Methods and apparatus for searching and comparing technical solutions

Abstract

This application discloses a method and apparatus for searching and comparing technical solutions. The method includes: acquiring a technical solution input by a user and extracting a set of keywords corresponding to the technical solution; constructing a search strategy based on the keyword set, selecting multiple relevant documents from a preset literature database according to the search strategy, calculating a quality index corresponding to the search strategy based on the selected multiple relevant documents, updating the search strategy based on the quality index, until the quality index meets preset conditions; determining multiple target documents from multiple relevant documents based on the technical elements corresponding to the technical solution; generating search comparison information based on the multiple target documents, and displaying the search comparison information. Thus, this application can effectively solve the problems of low efficiency and high false negative rate of existing search solutions, realize the automation of the search comparison process, and improve the generation quality of search comparison information of technical solutions.

Description

Technical Field

[0001] This application relates to the field of information retrieval technology, and in particular to a retrieval and comparison method and apparatus for a technical solution. Background Technology

[0002] In the existing technology, the search for technical solutions relies heavily on the manual experience of patent search practitioners. Patent search practitioners need to manually construct search strategies, screen patent documents, and compare technical features. The entire search and comparison process is time-consuming and costly, and there is a risk of missed detection due to personal knowledge limitations and fatigue. Summary of the Invention

[0003] This application provides a method and apparatus for searching and comparing technical solutions, which can effectively solve the problems of low efficiency and high false negative rate in the prior art, realize the automation of the search and comparison process of technical solutions, and improve the generation quality of search and comparison information of technical solutions.

[0004] Firstly, this application provides a method for searching and comparing technical solutions, applied to a search and comparison system, the method comprising: Obtain the technical solution input by the user, and extract the keyword set corresponding to the technical solution; A retrieval strategy is constructed based on the keyword set, and multiple relevant documents are selected from a preset literature database according to the retrieval strategy. The quality index corresponding to the retrieval strategy is calculated based on the selected multiple relevant documents, and the retrieval strategy is updated based on the quality index until the quality index meets the preset conditions. Based on the technical elements corresponding to the technical solution, multiple target documents are identified from the multiple related documents; Search comparison information is generated based on the multiple target documents, and the search comparison information is displayed.

[0005] In some embodiments, the quality metrics include recall and precision, and the calculation of the quality metrics corresponding to the retrieval strategy based on the selected plurality of relevant documents includes: The following steps are performed using the pre-trained iterative analysis module: A pre-search is performed on the literature database to obtain a seed literature set corresponding to the technical solution; Based on the seed document set, a complete retrieval is performed on the multiple related documents to obtain the recall rate corresponding to the retrieval strategy; Based on the iterative analysis module, precision retrieval is performed on the multiple relevant documents to obtain the precision rate corresponding to the retrieval strategy.

[0006] In some embodiments, the pre-retrieval includes the following steps: Based on the technical field to which the technical solution belongs, target owners whose ranking of the number of technical documents meets preset requirements are determined from the owners corresponding to the multiple related documents; The target owner's multiple technical documents in the technical field are used as a seed document set.

[0007] In some embodiments, the complete search includes the following steps: Calculate the number of intersections between the multiple related documents and the seed document set; The ratio of the number of intersections to the total number of documents in the seed document set is used as the recall rate of the retrieval strategy.

[0008] In some embodiments, the precision retrieval includes the following steps: Multiple document samples were identified from the aforementioned multiple related documents; Using the preset model within the iterative analysis module, the multiple document samples are semantically compared with the technical solution to obtain comparison results characterizing whether the document samples and the technical solution are related. Based on the comparison results, the proportion of relevant documents in the multiple document samples is determined, and the proportion is used as the precision of the retrieval strategy.

[0009] In some embodiments, the quality metrics include recall and precision, and updating the retrieval strategy based on the quality metrics includes: Based on the comparison results of the recall rate and the precision rate with the preset threshold, the retrieval strategy is adjusted. The adjustment process includes at least one of the following: When the recall rate is lower than a first preset threshold, a broadening strategy is executed to expand the search scope; When the precision rate is lower than the second preset threshold, a narrowing strategy is executed to reduce the search scope; The broadening strategy and the narrowing strategy are implemented by modifying the keywords, classification numbers or logical operators of the search query in the search strategy.

[0010] In some embodiments, extracting the keyword set corresponding to the technical solution includes: The technical feature set of the technical solution is extracted using a pre-trained feature extraction module; The keyword set corresponding to the technical feature set is extracted using a pre-trained keyword expansion module.

[0011] In some embodiments, extracting the set of technical features of the technical solution includes: Semantic analysis is performed on the text content of the technical solution to identify multiple technical elements contained in the technical solution; Based on preset technical feature classification rules, the multiple technical elements are classified into technical feature features, necessary technical features, and optional technical features. The set of technical features is generated based on the classification results, wherein the set of technical features includes at least one technical field feature and at least one essential technical feature.

[0012] In some embodiments, extracting the keyword set corresponding to the set of technical features includes: Each technical feature in the set of technical features is semantically expanded to generate at least one of the following: a set of synonyms, a set of near-synonyms, and a set of hyponyms and hypernyms corresponding to each technical feature. Based on the technical fields associated with the set of technical features, at least one international patent classification number is recommended from a preset classification number library; The set of words that have undergone semantic expansion processing is combined with the at least one international patent classification number to form the set of keywords.

[0013] In some embodiments, determining multiple target documents from the multiple related documents based on the technical elements corresponding to the technical solution includes: The pre-trained preliminary screening module is used to perform preliminary screening of the multiple relevant documents; By using a pre-trained technical feature comparison module, the technical features of the documents that have passed the initial screening are compared to identify multiple target documents.

[0014] In some embodiments, where the document is a patent, the preliminary screening includes the following steps: The multiple relevant documents are initially screened according to preset screening criteria, which include at least one of the following: International Patent Classification (IPC), applicant, legal status, application date, and patent type.

[0015] In some embodiments, the technical feature comparison includes the following steps: Technical feature comparison analysis is performed on the documents that have passed the initial screening. The set of technical features of the technical solution is compared one by one with the technical features recorded in the claims of each document. Based on the comparison results, documents with a technical feature matching degree higher than a preset threshold are identified as target documents.

[0016] In some embodiments, the preliminary screening of the plurality of relevant documents further includes: Receive intermediate files from the retrieval strategy stage, the intermediate files including retrieval result data of the plurality of relevant documents; The preliminary screening is performed on the literature data in the intermediate file based on preset screening conditions or configurable screening instructions. The configurable filtering instructions are generated based on the analysis results of the intermediate files.

[0017] In some embodiments, the technical feature comparison of the documents that have undergone the initial screening further includes: Receive intermediate files from the initial screening stage, the intermediate files including a set of documents that have passed the initial screening; Based on preset comparison rules or configurable comparison instructions, technical features are compared on the literature data in the intermediate files. The configurable comparison instructions are generated based on the analysis results of the document claims in the intermediate file.

[0018] In some embodiments, the search comparison information is an infringement search report; The infringement search report includes: a comparison table of the claims of the technical solution and the multiple target documents, and an infringement risk assessment conclusion based on the comparison results.

[0019] Secondly, this application provides a retrieval and comparison device for use in a retrieval and comparison system, the retrieval and comparison device comprising: The feature processing module is used to acquire the technical solution input by the user and extract the keyword set corresponding to the technical solution; The iterative analysis module is used to construct a retrieval strategy based on the keyword set, select multiple relevant documents from a preset literature database according to the retrieval strategy, calculate the quality index corresponding to the retrieval strategy based on the selected multiple relevant documents, and update the retrieval strategy based on the quality index until the quality index meets the preset conditions. The patent screening module is used to identify multiple target documents from the multiple relevant documents based on the technical elements corresponding to the technical solution. The report generation module is used to generate search comparison information based on the multiple target documents and to display the search comparison information.

[0020] In some embodiments, the feature processing module includes a feature extraction module and a keyword expansion module; when extracting the keyword set corresponding to the technical solution: The feature extraction module is used to extract a set of technical features of the technical solution; The keyword expansion module extracts the keyword set corresponding to the set of technical features.

[0021] As can be seen from the above, the technical solution retrieval and comparison method and apparatus provided in this application automatically extracts the keyword set of the technical solution to construct a retrieval strategy, and dynamically optimizes the retrieval strategy based on quality indicators until it meets the preset conditions. At the same time, it combines technical elements to screen target documents and uses a large language model to realize intelligent patent screening and technical feature comparison. This can solve the problems of traditional methods relying on human experience, long retrieval time, high cost and risk of missed detection. It can automate the retrieval and comparison process and improve the generation quality of technical solution retrieval and comparison information.

[0022] Of course, any product implementing this application does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A flowchart of a retrieval and comparison method for technical solutions provided in the embodiments of this application.

[0025] Figure 2 A schematic diagram of the overall architecture of the retrieval and comparison method for the technical solutions provided in the embodiments of this application.

[0026] Figure 3 This is a schematic diagram illustrating feature processing in the retrieval and comparison method of the technical solutions provided in the embodiments of this application.

[0027] Figure 4 This is a schematic diagram illustrating iterative analysis in the retrieval and comparison method of the technical solutions provided in the embodiments of this application.

[0028] Figure 5 An end-to-end flowchart of the retrieval and comparison method for the technical solutions provided in the embodiments of this application.

[0029] Figure 6 A schematic block diagram of a retrieval and comparison device for the technical solutions provided in the embodiments of this application.

[0030] Figure 7 A schematic block diagram of a control device provided in an embodiment of this application. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0032] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0033] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0034] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0035] In the existing technology, the search for technical solutions relies heavily on the manual experience of patent search practitioners. Patent search practitioners need to manually construct search strategies, screen patent documents, and compare technical features. The entire search and comparison process is time-consuming and costly, and there is a risk of missed detection due to personal knowledge limitations and fatigue.

[0036] Based on this, this application provides a method and apparatus for searching and comparing technical solutions, which can effectively solve the problems of low efficiency and high false negative rate in the prior art, realize the automation of the search and comparison process of technical solutions, and improve the generation quality of search and comparison information of technical solutions.

[0037] Therefore, refer to Figure 1 , Figure 1 A flowchart illustrating a retrieval and comparison method for technical solutions provided in embodiments of this application; in some embodiments, this application provides a retrieval and comparison method for technical solutions, applied to a retrieval and comparison system, the method including at least the following steps: Step 101: Obtain the technical solution input by the user and extract the keyword set corresponding to the technical solution; Step 102: Construct a retrieval strategy based on the keyword set, select multiple relevant documents from the preset literature database according to the retrieval strategy, and calculate the quality index corresponding to the retrieval strategy based on the selected multiple relevant documents, so as to update the retrieval strategy based on the quality index until the quality index meets the preset conditions. Step 103: Based on the technical elements corresponding to the technical solution, identify multiple target documents from multiple relevant documents; Step 104: Generate search comparison information based on multiple target documents and display the search comparison information.

[0038] Based on steps 101 to 104 above, this application can automatically extract the keyword set of technical solutions to construct a search strategy, and dynamically optimize the search strategy based on quality indicators until it meets the preset conditions. At the same time, it can screen target documents by combining technical elements and realize intelligent patent screening and technical feature comparison by using a large language model. This can solve the problems of traditional methods relying on human experience, long search time, high cost and risk of missed detection. It can automate the search comparison process and improve the quality of generated search comparison information of technical solutions.

[0039] For ease of understanding, the following explains some key terms in this embodiment: A technical solution can refer to a collection of technical features proposed to solve a certain technical problem. It can be presented in the form of text, diagrams or other forms, and carries specific technical content and innovative points.

[0040] A keyword set refers to a set of representative words extracted or expanded from a technical solution. These words can accurately summarize the core content of the technical solution and serve as the basis for constructing a search strategy.

[0041] A retrieval strategy is a set of rules or expressions designed to accurately find the required information from a large amount of literature. It can consist of keywords, classification numbers, logical operators, etc., and is used to guide the retrieval process.

[0042] A literature database is a collection that stores a large amount of technical literature, such as patent databases, academic paper databases, and industry report databases, providing a data source for the retrieval and comparison of technical solutions.

[0043] Relevant literature refers to literature selected from literature databases through retrieval strategies that has a certain technical relevance to the technical solution to be compared. These documents may contain similar technical features or solve similar technical problems.

[0044] Quality metrics are quantitative standards used to evaluate the effectiveness of search strategies, such as recall and precision, which reflect the comprehensiveness and accuracy of search results.

[0045] Technical elements refer to the various components or functional modules that constitute a technical solution, such as technical fields, necessary technical features, and optional technical features. These elements are the basis for comparing technical features.

[0046] Target documents refer to documents that have been further screened and compared among relevant documents and have been identified as being highly relevant to the technical solution to be compared or having specific comparative significance, such as patent documents that may pose an infringement risk.

[0047] Search and comparison information refers to reports or data generated after comparing target documents with technical solutions to be compared. It may include technical feature comparison tables, infringement risk assessment conclusions, etc., to assist users in making decisions.

[0048] Understandably, this method is first designed to acquire a technical solution input by the user and extract the corresponding keyword set. Specifically, users can directly enter a description of the technical solution through a text input box, or provide the technical solution by uploading electronic documents such as technical disclosure documents or draft patent applications. After acquiring the technical solution, its corresponding keyword set can be read and analyzed by human experts who can manually select core words, or high-frequency words can be automatically identified as a preliminary keyword set using text analysis tools based on word frequency statistics.

[0049] Furthermore, a search strategy is constructed based on this keyword set, and multiple relevant documents are selected from a pre-defined literature database according to this strategy. This search strategy can be constructed manually by the user using the keyword set, for example, by combining keywords using Boolean logic operators. Alternatively, the system can provide pre-defined search templates; the user only needs to enter the keywords to generate a preliminary search strategy. The literature database can be a publicly available patent database or an internally maintained technical literature repository. The selection of relevant documents involves executing this search strategy to obtain all search results that meet the criteria.

[0050] Based on this, a quality index corresponding to the retrieval strategy is calculated using the selected relevant literature. The retrieval strategy is then updated based on this quality index until it meets preset conditions. The calculation of this quality index can involve manual sampling and evaluation of the search results. For example, experts can judge the relevance of randomly selected literature to the technical solution to estimate the effectiveness of the retrieval strategy. Based on this quality index, the retrieval strategy will be adjusted. For instance, if manual evaluation finds that the number of search results is insufficient, the search terms can be manually modified to broaden the search scope; if a high proportion of irrelevant literature is found, the search terms can be manually adjusted to narrow the search scope. This updating process will continue until the quality index is manually judged to meet preset accuracy and comprehensiveness requirements.

[0051] Subsequently, based on the technical elements corresponding to the technical solution, multiple target documents were identified from the various related documents. The technical elements of the technical solution can be identified and listed by human experts from the text of the technical solution, such as listing its core components, functions, or application areas. When identifying target documents, human experts can review each relevant document and, based on their understanding of the technical elements of the technical solution, determine the degree of relevance of each document to the technical solution, thereby selecting target documents. For example, experts can compare the technical elements of the technical solution with the title, abstract, or keywords in the claims of a document to decide whether to identify it as a target document.

[0052] Furthermore, based on these multiple target documents, retrieval comparison information is generated and displayed. This retrieval comparison information can be a simple list outlining the titles, abstracts, and publication numbers of all target documents, accompanied by a brief, manually written analysis report explaining the similarities or differences between each target document and the technical solution. This information can be presented as a text report or displayed on a screen through a graphical user interface for user review and understanding.

[0053] This method effectively overcomes the problems of time-consuming, costly, and prone to missed detections in traditional manual retrieval and comparison by introducing automated keyword extraction, iterative optimization of retrieval strategy construction and updating mechanism, and target document screening process based on technical elements. As a result, it improves the efficiency, accuracy, and comprehensiveness of technical solution retrieval and comparison, providing users with a more reliable and convenient tool for technical solution analysis.

[0054] In some embodiments, the training process of the large language model used in this application may include enabling the model to master professional analytical frameworks and legal logic through positive FTO analysis report samples; enabling the model to identify common analytical pitfalls and missed detection patterns through negative patent infringement litigation cases; and enhancing the model's understanding of the boundaries of claim interpretation and the depth of technical feature comparison through patent examination history documents. In terms of training methods, a strategy combining instruction fine-tuning and reinforcement learning based on human feedback is adopted. Patent experts continuously optimize the intermediate results generated by the model to ensure the technical accuracy and legal rigor of its output.

[0055] refer to Figure 2 , Figure 2 This is a schematic diagram of the overall architecture of the retrieval and comparison method for the technical solutions provided in the embodiments of this application. For example... Figure 2As shown, this application starts with the technical solution input by the user, first enters the feature processing stage to complete feature extraction and keyword expansion in sequence to form a keyword set, then goes through the iterative analysis stage to calculate the recall and precision and update the search strategy based on this to obtain multiple relevant documents, then the relevant documents are passed to the patent screening stage for preliminary screening and technical feature comparison to determine multiple target documents, and then the search comparison information can be output through the report generation step to present the fully automated processing logic of the search comparison method of this application from technical solution input to result output.

[0056] In some embodiments, this application proposes updating the retrieval strategy by calculating quality indicators corresponding to the retrieval strategy to ensure the accuracy and completeness of the retrieval results. However, in practical applications, how to specifically define and quantify these quality indicators, and effectively calculate them to guide the optimization of the retrieval strategy, is a key challenge in achieving efficient retrieval comparison. If the calculation method of the quality indicators is not clear enough or lacks sufficient guidance, it will be difficult to accurately evaluate the performance of the retrieval strategy, thereby affecting the iterative optimization effect of subsequent retrieval strategies.

[0057] In response, this application further proposes that the quality indicators in the above method include recall and precision. The quality indicators corresponding to the retrieval strategy are calculated based on multiple relevant documents selected, including: performing the following steps through a pre-trained iterative analysis module: performing a pre-retrieval of documents in the literature database to obtain a seed document set corresponding to the technical solution; performing a recall retrieval of multiple relevant documents based on the seed document set to obtain the recall corresponding to the retrieval strategy; and performing a precision retrieval of multiple relevant documents based on the iterative analysis module to obtain the precision corresponding to the retrieval strategy.

[0058] The quality indicators include recall and precision. Recall refers to the proportion of relevant documents retrieved by the retrieval system from the total number of relevant documents in the database, reflecting the comprehensiveness of the search results, i.e., the degree to which relevant documents are omitted. Precision refers to the proportion of truly relevant documents retrieved by the retrieval system, reflecting the accuracy of the search results, i.e., the degree to which irrelevant documents are included. Using recall and precision as quality indicators allows for a quantitative evaluation of the performance of the retrieval strategy from both comprehensiveness and accuracy dimensions, providing a clear basis for subsequent strategy adjustments.

[0059] The pre-trained iterative analysis module is a collection of artificial intelligence models or algorithms trained and optimized on a large amount of data. Its core function is to perform in-depth analysis and evaluation of literature data. Before being put into use, this module has already learned from a large amount of literature data, mastering the ability to identify literature relevance, extract key information, and evaluate search results. Its implementation methods can include, but are not limited to, deep learning-based natural language processing models, machine learning classifiers, or expert system rule bases. This module can automatically perform complex literature analysis tasks, thereby improving the efficiency and accuracy of quality indicator calculations.

[0060] A pre-search is conducted on literature in the literature database to obtain a seed literature set corresponding to the technical solution. This aims to initially identify core literature highly relevant to the user-input technical solution before the formal search. The "seed literature set" refers to a group of literature closely related to the target technical solution, selected through preliminary, broad searches or based on expert experience. These documents serve as a benchmark for subsequent recall calculations, representing relevant literature that is known or expected to be retrieved. Pre-search can be implemented in various ways, such as using core keywords of the technical solution for initial searching, filtering literature from specific classification numbers or topics based on the technical field information of the technical solution, or searching for publicly available relevant literature based on the applicant or inventor information of the technical solution. This step aims to establish a reliable reference set to accurately evaluate the recall performance of the subsequent search strategy.

[0061] Based on a seed document set, a comprehensive search is performed on multiple relevant documents to obtain the recall rate corresponding to the search strategy. This involves comparing the multiple relevant documents selected by the current search strategy with the seed document set to assess the comprehensiveness of the current search strategy. In essence, recall is calculated by determining how many of the relevant documents retrieved by the current search strategy are included in the seed document set, and by considering the total number of documents in the seed document set. For example, the number of intersections between multiple relevant documents and the seed document set can be counted, and the ratio of this intersection to the total number of documents in the seed document set can be calculated. This step aims to quantify the current search strategy's ability to discover all known relevant documents.

[0062] Based on the iterative analysis module, precision retrieval is performed on multiple relevant documents to obtain the precision rate corresponding to the retrieval strategy. This aims to evaluate how many of the relevant documents selected by the current retrieval strategy are truly relevant to the user's technical solution. This process utilizes the intelligent analysis capabilities of the iterative analysis module to evaluate multiple relevant documents individually or in batches. Specific implementation methods may include: the iterative analysis module calculating the semantic similarity between each relevant document and the user's technical solution to determine their relevance; or the iterative analysis module identifying technical features in the documents and comparing them with the technical features of the user's technical solution to determine whether the documents are truly relevant. This step aims to quantify the current retrieval strategy's ability to exclude irrelevant documents, ensuring the accuracy of the retrieval results.

[0063] By explicitly defining quality indicators as recall and precision, and introducing a pre-trained iterative analysis module to execute specific calculation steps, this application overcomes the problems of vague indicators and low computational efficiency in traditional retrieval strategy evaluation. First, a seed literature set is established through pre-retrieval, providing a reliable benchmark for recall calculation and ensuring an accurate assessment of the comprehensiveness of the retrieval strategy. Second, the iterative analysis module performs precision retrieval, intelligently and efficiently judging the accuracy of search results, avoiding the time-consuming and subjective nature of manual judgment. This explicit and automated method of calculating quality indicators makes the performance evaluation of retrieval strategies more objective and efficient, thereby more accurately guiding the iterative updates of retrieval strategies and ensuring that the selected relevant literature is both comprehensive and accurate.

[0064] In some embodiments, this application further proposes a pre-search step including: determining the target owner whose ranking of the number of technical documents meets preset requirements from multiple relevant document owners based on the technical field to which the technical solution belongs; and using multiple technical documents of the target owner in the technical field as a seed document set.

[0065] Understandably, when determining the seed document set, the first step is to identify target patent holders whose technical document quantity ranking meets preset requirements from among multiple relevant document holders, based on the technical field to which the technical solution belongs. The technical field to which the technical solution belongs can be determined through text analysis, keyword extraction, or classification number identification of the user-input technical solution. For example, natural language processing technology can be used to identify the core concepts in the technical solution and map them to a preset technical field classification system. The patent holders corresponding to multiple relevant documents refer to the entities represented by information such as applicants, inventors, or publishing institutions in the relevant documents initially screened by the search strategy. Determining target patent holders can involve statistically analyzing and ranking the number of technical documents held by these holders within the technical field, and then selecting major participants with high activity and influence in a specific technical field based on preset requirements (e.g., the top 10% of patent holders, or patent holders with more than a certain threshold of technical documents). For example, the number of patents held by each patent holder under a specific international patent classification number can be counted, and the top N patent holders with the largest number can be selected as target patent holders.

[0066] Subsequently, multiple technical documents from the target patent holder within the technical field are used as a seed document set. This means that once a representative target patent holder in a specific technical field is identified, all publicly published technical documents (such as patent documents, academic papers, etc.) within that technical field will be collected to form an initial, high-quality seed document set, which can provide a reliable benchmark for subsequent recall and precision calculations.

[0067] By focusing on influential rights holders within the technical field to which the technical solution pertains, and using their technical documents as a seed document set, this application ensures a higher representativeness of the constructed seed document set. This helps to more accurately reflect the actual document distribution of the technical solution within a specific technical field, thereby making the calculation results of recall and precision more accurate. Based on these more accurate quality indicators, the iterative update process of the search strategy will be more efficient and effective, converging to the optimal search strategy more quickly, and improving the accuracy and efficiency of search comparison.

[0068] In some embodiments, this application further proposes specific steps for recall retrieval, including calculating the number of intersections between multiple relevant documents and the seed document set, and using the ratio of the number of intersections to the total number of documents in the seed document set as the recall rate of the retrieval strategy.

[0069] Understandably, calculating the intersection of multiple relevant documents with the seed document set refers to identifying the number of documents shared by both sets by comparing the selected relevant documents with those in the pre-determined seed document set. The seed document set can represent known documents highly relevant to the technical solution and can serve as a benchmark for evaluating the recall capability of a retrieval strategy. This intersection can be calculated by matching document identifiers (e.g., patent application numbers, publication numbers, document IDs, etc.). For example, the system can traverse two document sets, find all documents present in both sets, and count their number. This step aims to quantify the proportion of known relevant documents covered by the current retrieval strategy.

[0070] Based on this, the ratio of the number of intersections to the total number of documents in the seed document set is used as the recall rate of the retrieval strategy. This means that after obtaining the number of intersections, the total number of documents in the seed document set is further calculated, and then the number of intersections is divided by the total number of documents in the seed document set; the quotient is the recall rate. Recall rate is an important indicator for measuring the recall capability of a retrieval system. It reflects what proportion of documents the retrieval strategy can successfully retrieve from a given set of relevant documents. For example, if the number of intersections is N_intersect and the total number of documents in the seed document set is N_seed, then the recall rate = N_intersect / N_seed. This ratio provides a standardized, quantifiable indicator for objectively evaluating the completeness of the retrieval strategy.

[0071] Through the above technical solution, this application clarifies the specific calculation method for recall retrieval. Specifically, it calculates the intersection of multiple relevant documents with the seed document set, and then compares this intersection with the total number of documents in the seed document set to obtain the recall rate of the retrieval strategy. This quantitative method provides an objective and repeatable recall rate evaluation standard, avoiding the bias of subjective judgment and ensuring the accuracy of recall rate calculation. Based on the aforementioned recall rate, the system can more reliably evaluate the completeness of the current retrieval strategy in covering relevant documents.

[0072] refer to Figure 3 , Figure 3 This diagram illustrates feature processing in the retrieval and comparison method for the technical solutions provided in this application. For example... Figure 3 As shown, Figure 3The feature processing flow shown is as follows: Starting from the technical solution input by the user, the feature extraction agent first completes the extraction of technical features through the overall scheduling of the FTO workflow manager and global state management. These features are then divided into three categories: technical features, necessary technical features, and optional technical features. The keyword expansion agent then performs semantic expansion on these technical features, generating corresponding sets of synonyms, near-synonyms, and hyponyms. Combined with the matched IPC classification number, a set of keywords is formed to support subsequent retrieval, providing accurate and rich retrieval elements for the construction and optimization of precision retrieval strategies.

[0073] In some embodiments, this application further proposes specific steps for precision retrieval. First, multiple document samples are determined from a large number of relevant documents. This step aims to select a representative subset for analysis from a massive amount of relevant documents, thereby reducing computational complexity and improving processing efficiency. The determination of the document samples can employ various methods; for example, a certain proportion of documents can be randomly selected, or stratified sampling can be performed based on characteristics such as publication time, technical field, and relevance score, to ensure that the samples fully reflect the overall distribution of relevant documents. In this way, resource consumption can be effectively controlled while ensuring the accuracy of precision assessment.

[0074] Secondly, through the iterative analysis module's pre-set model, multiple document samples and technical solutions are semantically compared to obtain comparison results representing whether the document samples and technical solutions are related. The pre-set model can be a deep learning-based natural language processing model, such as a pre-trained BERT, RoBERTa, or its variants, which can capture deep semantic information of the text. During semantic comparison, the model can receive the text content of the document samples and the text content of the technical solutions as input, calculate the semantic similarity, relevance score, or perform binary classification, and output a comparison result. This comparison result can be a continuous similarity score (e.g., between 0 and 1) or a discrete classification label (e.g., "relevant" or "irrelevant"), used to quantify the degree of association between the document samples and the technical solutions.

[0075] Furthermore, based on the comparison results, the proportion of relevant documents among multiple document samples is determined, and this proportion is used as the precision of the retrieval strategy. Understandably, if the comparison result is a binary label, the number of document samples judged as "relevant" is directly counted and divided by the total number of document samples to obtain the proportion of relevant documents. If the comparison result is a continuous similarity score, a preset threshold can be set, and document samples with similarity scores higher than this threshold can be judged as relevant documents, and then their proportion is calculated. This proportion intuitively reflects how many proportions of the selected document samples are indeed relevant to the technical solution input by the user, thus providing a quantitative evaluation indicator for the effectiveness of the retrieval strategy.

[0076] In some embodiments, this application further proposes adjusting the retrieval strategy based on a comparison of recall and precision with preset thresholds. It is understood that after calculating the recall and precision of the current retrieval strategy, the system compares them with a first preset threshold and a second preset threshold. This comparison can be a simple numerical judgment, such as determining whether the recall is lower than the first preset threshold or whether the precision is lower than the second preset threshold. Based on the comparison results, the system will automatically trigger corresponding adjustment processing to optimize the current retrieval strategy.

[0077] The adjustment process includes at least one of the following: when the recall rate is lower than a first preset threshold, an expansion strategy is implemented to broaden the search scope; when the precision rate is lower than a second preset threshold, a narrowing strategy is implemented to narrow the search scope. When the system detects that the recall rate of the current search strategy fails to meet the preset requirements, meaning that a large number of relevant documents may not be retrieved, the system will automatically implement the expansion strategy. This strategy aims to increase the number of search results by modifying the search criteria, thereby improving the recall rate. For example, it can relax the restrictions on keywords or add classification numbers for relevant fields. When the system detects that the precision rate of the current search strategy is lower than the preset requirements, meaning that the search results contain a large number of irrelevant documents, the system will automatically implement the narrowing strategy. This strategy aims to reduce the occurrence of irrelevant documents by modifying the search criteria, thereby improving the precision rate. For example, it can add keyword restrictions or exclude documents under specific classification numbers.

[0078] Understandably, broadening and narrowing strategies are implemented by modifying the keywords, classification numbers, or logical operators in the search query. To broaden or narrow a search strategy, the system can modify specific elements of the search query. When implementing a broadening strategy, synonyms, hypernyms, or related terms can be added, or the "AND" logical operator in the search query can be replaced with an "OR" logical operator to expand the search scope; a wider range of International Patent Classification (IPC) numbers can also be introduced. When implementing a narrowing strategy, vague or generalized keywords can be deleted, more specific qualifiers can be added, or the "OR" logical operator in the search query can be replaced with an "AND" logical operator to strengthen the restriction of the search conditions; a more refined IPC number can also be used. These modifications can be performed automatically by pre-trained models or rule-based expert systems.

[0079] In some embodiments, this application further proposes to extract the keyword set corresponding to the technical solution, including: extracting the technical feature set of the technical solution through a pre-trained feature extraction module; and extracting the keyword set corresponding to the technical feature set through a pre-trained keyword expansion module.

[0080] Understandably, the feature extraction module is an artificial intelligence model or algorithm component pre-trained on a large amount of professional text data. Its main function is to automatically identify, parse, and extract information units with clear technical meaning and distinctiveness from the unstructured technical solution text input by the user. These information units collectively constitute the technical feature set of the technical solution. This module can be implemented using various technologies. For example, it can be based on deep learning natural language processing models, such as Named Entity Recognition (NER) and relation extraction techniques, to identify key entities (such as components, materials, methods, parameters, etc.) and their interrelationships in the technical solution. Alternatively, it can identify technical features by constructing a domain-specific rule base and dictionary, combined with syntactic analysis and pattern matching. In this way, the original technical solution text can be transformed into a structured set of technical features, providing accurate input for subsequent keyword generation.

[0081] The keyword expansion module can also be a pre-trained module. Its function is to generate a more comprehensive and representative keyword set based on the extracted set of technical features. This module aims to overcome the limitations of a single technical feature in expression by expanding the scope of the search by expanding related vocabulary. For example, this module can use word vector models or pre-trained language models to calculate the semantic similarity of words in the technical features, thereby finding synonyms, near-synonyms, or related words; it can also combine a pre-defined technical field ontology or knowledge graph to expand based on the hierarchical relationship (such as hypernyms and hyponyms) and association relationship of concepts; in addition, it can analyze the co-occurrence frequency of words in a large number of technical documents and use words that frequently co-occur with words in the technical features as expanded keywords. Through these expansion methods, it is ensured that the generated keyword set not only includes the core terminology of the technical solution, but also covers various expressions and concepts closely related to it, thus providing a richer vocabulary foundation for building efficient search strategies.

[0082] refer to Figure 4 , Figure 4 This diagram illustrates feature processing in the retrieval and comparison method for the technical solutions provided in this application. For example... Figure 4 As shown, this application can take a keyword set as input and enter the iterative analysis stage. First, it calculates recall and precision and generates corresponding optimization suggestions. Then, it aggregates and integrates search elements. Subsequently, it verifies the quantity and other dimensions of the search results through the search and quantity verification stage, and then enters the search verification step. If the search verification fails, it will feed back to the iterative analysis agent to readjust the search strategy and execute the process again. If the verification passes, it will output multiple relevant documents. This iterative cycle can dynamically optimize the search strategy, thereby ensuring the quality and accuracy of the search results. In some embodiments, this application further proposes a step of extracting a set of technical features of a technical solution, including: performing semantic analysis on the text content of the technical solution to identify multiple technical elements contained in the technical solution; classifying the multiple technical elements into technical field features, necessary technical features, and optional technical features based on preset technical feature classification rules; and generating a set of technical features based on the classification results, wherein the set of technical features includes at least one technical field feature and at least one necessary technical feature.

[0083] It is understandable that semantic parsing of the textual content of a technical solution refers to using natural language processing technology to deeply understand and analyze the textual content of the technical solution. This includes, but is not limited to, lexical analysis (such as word segmentation and part-of-speech tagging), syntactic analysis (such as dependency parsing and constituent parsing), and semantic analysis (such as named entity recognition, relation extraction, and event extraction). Through semantic parsing, the system can accurately identify various entities mentioned in the text (such as components, materials, methods, effects, etc.) and their interrelationships, thereby transforming this discrete textual information into structured technical elements. For example, for "a device that achieves rotation by driving gears with a motor," semantic parsing can identify technical elements such as "motor," "gear," "rotation," "drive," and "achieve," and understand the driving relationships between them. Technical elements are the basic building blocks of a technical solution; they can be components, methods, materials, structures, functions, effects, etc., involved in the technical solution. Identifying these technical elements is the foundation for subsequent technical feature classification and set construction, ensuring a comprehensive understanding of the technical solution.

[0084] Based on this, this application can categorize multiple technical elements into technical field features, necessary technical features, and optional technical features according to preset technical feature classification rules. Preset technical feature classification rules are a set of well-defined standards or models used to guide the system in classifying identified technical elements. These rules can be built based on an expert knowledge base; for example, defining which words or phrases represent technical fields, which represent core functions or structures, and which represent auxiliary or improving features. Furthermore, they can be trained through machine learning models to automatically learn and identify different types of technical features. Technical field features refer to elements that define the application scope or technical background of a technical solution, such as "communication field," "medical devices," and "artificial intelligence." These features help determine the macro-direction of the search. Necessary technical features refer to the core, indispensable elements of a technical solution; they are essential for achieving the main purpose of the technical solution. For example, in "a device that achieves rotation by driving gears with a motor," "motor drive" and "gears achieve rotation" can be necessary technical features. These features are key to constructing core search terms. Optional technical features refer to auxiliary elements that may or may not exist in a technical solution, or that can have multiple alternatives. These features can be used to further optimize, improve, or limit the technical solution, but they are not essential for achieving the basic function. For example, "the motor is a stepper motor" can be an optional technical feature. Such features can be used to broaden or narrow the search scope.

[0085] The technical feature set includes at least one technical field feature and at least one necessary technical feature. This is to ensure that the extracted technical feature set can accurately reflect the field to which the technical solution belongs to, so as to ensure the correctness of the search direction, and can fully cover its core innovation points to ensure the accuracy and relevance of the search content, so as to avoid the problem of incomplete or inaccurate search due to the omission of key information.

[0086] In some embodiments, this application further proposes a specific method for extracting a set of keywords corresponding to a set of technical features, which includes the following steps: First, semantic expansion processing is performed on each technical feature in the technical feature set to generate at least one of the following sets: a set of synonyms, a set of near-synonyms, and a set of hyponyms and hypernyms corresponding to each technical feature. Semantic expansion processing aims to broaden the retrieval coverage by identifying words semantically related to the original technical features. For example, for a technical feature "motor," its synonyms may include "motor," its near-synonyms may include "engine," its hypernyms may include "power unit," and its hyponyms may include "DC motor," "stepper motor," etc. This processing can be implemented in various ways. For example, a pre-trained word vector model can be used to calculate the semantic similarity between words, and words with similarity higher than a preset threshold can be used as expanded vocabulary; alternatively, based on a domain knowledge graph or ontology, expanded vocabulary can be obtained by querying the hyponyms, hypernyms, and synonyms between words; or, a large-scale corpus can be used for statistical analysis to identify words that frequently co-occur with the technical features as potential expanded vocabulary. In this way, it can be ensured that the retrieval strategy can cover all possible expressions in the technical solution, thereby improving recall.

[0087] Secondly, based on the technical fields associated with the set of technical features, at least one International Patent Classification (IPC) number is recommended from a pre-defined classification database. The International Patent Classification (IPC) is an internationally recognized patent document classification system that accurately identifies the technical field to which a technical solution belongs. The pre-defined classification database stores IPC classification numbers and their corresponding technical field descriptions or keywords. The system can map the text content of the technical feature set to the most relevant IPC classification number using a text classification algorithm; or, through keyword matching, compare the keywords in the technical feature set with the descriptions of the classification numbers in the database, recommending the classification number with the highest matching degree. Introducing IPC classification numbers helps to limit the search scope to highly relevant technical fields, thereby improving the accuracy of the search.

[0088] Furthermore, the semantically expanded vocabulary set is combined with at least one International Patent Classification (IPC) number to form a keyword set. This combination method fully utilizes the breadth of semantic expansion and the depth of IPC classification numbers. Understandably, synonyms, near-synonyms, and hyponyms obtained from semantic expansion can be connected using Boolean logic operators (such as "OR") to form a vocabulary group; simultaneously, the recommended IPC classification number is also included as part of the search criteria, combined with the vocabulary group using Boolean logic operators (such as "AND"). This ensures that the keyword set not only covers various expressions of technical solutions but also accurately targets specific technical fields, providing high-quality input for subsequent search strategy construction.

[0089] In some embodiments, this application further proposes the following steps when determining multiple target documents from multiple related documents based on the technical elements corresponding to the technical solution: performing preliminary screening on multiple related documents through a pre-trained preliminary screening module; and performing technical feature comparison on the pre-screened documents through a pre-trained technical feature comparison module to determine multiple target documents.

[0090] The preliminary screening module is a pre-trained and optimized functional unit whose main function is to quickly and coarsely filter a large number of relevant documents. This module can be trained based on various preset rules or machine learning models. For example, it can learn the mapping relationship between metadata features of documents (such as classification number, applicant, application date, etc.) and the relevance of the technical solution, or it can identify the matching degree between high-frequency keywords in the document text and the core concepts of the technical solution. Its goal is to quickly eliminate obviously irrelevant documents without performing deep semantic analysis, thereby narrowing the document set to a more targeted range. Preliminary screening of multiple relevant documents refers to the filtering operation performed on multiple relevant documents selected from the document database using the aforementioned preliminary screening module. This process aims to divide the original set of relevant documents into a smaller subset that can better contain the target documents by quickly evaluating the surface features or coarse-grained content of the documents.

[0091] The technical feature comparison module is a pre-trained and optimized functional unit specifically designed for refined technical feature comparison of documents. This module integrates advanced natural language processing technologies and semantic understanding models, enabling in-depth analysis of technical features in technical solutions and document texts, and assessment of their semantic similarity or logical connections. For example, this module can be trained to identify synonyms, near-synonyms, hypernyms, and hyponyms of technical features, or to identify the same or similar technical concepts under different expressions by constructing a technical knowledge graph. Its goal is to achieve accurate identification and high-precision matching of technical features. Technical feature comparison of pre-screened documents refers to the detailed comparison operation performed on a subset of documents after preliminary screening using the aforementioned technical feature comparison module. This process may involve semantically comparing the technical elements contained in the user-input technical solution with the technical features recorded in each of the pre-screened documents, either individually or as a whole. The purpose of the comparison can be to assess the degree of fit between the document and the technical solution in terms of technical content, for example, by calculating the matching score of technical features, or to determine whether the document covers the key technical points of the technical solution. This refined comparison allows for the accurate identification of target documents that are highly relevant to the technical solution.

[0092] Furthermore, after completing the technical feature comparison, the system can set a preset threshold based on the comparison results (e.g., technical feature matching score). Documents with a matching degree reaching or exceeding this threshold are identified as target documents highly relevant to the user-input technical solution. These target documents will serve as the basis for generating subsequent search comparison information (such as infringement search reports), ensuring the accuracy and effectiveness of subsequent analysis.

[0093] In some embodiments, this application further proposes that when the document is a patent, the preliminary screening includes preliminary screening of multiple relevant documents according to preset screening criteria, wherein the screening criteria include at least one of the following: International Patent Classification, applicant, legal status, application date, and patent type.

[0094] It is understood that the document is a patent, which clarifies that the document type processed in this embodiment is patent literature. Patent literature can have specific structures and metadata, such as International Patent Classification (IPC), applicant, legal status, application date, and patent type. This metadata provides a basis for refined screening. Preset screening criteria refer to standards pre-set in the initial screening stage for filtering relevant literature. These criteria can be configured according to user needs, technical field characteristics, or search objectives to ensure that the selected literature is more targeted. The International Patent Classification (IPC) is an internationally recognized patent literature classification system used to classify patent technical content. Using it as a screening criterion allows for the rapid location of highly relevant patent literature based on the specific technical field or subfield to which the technical solution belongs, while excluding irrelevant technical fields.

[0095] In some embodiments, this application further proposes to perform technical feature comparison analysis on the preliminarily screened documents, compare the set of technical features of the technical solution with the technical features recorded in the claims of each document one by one, and determine the documents with a technical feature matching degree higher than a preset threshold as target documents based on the comparison results.

[0096] Understandably, performing technical feature comparison analysis on initially screened documents aims to conduct a deeper review of the technical content of these documents, identifying the parts that are highly relevant to the user-input technical solution in terms of technical features. The goal is precise matching at the technical level, rather than relying solely on metadata. This analysis can be achieved through natural language processing techniques, parsing the textual content of the documents, especially the claims, to extract the technical features contained within. For example, techniques such as dependency parsing and named entity recognition can be used to identify key technical elements in the technical solution, such as the subject, object, action, and attributes, and represent them in a structured manner.

[0097] Based on this, the set of technical features of the technical solution is compared one by one with the technical features described in the claims of each document, aiming to quantify the similarity of the technical content between the user's technical solution and the document. According to the comparison results, documents with a technical feature matching degree higher than a preset threshold are identified as target documents. This is a decision-making step based on the quantitative results of the comparison, used to filter out documents that are highly relevant to the user's technical solution in terms of technical content. It is understandable that the setting of the preset threshold affects the recall and precision of the target documents. This preset threshold can be adjusted according to the actual application scenario and the requirements for matching accuracy.

[0098] Furthermore, based on the above comparison results, this application can identify documents with a technical feature matching degree higher than a preset threshold as target documents. This is a decision-making step based on the quantitative results of the comparison, used to ultimately filter out documents that are highly relevant to the user's technical solution in terms of technical content. The setting of the preset threshold directly affects the recall and precision of the target documents. This preset threshold can be adjusted according to the actual application scenario and the requirements for matching accuracy. For example, in infringement retrieval, the threshold can be set higher to ensure that the selected documents are highly similar to the technical solution; while in novelty retrieval, the threshold can be set relatively loosely to discover more potential related technologies. The matching degree can be a comprehensive score, for example, by weighted averaging of the comparison scores of multiple technical features, or by using a more complex machine learning model to predict the matching degree.

[0099] refer to Figure 5 , Figure 5 An end-to-end workflow diagram of the retrieval and comparison method for the technical solutions provided in the embodiments of this application. For example... Figure 5 As shown, after the project management phase is initiated, this application can proceed sequentially to understanding the technical solution, conducting anti-infringement searches, performing initial screening to obtain search results data of multiple relevant documents, then performing feature comparison calculations on the document set that has passed the initial screening, generating an FTO report based on the comparison results and outputting corresponding action suggestions, and feeding back the search comparison information to the project management phase, forming a complete closed-loop process from project initiation to result application, which can realize full-process management of search comparison work.

[0100] In some embodiments, this application further proposes a step of preliminary screening of multiple relevant documents, which further includes: receiving an intermediate file from the retrieval strategy stage, the intermediate file including retrieval result data of multiple relevant documents; performing preliminary screening on the document data in the intermediate file based on preset screening conditions or configurable screening instructions; wherein the configurable screening instructions are generated based on the analysis results of the intermediate file.

[0101] As is understandable, intermediate files refer to temporary data files containing search results generated and stored by the system after the search strategy phase is completed. These files may include detailed information from multiple relevant documents selected from the literature database, such as metadata like document ID, title, abstract, keywords, classification number, applicant, application date, and legal status, as well as search scores or relevance ranking information. Receiving these intermediate files aims to ensure that the initial screening phase is based on the latest and most comprehensive search results data, avoiding duplicate data acquisition and providing a data foundation for subsequent screening.

[0102] Based on this, preliminary screening can be conducted in two ways. One is based on preset screening criteria, which are predefined by the system, such as documents within a specific time frame, documents from a specific applicant, or documents related to a specific legal status. These preset criteria can be general, empirical screening rules. The other is based on configurable screening instructions, which are dynamically generated and can be adjusted according to the output results of the retrieval strategy stage. Performing preliminary screening involves filtering the document data in the intermediate files according to these criteria or instructions, eliminating documents that do not meet the requirements, thereby narrowing the scope of subsequent processing and improving the efficiency of subsequent technical feature comparison.

[0103] In some embodiments, this application further proposes that when comparing the technical features of pre-screened documents, an intermediate file is first received from the pre-screening stage. This intermediate file includes a set of documents that have passed the pre-screening. As the output of the pre-screening stage, this intermediate file carries more relevant document data that has been pre-filtered, ensuring that the technical feature comparison module can operate based on a more refined set of documents, avoiding comparisons with a large number of irrelevant documents, thereby improving the efficiency and accuracy of subsequent comparisons. This intermediate file can be a structured data format, such as an XML or JSON file, or a database query result set, containing key information such as document identifiers, abstracts, claims, and specifications. The receiving operation can be implemented through a file reading interface, database connection, or API call.

[0104] Based on this, the system performs technical feature comparisons on the literature data in the intermediate files, using preset comparison rules or configurable comparison instructions. The preset comparison rules provide a standardized comparison method, which can include various methods such as keyword matching, semantic similarity calculation, and structured comparison of technical elements. A match is defined as when a necessary technical feature in the technical solution appears as a synonym or hypernym in the document's claims. Furthermore, the configurable comparison instructions are generated based on the analysis results of the document claims in the intermediate files. This means the system can utilize natural language processing technologies, such as text segmentation, part-of-speech tagging, named entity recognition, and syntactic analysis, to conduct in-depth analysis of the document claims to extract key technical elements, limiting conditions, and logical relationships. For example, if the analysis reveals that the claims of a batch of documents generally adopt a specific limiting method, the system can generate instructions to use this limiting method as an important consideration during comparison; or, if a batch of documents' claims contain a large number of hypernyms, the system can generate instructions to appropriately relax the matching requirements for hyponyms during comparison. This adaptive instruction generation mechanism enables the comparison process to be dynamically adjusted according to the actual characteristics of the documents. For example, it allows users or the system to dynamically adjust parameters such as comparison weight, similarity threshold, and comparison range according to the characteristics of the documents.

[0105] In some embodiments, the search comparison information specifically refers to an infringement search report. This infringement search report includes a comparison table of the technical solution and the claims of multiple target documents, as well as an infringement risk assessment conclusion based on the comparison results.

[0106] Understandably, an infringement search report is a document specifically designed to assess whether a technical solution carries the risk of infringement. It aims to present complex search and comparison results to users in a structured and easily understandable manner, enabling them to quickly determine the relationship between their technical solution and existing technologies, especially regarding patent infringement risks. This report can be automatically generated by the system, or it can be edited and refined by professionals after the system generates initial data.

[0107] The infringement search report includes a comparison table of technical solutions and claims from multiple target documents. This table compares the technical solution to be analyzed with the claims of the target documents (which may be patent documents) line by line. The comparison table details each technical feature of the technical solution and compares it with the corresponding technical features in the independent and dependent claims of the target documents.

[0108] In some embodiments, the infringement risk assessment conclusion included in the infringement search report, based on the comparison results, is a summary opinion that qualitatively or quantitatively judges the potential infringement risks of the technical solution based on the comparison results between the technical solution and the claims of the target document. The assessment conclusion can comprehensively consider factors such as the matching situation in the claim comparison table, the legal status of the target document, and the innovative points of the technical solution. For example, if all the necessary technical features of the technical solution fall within the scope of protection of the independent claims of the target document, a high-risk conclusion can be drawn; if only some features match, or if there are technical differences, a medium-low risk or no risk conclusion can be drawn. The assessment conclusion may include a risk level (e.g., high, medium, low), risk cause analysis, and suggestions for risk avoidance. This can provide users with direct decision support, helping them understand the potential legal risks of their technical solutions.

[0109] refer to Figure 6 , Figure 6 A schematic block diagram of a retrieval and comparison device for a technical solution provided in an embodiment of this application; in some embodiments, this application provides a retrieval and comparison device 600 for a technical solution, applied to a retrieval and comparison system, the retrieval and comparison device for the technical solution comprising: The feature processing module 601 is used to acquire the technical solution input by the user and extract the keyword set corresponding to the technical solution; The iterative analysis module 602 is used to construct a retrieval strategy based on a set of keywords, select multiple relevant documents from a preset literature database according to the retrieval strategy, calculate the quality index corresponding to the retrieval strategy based on the selected multiple relevant documents, and update the retrieval strategy based on the quality index until the quality index meets the preset conditions. The patent screening module 603 is used to identify multiple target documents from multiple relevant documents based on the technical elements corresponding to the technical solutions. The report generation module 604 is used to generate search comparison information based on multiple target documents and to display the search comparison information.

[0110] In some embodiments, the feature processing module includes a feature extraction module and a keyword expansion module; when extracting the keyword set corresponding to the technical solution: the feature extraction module is used to extract the technical feature set of the technical solution; the keyword expansion module is used to extract the keyword set corresponding to the technical feature set.

[0111] In addition, embodiments of this application also provide a control device, including: One or more processors; and A memory associated with one or more processors, the memory being used to store program instructions that, when read and executed by one or more processors, perform the steps of any of the methods in the foregoing method embodiments.

[0112] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the methods in the foregoing method embodiments.

[0113] in, Figure 7 An exemplary architecture of a control device is shown, which may include a processor 710, a video display adapter 711, a disk drive 712, an input / output interface 713, a network interface 714, and a memory 720. The processor 710, video display adapter 711, disk drive 712, input / output interface 713, network interface 714, and memory 720 can communicate with each other via a communication bus 730.

[0114] The processor 710 can be implemented using a general-purpose CPU, microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits to execute relevant programs and implement the technical solution provided in this application.

[0115] The memory 720 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 720 can store the operating system 721 for controlling the operation of the device 700, and the basic input / output system (BIOS) 722 for controlling the low-level operations of the device 700. Additionally, it can store a web browser 723, a data storage management system 724, and a technical solution retrieval and comparison device 600, etc. The aforementioned technical solution retrieval and comparison device 600 can be an application program that specifically implements the aforementioned steps in the embodiments of this application. In summary, when the technical solution provided in this application is implemented through software or firmware, the relevant program code is stored in the memory 720 and is called and executed by the processor 710.

[0116] Input / output interface 713 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touch screens, microphones, various sensors, etc., and output devices may include displays, speakers, vibrators, indicator lights, etc. Network interface 714 is used to connect communication modules (not shown in the figure) to realize communication interaction between this device and other devices. The communication module can communicate via wired means (e.g., USB, network cable, etc.) or via wireless means (e.g., mobile network, WIFI, Bluetooth, etc.). Bus 730 includes a path for transmitting information between various components of the device (e.g., processor 710, video display adapter 711, disk drive 712, input / output interface 713, network interface 714, and memory 720).

[0117] It should be noted that although the above-described device only shows the processor 710, video display adapter 711, disk drive 712, input / output interface 713, network interface 714, memory 720, bus 730, etc., in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the solution of this application, and does not necessarily include all the components shown in the figures.

[0118] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer program product. This computer program product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of the embodiments of this application.

[0119] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and embodiments of this application. The descriptions of the embodiments above are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in specific embodiments and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for searching and comparing technical solutions, characterized in that, The method, applied to a retrieval and comparison system, includes: Obtain the technical solution input by the user, and extract the keyword set corresponding to the technical solution; A retrieval strategy is constructed based on the keyword set, and multiple relevant documents are selected from a preset literature database according to the retrieval strategy. The quality index corresponding to the retrieval strategy is calculated based on the selected multiple relevant documents, and the retrieval strategy is updated based on the quality index until the quality index meets the preset conditions. Based on the technical elements corresponding to the technical solution, multiple target documents are identified from the multiple related documents; Search comparison information is generated based on the multiple target documents, and the search comparison information is displayed.

2. The method according to claim 1, characterized in that, The quality indicators include recall and precision. The calculation of the quality indicators corresponding to the retrieval strategy based on the selected multiple relevant documents includes: The following steps are performed using the pre-trained iterative analysis module: A pre-search is performed on the literature database to obtain a seed literature set corresponding to the technical solution; Based on the seed document set, a complete retrieval is performed on the multiple related documents to obtain the recall rate corresponding to the retrieval strategy; Based on the iterative analysis module, precision retrieval is performed on the multiple relevant documents to obtain the precision rate corresponding to the retrieval strategy.

3. The method according to claim 2, characterized in that, The pre-search includes the following steps: Based on the technical field to which the technical solution belongs, target owners whose ranking of the number of technical documents meets preset requirements are determined from the owners corresponding to the multiple related documents; The target owner's multiple technical documents in the technical field are used as a seed document set.

4. The method according to claim 2, characterized in that, The comprehensive search includes the following steps: Calculate the number of intersections between the multiple related documents and the seed document set; The ratio of the number of intersections to the total number of documents in the seed document set is used as the recall rate of the retrieval strategy.

5. The method according to claim 2, characterized in that, The precise search includes the following steps: Multiple document samples were identified from the aforementioned multiple related documents; Using the preset model within the iterative analysis module, the multiple document samples are semantically compared with the technical solution to obtain comparison results characterizing whether the document samples and the technical solution are related. Based on the comparison results, the proportion of relevant documents in the multiple document samples is determined, and the proportion is used as the precision of the retrieval strategy.

6. The method according to claim 1, characterized in that, The quality metrics include recall and precision, and updating the retrieval strategy based on the quality metrics includes: Based on the comparison results of the recall rate and the precision rate with the preset threshold, the retrieval strategy is adjusted. The adjustment process includes at least one of the following: When the recall rate is lower than a first preset threshold, a broadening strategy is executed to expand the search scope; When the precision rate is lower than the second preset threshold, a narrowing strategy is executed to reduce the search scope; The broadening strategy and the narrowing strategy are implemented by modifying the keywords, classification numbers or logical operators of the search query in the search strategy.

7. The method according to claim 1, characterized in that, The extraction of the keyword set corresponding to the technical solution includes: The technical feature set of the technical solution is extracted using a pre-trained feature extraction module; The keyword set corresponding to the technical feature set is extracted using a pre-trained keyword expansion module.

8. The method according to claim 7, characterized in that, The set of technical features extracted from the technical solution includes: Semantic analysis is performed on the text content of the technical solution to identify multiple technical elements contained in the technical solution; Based on preset technical feature classification rules, the multiple technical elements are classified into technical feature features, necessary technical features, and optional technical features. The set of technical features is generated based on the classification results, wherein the set of technical features includes at least one technical field feature and at least one essential technical feature.

9. The method according to claim 7, characterized in that, The extraction of the keyword set corresponding to the set of technical features includes: Each technical feature in the set of technical features is semantically expanded to generate at least one of the following: a set of synonyms, a set of near-synonyms, and a set of hyponyms and hypernyms corresponding to each technical feature. Based on the technical fields associated with the set of technical features, at least one international patent classification number is recommended from a preset classification number library; The set of words that have undergone semantic expansion processing is combined with the at least one international patent classification number to form the set of keywords.

10. The method according to claim 1, characterized in that, Based on the technical elements corresponding to the technical solution, multiple target documents are determined from the multiple relevant documents, including: The pre-trained preliminary screening module is used to perform preliminary screening of the multiple relevant documents; By using a pre-trained technical feature comparison module, the technical features of the documents that have passed the initial screening are compared to identify multiple target documents.

11. The method according to claim 10, characterized in that, If the document is a patent, then the preliminary screening includes the following steps: The multiple relevant documents are initially screened according to preset screening criteria, which include at least one of the following: International Patent Classification (IPC), applicant, legal status, application date, and patent type.

12. The method according to claim 10, characterized in that, The technical feature comparison includes the following steps: Technical feature comparison analysis is performed on the documents that have undergone the preliminary screening. The set of technical features of the technical solution is compared one by one with the technical features recorded in the claims of each document. Based on the comparison results, documents with a technical feature matching degree higher than a preset threshold are identified as target documents.

13. The method according to claim 10, characterized in that, The preliminary screening of the multiple relevant documents also includes: Receive intermediate files from the retrieval strategy stage, the intermediate files including retrieval result data of the plurality of relevant documents; The preliminary screening is performed on the literature data in the intermediate file based on preset screening conditions or configurable screening instructions. The configurable filtering instructions are generated based on the analysis results of the intermediate files.

14. The method according to claim 10, characterized in that, The technical feature comparison of the documents that have passed the initial screening also includes: Receive intermediate files from the initial screening stage, the intermediate files including a set of documents that have passed the initial screening; Based on preset comparison rules or configurable comparison instructions, technical features are compared on the literature data in the intermediate files. The configurable comparison instructions are generated based on the analysis results of the document claims in the intermediate file.

15. The method according to claim 1, characterized in that, The search comparison information is an infringement search report; The infringement search report includes: a comparison table of the claims of the technical solution and the multiple target documents, and an infringement risk assessment conclusion based on the comparison results.

16. A retrieval and comparison device, characterized in that, The retrieval and comparison device is used in a retrieval and comparison system and includes: The feature processing module is used to acquire the technical solution input by the user and extract the keyword set corresponding to the technical solution; The iterative analysis module is used to construct a retrieval strategy based on the keyword set, select multiple relevant documents from a preset literature database according to the retrieval strategy, calculate the quality index corresponding to the retrieval strategy based on the selected multiple relevant documents, and update the retrieval strategy based on the quality index until the quality index meets the preset conditions. The patent screening module is used to identify multiple target documents from the multiple relevant documents based on the technical elements corresponding to the technical solution. The report generation module is used to generate search comparison information based on the multiple target documents and to display the search comparison information.

17. The retrieval and comparison device according to claim 16, characterized in that, The feature processing module includes a feature extraction module and a keyword expansion module; when extracting the keyword set corresponding to the technical solution: The feature extraction module is used to extract a set of technical features of the technical solution; The keyword expansion module extracts the keyword set corresponding to the set of technical features.

Images