] The above examples are only for understanding this application and do not constitute a limitation on the demand information deduplication method based on the large model in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0087] This application also provides a requirement information deduplication device based on a large model. Please refer to... Figure 6 The demand information deduplication device based on the large model includes: Text processing module 10 is used to perform structured processing on the user-uploaded requirement text to obtain structured requirement information containing requirement entity annotations; The demand retrieval module 20 is used to perform near nearest neighbor retrieval in a similar historical demand vector library based on structured demand information and through a large language model with integrated retrieval enhancement generation technology, and to reorder the retrieval results based on the relevance of context semantics in order to filter out several similar historical demands. The similarity comparison module 30 is used to compare the structured requirement information with each similar historical requirement. By calculating the character-level similarity score and the semantic-level similarity score, and by weighting and fusing the character-level similarity score and the semantic-level similarity score according to the preset weight configuration, a comprehensive similarity score corresponding to each similar historical requirement is obtained. The analysis conclusion feedback module 40 is used to obtain the demand duplication determination result based on the similarity score, and to feed back the demand duplication determination result to the user.

[0088] In one embodiment, the text processing module 10 is further configured to extract keywords from the requirement text to obtain a keyword set; match the keyword set with a preset entity type to obtain requirement entity annotations; and logically combine the requirement entity annotations with predefined association rules to obtain structured requirement information.

[0089] In one embodiment, the demand retrieval module 20 is further configured to generate a query vector based on structured demand information to obtain a vectorized query request; perform an approximate nearest neighbor search in a similar historical demand vector library based on the vectorized query request to obtain a preliminary search result set; reorder the preliminary search result set based on contextual semantic relevance using a large language model to obtain optimized search results; and select a preset number of similar historical demands based on the ranking score of the optimized search results.

[0090] In one embodiment, the similarity comparison module 30 is further configured to perform character-level similarity calculation based on the character sequence of the structured requirement information and the character sequence of each similar historical requirement to obtain a character-level similarity score corresponding to each similar historical requirement; perform semantic-level similarity calculation based on the semantic vector representation of the structured requirement information and the semantic vector representation of each similar historical requirement to obtain a semantic-level similarity score corresponding to each similar historical requirement; and perform weighted fusion of the character-level similarity score and the semantic-level similarity score corresponding to each similar historical requirement according to a preset weight configuration to obtain a comprehensive similarity score.

[0091] In one embodiment, the analysis conclusion feedback module 40 is further configured to, if the comprehensive similarity score is less than a preset threshold, obtain the semantic feature vector of the current requirement based on the semantic level similarity score; obtain the structured encoding of the requirement text based on the character level similarity score; and perform the entry operation of the similar historical requirement database based on the semantic feature vector and the structured encoding, and send an entry completion notification to the user.

[0092] In one embodiment, the analysis conclusion feedback module 40 is further configured to generate a demand duplication determination conclusion and feed it back to the client terminal if the comprehensive similarity score is greater than a preset threshold; obtain a functional intent similarity description between the current demand and similar historical demands based on the semantic similarity score; obtain a field overlap analysis description of the demand text based on the character similarity score; and generate demand duplication confirmation consultation information and send it to the user terminal based on the functional intent similarity description and the field overlap analysis description.

[0093] The large-model-based demand information deduplication device provided in this application, employing the large-model-based demand information deduplication method described in the above embodiments, can solve the technical problem of accurately identifying repetitive improvement needs in intelligent manufacturing scenarios and achieving efficient reuse of production optimization solutions. Compared with the prior art, the beneficial effects of the large-model-based demand information deduplication device provided in this application are the same as those of the large-model-based demand information deduplication method described in the above embodiments, and other technical features in the large-model-based demand information deduplication device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0094] This application provides a large-model-based demand information deduplication device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the large-model-based demand information deduplication method in the above embodiment 1.

[0095] The following is for reference. Figure 7 This diagram illustrates a structural schematic suitable for implementing a large-model-based demand information deduplication device in the embodiments of this application. The large-model-based demand information deduplication device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 7 The demand information deduplication device based on a large model shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0096] like Figure 7As shown, the demand information deduplication device based on a large model may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the demand information deduplication device based on the large model. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the large-model-based demand information deduplication device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows a large-model-based demand information deduplication device with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0097] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0098] The large-model-based demand information deduplication device provided in this application, employing the large-model-based demand information deduplication method described in the above embodiments, can solve the technical problem of accurately identifying repetitive improvement needs in intelligent manufacturing scenarios and achieving efficient reuse of production optimization solutions. Compared with the prior art, the beneficial effects of the large-model-based demand information deduplication device provided in this application are the same as those of the large-model-based demand information deduplication method provided in the above embodiments, and other technical features of this large-model-based demand information deduplication device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0099] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0100] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0101] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the large-model-based demand information deduplication method in the above embodiments.

[0102] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0103] The aforementioned computer-readable storage medium may be included in a demand information deduplication device based on a large model; or it may exist independently and not be assembled into a demand information deduplication device based on a large model.

[0104] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by a large-model-based demand information deduplication device, the large-model-based demand information deduplication device performs the following: structuring the user-uploaded demand text to obtain structured demand information; searching a similar historical demand database based on the structured demand information to obtain several similar historical demands; calculating the similarity between the structured demand information and the similar historical demands to obtain a similarity score; obtaining a demand duplication determination result based on the similarity score, and feeding back the demand duplication determination result to the user.

[0105] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via an Internet service provider through an Internet connection).

[0106] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0107] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0108] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the aforementioned large-model-based demand information deduplication method. This solves the technical problem of accurately identifying repetitive improvement needs in intelligent manufacturing scenarios and achieving efficient reuse of production optimization solutions. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the large-model-based demand information deduplication method provided in the above embodiments, and will not be elaborated upon here.

[0109] The computer program product provided in this application can solve the technical problem of demand information deduplication based on large models. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the demand information deduplication method based on large models provided in the above embodiments, and will not be repeated here.

[0110] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application through the content of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for deduplication of demand information based on a large model, characterized in that, The method for deduplication of demand information based on a large model includes: The user-uploaded requirement text is processed into a structured requirement information containing requirement entity annotations. Based on the structured demand information, a large language model with integrated retrieval enhancement generation technology is used to perform near nearest neighbor retrieval in a similar historical demand vector library, and the retrieval results are reordered based on the relevance of context semantics to select several similar historical demands. The structured requirement information is compared with each of the similar historical requirements. Character-level similarity scores and semantic-level similarity scores are calculated, and the character-level similarity scores and semantic-level similarity scores are weighted and fused according to a preset weight configuration to obtain a comprehensive similarity score corresponding to each similar historical requirement. Based on the comprehensive similarity score, the result of the demand duplication determination is obtained.

2. The method for deduplication of demand information based on a large model according to claim 1, characterized in that, The step of structuring the user-uploaded requirement text to obtain structured requirement information containing requirement entity annotations includes: Extract keywords from the required text to obtain a keyword set; Based on the keyword set, a preset entity type is matched to obtain the required entity annotation; The structured requirement information is obtained by logically combining the required entity annotations with predefined association rules.

3. The method for deduplication of demand information based on a large model according to claim 1, characterized in that, Based on the structured demand information, a large language model integrating retrieval enhancement generation technology is used to perform an approximate nearest neighbor search in a similar historical demand vector database. The search results are then reordered based on contextual semantic relevance to select several similar historical demands, including: A query vector is generated based on the structured requirement information to obtain a vectorized query request; Based on the vectorized query request, an approximate nearest neighbor search is performed in the similar historical demand vector database to obtain a preliminary search result set; The preliminary search result set is reordered based on contextual semantic relevance using the large language model to obtain optimized search results. Based on the ranking score of the optimized search results, a preset number of similar historical requests are selected.

4. The method for deduplication of demand information based on a large model according to claim 1, characterized in that, The step of calculating character-level similarity scores and semantic-level similarity scores, and then weighting and fusing the character-level similarity scores and semantic-level similarity scores according to a preset weight configuration, includes: Based on the character sequence of the structured requirement information and the character sequence of each similar historical requirement, a character-level similarity score is calculated to obtain the character-level similarity score corresponding to each similar historical requirement. Semantic similarity is calculated based on the semantic vector representation of the structured demand information and the semantic vector representation of each similar historical demand to obtain a semantic similarity score corresponding to each similar historical demand. Based on the preset weight configuration, the character-level similarity score and semantic-level similarity score corresponding to each similar historical requirement are weighted and fused to obtain the comprehensive similarity score.

5. The method for deduplication of demand information based on a large model according to claim 4, characterized in that, After obtaining the demand duplication determination result based on the comprehensive similarity score, the process includes: If the overall similarity score is less than a preset threshold, then the semantic feature vector required at the current time is obtained based on the semantic similarity score. Based on the character-level similarity score, the structured encoding of the required text is obtained; Based on the semantic feature vector and structured encoding, perform the database entry operation for similar historical requirements and send a database entry completion notification to the user.

6. The method for deduplication of demand information based on a large model according to claim 4, characterized in that, After obtaining the demand duplication determination result based on the comprehensive similarity score, the process further includes: If the overall similarity score is greater than a preset threshold, a duplicate demand determination conclusion is generated and fed back to the client terminal; Based on the semantic similarity score, a description of the functional intent similarity between the current requirement and similar historical requirements is obtained; Based on the character-level similarity score, an analysis of the field overlap of the required text is obtained; Based on the functional intent similarity description and field overlap analysis description, a request for repeated confirmation is generated and sent to the user terminal.

7. A demand information deduplication device based on a large model, characterized in that, The demand information deduplication device based on the large model includes: The text processing module is used to process the user-uploaded requirement text into a structured form, resulting in structured requirement information containing requirement entity annotations. The demand retrieval module is used to perform near nearest neighbor retrieval in a similar historical demand vector library based on the structured demand information and a large language model with integrated retrieval enhancement generation technology, and to reorder the retrieval results based on the relevance of context semantics in order to filter out several similar historical demands. The similarity comparison module is used to compare the structured requirement information with each of the similar historical requirements. By calculating the character-level similarity score and the semantic-level similarity score, and by weighting and fusing the character-level similarity score and the semantic-level similarity score according to the preset weight configuration, a comprehensive similarity score corresponding to each similar historical requirement is obtained. The analysis conclusion feedback module is used to obtain the demand duplication determination result based on the similarity score, and to feed back the demand duplication determination result to the user.

8. A demand information deduplication device based on a large model, characterized in that, The large-model-based demand information deduplication device includes: a memory, a processor, and a large-model-based demand information deduplication program stored in the memory and executable on the processor, wherein the large-model-based demand information deduplication program is configured to implement the large-model-based demand information deduplication method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium stores a requirement information deduplication program based on a large model, which, when executed by a processor, implements the requirement information deduplication method based on a large model as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the demand information deduplication method based on a large model as described in any one of claims 1 to 6.

Images