LLM large model-based patent report intelligent generation method and system

Abstract

The present application relates to the technical field of patent analysis, and is used to solve the problem that the existing patent analysis technology can only provide fixed-dimension analysis and cannot meet the diversified needs of users, and generally lacks accurate modeling means for annual data change trend, making it difficult to identify patent expansion years and unable to dynamically adjust the effective query period, resulting in lack of pertinence and timeliness of the analysis results, specifically, a patent report intelligent generation method and system based on LLM large model, the present application provides patent report generation including technology development and trend analysis, applicant ranking analysis and regional analysis dimensions, and introduces inflection point identification and slope calculation method, quantifies the patent change value of each year, and dynamically determines the effective duration of the query period through the expansion year proportion and multi-level interval comparison mechanism, which can meet the diversified needs of users, provide more comprehensive and in-depth patent analysis, and ensure the timeliness and accuracy of the analysis results.

Description

Technical Field

[0001] This invention relates to the field of patent analysis technology, specifically to a method and system for intelligent generation of patent reports based on an LLM large model. Background Technology

[0002] With the development of science and technology, significant progress has been made in fields such as artificial intelligence, patent analysis, and information processing. Artificial intelligence, through machine learning and deep learning algorithms, can simulate human intelligent behavior and process and analyze large amounts of data. Patent analysis, through the collection, organization, and analysis of patent data, can provide insights into the patent landscape and technological development trends in a particular field. Information processing refers to the technologies for collecting, storing, processing, and transmitting data, providing data support for the aforementioned two technologies.

[0003] In existing technological solutions, patent analysis techniques primarily generate patent reports by mining and analyzing data from patent databases. While these techniques offer some useful information, they also have several limitations. First, they often provide only fixed-dimensional analysis, failing to meet diverse user needs. Furthermore, they generally lack precise modeling methods for annual data trends, making it difficult to identify patent expansion years or dynamically adjust effective query cycles, resulting in a lack of targeted and timely analysis. Second, LLM models, due to their complex structure and large number of parameters, often lack interpretability. This means that even if a model generates a patent report, it's difficult to explain the underlying reasons and logic, reducing user trust. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for intelligent generation of patent reports based on an LLM large model, which solves the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent generation of patent reports based on an LLM large model, comprising the following steps: S1. Perform semantic recognition on the natural semantic query statement input by the current user, and in combination with the preset initial query period, filter out a number of patent documents within the initial query period, record the publication time of the corresponding patent documents, and count the number of patent documents in each unit year by year. S2. Perform numerical model analysis on the changing trend of the number of patent documents in each unit year to obtain the patent change value of the corresponding unit year. Combined with the statistical averaging algorithm, obtain the average patent change value. Based on the comparative analysis process, count the number of marked patent expansion years and calculate the proportion of expansion years in the initial query period. Perform query period confirmation analysis on the proportion of expansion years in the initial query period to obtain the effective query time of the corresponding level. S3. Based on the confirmed effective query duration of the corresponding class, extract the patent change value and the number of patent documents in each unit year within the effective query duration. After feature extraction, analyze the fluctuation level of patent changes within the effective query duration and generate a patent report on technology development trend information accordingly. S4. Filter the patent documents within the initial query period, extract the valid period patent documents, and download the text information of the corresponding valid period patent documents. After feature extraction, analyze the concentration of patent applications filed by each applicant in different technology subcategories within the valid query period, and generate an applicant ranking information patent report accordingly. S5. Based on the text information of the corresponding valid period patent documents, after feature extraction, analyze the patent competition level of each province within the valid query period, and generate a regional competition distribution image information patent report accordingly. S6. Store the generated patent report content in the cloud database and mark the corresponding patent report ID. Display the information on the user report details page for users to view and download.

[0006] Preferably, step S1 specifically includes: S11. Based on the natural semantic query statement entered by the current user on the front-end interface, after semantic recognition by the LLM large model, extract the main information of the current user's query statement, which includes technical field keywords. The semantic recognition process includes: identifying the distribution of verbs, noun phrases, and adjective structures in the natural semantic query statement structure, removing irrelevant modifiers, and extracting the main information of the query statement; S12. Using the technical field keywords in the main information of the query statement entered by the current user, a comprehensive query is performed in the patent resource database. The preset historical query duration is used as the initial query period. Several patent documents containing the corresponding technical field keywords are selected within the initial query period, and the publication time of the corresponding patent documents is recorded. Based on the selected initial query period containing several patent documents containing keywords of the corresponding technical field, and combined with the publication time of the corresponding patent documents, the publication time of each patent document is classified by year, and the number of patent documents in each year is counted.

[0007] Preferably, step S2 specifically includes: S21. Perform a numerical model analysis on the changing trend of the number of patent documents in each unit year. Use the unit year as the horizontal axis and the number of patent documents in the corresponding unit year as the vertical axis. Based on this, establish a two-dimensional coordinate system and draw a line graph of the patent change trend. Count the total number of inflection points of the change line in the two-dimensional coordinate system of the patent change trend. Using the inflection point of the change line in the two-dimensional coordinate system of patent change trend as the dividing point, the patent change trend line graph is divided into line segments to obtain the unit change line segments for each unit year. By calculating the slope value formed between the unit change line segment for each unit year and the horizontal axis of the two-dimensional coordinate system, it is recorded as the patent change value for the corresponding unit year. Combined with the statistical mean calculation algorithm, the average patent change value is obtained. S22. Compare and analyze the patent change value of each unit year with the average patent change. If the patent change value of the corresponding unit year exceeds the average patent change, mark the corresponding unit year as a patent expansion year and count the number of patent expansion years. Conversely, mark the corresponding unit year as a patent contraction year. The total number of units per year is determined based on the preset historical query duration. The ratio of the total number of units per year to the number of patent expansion years is calculated to obtain the proportion of expansion years within the initial query period.

[0008] Preferably, step S2 further includes: S23. Based on the proportion of expansion years in the initial query period, and pre-setting gradient comparison intervals A-range, B-range and C-range of expansion year proportion, the expansion year proportion in the initial query period is compared with the pre-set gradient comparison intervals A-range, B-range and C-range of expansion year proportion by performing interval substitution analysis, thereby obtaining the first-level query period confirmation instruction, the second-level query period confirmation instruction and the third-level query period confirmation instruction respectively. The specific process of interval substitution analysis includes: If the proportion of expansion years in the initial query period is within the preset gradient comparison range A-range, a first-level query period confirmation instruction will be generated. If the proportion of expansion years in the initial query period is within the preset gradient comparison range B-range, a second-level query period confirmation instruction will be generated. If the proportion of expansion years in the initial query period is within the preset gradient comparison range C-range, a third-level query period confirmation instruction will be generated. Based on the generated query cycle confirmation instruction of the corresponding level, the query cycle of the natural semantic query statement entered by the current user on the front-end interface is determined as the first-order effective query duration t1, the second-order effective query duration t2, and the third-order effective query duration t3, respectively, where t1 > t2 > t3.

[0009] Preferably, step S3 specifically includes: S31. Based on the confirmed valid query duration for the corresponding class, extract the patent change value and the number of patent documents for each unit year within the valid query duration to construct patent change data information. S32. Perform feature identification on patent change data, extract the minimum patent change value, maximum patent change value, minimum number of patent documents, and maximum number of patent documents respectively, and correlate the four data items. After dimensionless processing, analyze the fluctuation level of patent changes within the effective query period to determine the patent change fluctuation value, specifically: In the formula, This indicates the value of patent change fluctuation. and These represent the maximum and minimum values ​​of patent change, respectively. and These represent the maximum and minimum number of patent documents, respectively. Here, a1 and a2 both represent weight values, the specific values ​​of which are set by those skilled in the art. S33. Compare and analyze the patent change fluctuation value with the pre-set fluctuation threshold. If the patent change fluctuation value exceeds the fluctuation threshold, issue a first-level patent technology development prompt message. The prompt message is: the patent technology development trend is good within the effective query period. If the patent change fluctuation value does not exceed the fluctuation threshold, issue a second-level patent technology development prompt message. The prompt message is: the patent technology development trend is stable within the effective query period. Based on the patent technology development prompts issued at the corresponding level, a patent report on technology development trends is generated.

[0010] Preferably, step S4 specifically includes: S41. Based on the confirmed valid query duration of the corresponding class, filter each patent document within the initial query period, extract the patent documents whose publication time is within the valid query duration, and mark them as valid period patent documents. At the same time, download the text information of the corresponding valid period patent documents and send the text information of each valid period patent document to the cloud database for storage. The cloud database is also used to store the patent status determination table, wherein the text information includes the applicant's name, application address, claims, description, legal status text information and technical subclassification. S42. Perform feature recognition on the text information of each valid period patent document stored in the cloud database, extract the applicant name and technology subclass from the text information of each valid period patent document within the valid query time, construct patent technology classification data information, and count the number of patents applied for by each applicant in each technology subclass within the valid query time in the patent technology classification data information, and obtain the total number of patents applied for in each technology subclass within the valid query time by combining statistical summation algorithm.

[0011] Preferably, step S4 further includes: S43. Correlate the number of patents filed by each applicant in each technology subclass with the total number of patents filed in each technology subclass, analyze the concentration of patents filed by each applicant in different technology subclasses within the effective query period, and determine the concentration value of each applicant's applicant entity, specifically as follows: In the formula, This represents the concentration value of the applicant's entities for the i-th applicant. This represents the number of patents filed by the i-th applicant in the j-th technology subclass. This represents the total number of patents applied for in the j-th technology subclass, where j = 1, 2, 3, ..., n, and n represents the number of technology subclasses. S44. Based on the numerical value of each applicant's subject concentration value, the corresponding applicants are arranged in descending order to obtain the applicant subject ranking sequence. The applicants ranked in the top 5% of the applicant subject ranking sequence are selected as priority display applicants. The text information of the corresponding valid period patent documents applied for by the priority display applicants is extracted to generate an applicant ranking information patent report.

[0012] Preferably, the specific steps of S5 include: S51. Perform feature recognition on the text information of each valid period patent document stored in the cloud database, extract the application address and legal status text information from the text information of each valid period patent document within the valid query period, compare the legal status text information of each valid period patent document with the patent status determination table stored in the cloud database, output the patent status type of each valid period patent document, and combine it with the application address of the corresponding valid period patent document to count the overall number of patents and the number of authorized patents in each province, and construct regional patent status data information. Among them, the patent status types include authorized status, substantive examination status, invalid status, and expired status. S52. Based on the ratio of the total number of patents to the number of authorized patents in each province in the regional patent status data, obtain the percentage of authorized patents in each province. Add up the total number of patents in each province to calculate the total number of patents in all provinces. Based on the ratio of the total number of patents in all provinces to the total number of patents in each province, obtain the percentage of patents in each province.

[0013] Preferably, step S5 further includes: S53. Correlate the percentage of patents held by each province with the overall percentage of patents held by the corresponding province, analyze the patent competition level of each province within the effective query period, and determine the regional competition value of each province, specifically as follows: In the formula, This represents the regional competitiveness value of the k-th province. and These represent the percentage of patents held by the k-th province and the percentage of patents held by the whole province, respectively. Here, b1 and b2 represent the weight values ​​of the percentage of patents held by the province and the percentage of patents held by the whole province, respectively. The specific values ​​are set by those skilled in the art. S54. Based on the obtained regional competition values ​​of each province, and combined with the statistical averaging algorithm, obtain the regional competition average; compare and analyze the regional competition values ​​of each province with the regional competition average. If the regional competition value of the corresponding province is greater than the regional competition average, the corresponding province is marked as a highly competitive province and marked with a red mark on the corresponding province map. If the regional competition value of the corresponding province is equal to the regional competition average, the corresponding province is marked as a moderately competitive province and marked with a yellow mark on the corresponding province map. If the regional competition value of the corresponding province is less than the regional competition average, the corresponding province is marked as a low-competition province and marked with a green mark on the corresponding province map. Based on the corresponding color markings applied to each province, a patent report on the regional competition distribution image information is generated.

[0014] A patent report intelligent generation system based on an LLM large model includes a semantic recognition module, a query confirmation module, a technical analysis module, a ranking analysis module, a regional analysis module, and a report display module; The semantic recognition module is used to perform semantic recognition on the natural semantic query statement input by the current user, and combined with the preset initial query period, to filter out a number of patent documents within the initial query period, while recording the publication time of the corresponding patent documents, and counting the number of patent documents in each unit year by year. The query confirmation module is used to perform numerical model analysis on the changing trend of the number of patent documents in each unit year, obtain the patent change value of the corresponding unit year, and obtain the average patent change value by combining statistical averaging algorithm. Based on the comparative analysis process, the number of marked patent expansion years is counted, and the proportion of expansion years in the initial query period is calculated. The query period confirmation analysis is performed on the proportion of expansion years in the initial query period to obtain the effective query time of the corresponding level. The technology analysis module is used to extract the patent change value and the number of patent documents in each unit year within the valid query duration based on the confirmed valid query duration of the corresponding class. After feature extraction, it analyzes the fluctuation level of patent changes within the valid query duration and generates a patent report on technology development trends. The ranking analysis module is used to filter patent documents within the initial query period, extract valid period patent documents, and download the text information of the corresponding valid period patent documents. After feature extraction, it analyzes the concentration of patent applications filed by each applicant in different technology subcategories within the valid query period, and generates an applicant ranking information patent report accordingly. The regional analysis module is used to analyze the patent competition level of each province within the valid query period based on the text information of the corresponding valid period patent documents after feature extraction, and generate a regional competition distribution image information patent report accordingly. The report display module stores the generated patent reports in a cloud database and marks them with corresponding patent report IDs. These reports are then displayed on the user report details page for users to view and download.

[0015] This invention provides a method and system for intelligent generation of patent reports based on an LLM large model, which has the following beneficial effects: (1) By providing patent report generation including technology development and trend analysis, applicant ranking analysis and regional analysis, compared with the existing technology which can only provide single-dimensional analysis, the present invention can meet the diversified needs of users and provide more comprehensive and in-depth patent analysis.

[0016] (2) By introducing the inflection point identification and slope calculation method in S2, the patent change value of each year is quantified, and the effective duration of the query period is dynamically determined by expanding the proportion of the year and the multi-level interval comparison mechanism, so as to realize the generation of patent reports with more responsiveness and trend sensitivity, and ensure the timeliness and accuracy of the analysis results.

[0017] (3) The user report details page allows users to view and download the report content. Users do not need to have high professional knowledge and skills to easily use and understand the report results, which lowers the user threshold. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the process of a patent report intelligent generation method based on an LLM large model according to the present invention; Figure 2 This is a block diagram of a patent report intelligent generation system based on an LLM large model according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1 Please see Figure 1 This invention provides a method for intelligent generation of patent reports based on an LLM large model, comprising the following steps: The specific steps in S1 include: S11. Based on the natural semantic query statement entered by the current user on the front-end interface, after semantic recognition by the LLM large model, extract the main information of the current user's query statement, which includes technical field keywords. The technical field keywords refer to words extracted from the user's natural language query that accurately reflect the core content of the technical topic or field of interest. These typically include technical objects, methods, uses, principles, and specific application scenarios. In this invention, the technical field keywords, as the output of the LLM large-scale model semantic recognition, are used for targeted searches in the patent resource database. For example, if a user inputs "intelligent algorithm optimization technology for electric vehicle power management," the extracted technical keywords include electric vehicles, power management, and intelligent algorithm optimization. These words can directly guide and filter relevant patent documents. The semantic recognition process includes: identifying the distribution of verbs, noun phrases, and adjective structures in the natural semantic query statement structure, removing irrelevant modifiers, and extracting the main information of the query statement; It should be noted that the LLM large model is a natural language processing model based on a deep learning architecture, which has the ability to train on a large scale of parameters and corpus, and can understand, generate and reason about natural language text. In this invention, the LLM large model is used to perform semantic recognition on the natural semantic query statement input by the user, extract keywords in the core technical field, and remove irrelevant modifiers, thereby realizing intelligent parsing of the patent query intent and laying a semantic foundation for subsequent data screening and analysis. S12. Using the technical field keywords in the main information of the query statement input by the current user, a comprehensive query is performed in the patent resource database. The preset historical query duration is used as the initial query period. Several patent documents containing the corresponding technical field keywords are selected within the initial query period, and the publication time of the corresponding patent documents is recorded. The specific value of the historical query duration is set by a person skilled in the art in a specific case. Based on the selected initial query period containing several patent documents containing keywords of the corresponding technical field, and combined with the publication time of the corresponding patent documents, the publication time of each patent document is classified by year, and the number of patent documents in each year is counted.

[0021] The specific steps in S2 include: S21. Perform a numerical model analysis on the changing trend of the number of patent documents in each unit year. Use the unit year as the horizontal axis and the number of patent documents in the corresponding unit year as the vertical axis. Based on this, establish a two-dimensional coordinate system and draw a line graph of the patent change trend. Count the total number of inflection points of the change line in the two-dimensional coordinate system of the patent change trend. Using the inflection point of the change line in the two-dimensional coordinate system of patent change trend as the dividing point, the patent change trend line graph is divided into line segments to obtain the unit change line segments for each unit year. By calculating the slope value formed between the unit change line segment for each unit year and the horizontal axis of the two-dimensional coordinate system, it is recorded as the patent change value for the corresponding unit year. Combined with the statistical mean calculation algorithm, the average patent change value is obtained. S22. Compare and analyze the patent change value of each unit year with the average patent change. If the patent change value of the corresponding unit year exceeds the average patent change, mark the corresponding unit year as a patent expansion year and count the number of patent expansion years. Conversely, mark the corresponding unit year as a patent contraction year. The total number of units per year is determined based on the preset historical query duration. The ratio of the total number of units per year to the number of patent expansion years is calculated to obtain the proportion of expansion years within the initial query period.

[0022] Specifically, the S2 steps also include: S23. Based on the proportion of expansion years in the initial query period, and pre-setting gradient comparison intervals A-range, B-range and C-range of expansion year proportion, the expansion year proportion in the initial query period is compared with the pre-set gradient comparison intervals A-range, B-range and C-range of expansion year proportion by performing interval substitution analysis, thereby obtaining the first-level query period confirmation instruction, the second-level query period confirmation instruction and the third-level query period confirmation instruction respectively. The specific process of interval substitution analysis includes: If the proportion of expansion years in the initial query period is within the preset gradient comparison range A-range, a first-level query period confirmation instruction will be generated. If the proportion of expansion years in the initial query period is within the preset gradient comparison range B-range, a second-level query period confirmation instruction will be generated. If the proportion of expansion years in the initial query period is within the preset gradient comparison range C-range, a third-level query period confirmation instruction will be generated. Based on the generated query cycle confirmation instruction of the corresponding level, the query cycle of the natural semantic query statement entered by the current user on the front-end interface is determined as the first-order effective query duration t1, the second-order effective query duration t2, and the third-order effective query duration t3, respectively, where t1 > t2 > t3, and t1, t2, and t3 are all integer multiples of the unit year duration.

[0023] The specific steps of S3 include: S31. Based on the confirmed valid query duration for the corresponding class, extract the patent change value and the number of patent documents for each unit year within the valid query duration to construct patent change data information. S32. Perform feature identification on patent change data, extract the minimum patent change value, maximum patent change value, minimum number of patent documents, and maximum number of patent documents respectively, and correlate the four data items. After dimensionless processing, analyze the fluctuation level of patent changes within the effective query period to determine the patent change fluctuation value, specifically: In the formula, This indicates the value of patent change fluctuation. and These represent the maximum and minimum values ​​of patent change, respectively. and These represent the maximum and minimum number of patent documents, respectively. Here, a1 and a2 both represent weight values, the specific values ​​of which are set by those skilled in the art. It should be noted that the patent change fluctuation value is a comprehensive evaluation indicator used to measure the activity level and fluctuation range of patent technology development within the effective query period. By combining the maximum and minimum patent changes within a unit year with the maximum and minimum number of patent documents within the corresponding unit year, after dimensionless processing, it reflects the joint fluctuation between the number of patents and the trend of change. The larger the value, the more drastic the technological changes and the more active the innovation during that period; the smaller the value, the more stable the technological development. The role of this indicator is to provide quantitative basis for subsequent trend judgment and report generation, and to help determine the current stage of development in the technological field. S33. Compare and analyze the patent change fluctuation value with the pre-set fluctuation threshold. If the patent change fluctuation value exceeds the fluctuation threshold, issue a first-level patent technology development prompt message. The prompt message is: the patent technology development trend is good within the effective query period. If the patent change fluctuation value does not exceed the fluctuation threshold, issue a second-level patent technology development prompt message. The prompt message is: the patent technology development trend is stable within the effective query period. Based on the patent technology development prompts issued at the corresponding level, a patent report on technology development trends is generated.

[0024] The specific steps of S4 include: S41. Based on the confirmed valid query duration of the corresponding class, filter each patent document within the initial query period, extract the patent documents whose publication time is within the valid query duration, and mark them as valid period patent documents. At the same time, download the text information of the corresponding valid period patent documents and send the text information of each valid period patent document to the cloud database for storage. The cloud database is also used to store the patent status determination table, wherein the text information includes the applicant's name, application address, claims, description, legal status text information and technical subclassification. S42. Perform feature recognition on the text information of each valid period patent document stored in the cloud database, extract the applicant name and technology subclass from the text information of each valid period patent document within the valid query time, construct patent technology classification data information, and count the number of patents applied for by each applicant in each technology subclass within the valid query time in the patent technology classification data information, and obtain the total number of patents applied for in each technology subclass within the valid query time by combining statistical summation algorithm.

[0025] Specifically, the S4 steps also include: S43. Correlate the number of patents filed by each applicant in each technology subclass with the total number of patents filed in each technology subclass, analyze the concentration of patents filed by each applicant in different technology subclasses within the effective query period, and determine the concentration value of each applicant's applicant entity, specifically as follows: In the formula, This represents the concentration value of the applicant's entities for the i-th applicant. This represents the number of patents filed by the i-th applicant in the j-th technology subclass. This represents the total number of patents applied for in the j-th technology subclass, where j = 1, 2, 3, ..., n, and n represents the number of technology subclasses. It should be noted that the applicant concentration value is an indicator used to measure the degree of concentration of applicants' patent applications across different technology subcategories. It is calculated by taking the ratio of the number of patents in each technology subcategorie to the total number of patents in that subcategorie and averaging the ratio across all subcategories. This reflects whether the applicant has a significant patent focus in a specific technology direction. This indicator helps to identify core applicants who have a dominant or competitive advantage in a specific technology direction. S44. Based on the numerical value of each applicant's subject concentration value, the corresponding applicants are arranged in descending order to obtain the applicant subject ranking sequence. The applicants ranked in the top 5% of the applicant subject ranking sequence are selected as priority display applicants. The text information of the corresponding valid period patent documents applied for by the priority display applicants is extracted to generate an applicant ranking information patent report.

[0026] The specific steps of S5 include: S51. Perform feature recognition on the text information of each valid period patent document stored in the cloud database, extract the application address and legal status text information from the text information of each valid period patent document within the valid query period, compare the legal status text information of each valid period patent document with the patent status determination table stored in the cloud database, output the patent status type of each valid period patent document, and combine it with the application address of the corresponding valid period patent document to count the overall number of patents and the number of authorized patents in each province, and construct regional patent status data information. Among them, the patent status types include authorized status, substantive examination status, invalid status, and expired status. S52. Based on the ratio of the total number of patents to the number of authorized patents in each province in the regional patent status data, obtain the percentage of authorized patents in each province. Add up the total number of patents in each province to calculate the total number of patents in all provinces. Based on the ratio of the total number of patents in all provinces to the total number of patents in each province, obtain the percentage of patents in each province.

[0027] Specifically, the S5 steps also include: S53. Correlate the percentage of patents held by each province with the overall percentage of patents held by the corresponding province, analyze the patent competition level of each province within the effective query period, and determine the regional competition value of each province, specifically as follows: In the formula, This represents the regional competitiveness value of the k-th province. and These represent the percentage of patents held by the k-th province and the percentage of patents held by the whole province, respectively. Here, b1 and b2 represent the weight values ​​of the percentage of patents held by the province and the percentage of patents held by the whole province, respectively. The specific values ​​are set by those skilled in the art. It should be noted that the regional competition value is a comprehensive regional indicator used to measure the level of patent competition activity in each province within the effective search period. By weighting the proportion of authorized patents of the corresponding province with the overall patent proportion, it reflects the dual performance of the corresponding province in terms of total number of patents and high-quality patents. The higher the regional competition value, the more the province not only has a large total number of patents but also a large proportion of valid patents, representing that the region has stronger innovation capabilities and competitive advantages in related technology fields. This indicator can be used to assess the patent strength of different regions. S54. Based on the obtained regional competition values ​​of each province, and combined with the statistical averaging algorithm, obtain the regional competition average; compare and analyze the regional competition values ​​of each province with the regional competition average. If the regional competition value of the corresponding province is greater than the regional competition average, the corresponding province is marked as a highly competitive province and marked with a red mark on the corresponding province map. If the regional competition value of the corresponding province is equal to the regional competition average, the corresponding province is marked as a moderately competitive province and marked with a yellow mark on the corresponding province map. If the regional competition value of the corresponding province is less than the regional competition average, the corresponding province is marked as a low-competition province and marked with a green mark on the corresponding province map. Based on the corresponding color markings applied to each province, a patent report on the regional competition distribution image information is generated.

[0028] The specific steps of S6 include: The generated patent reports are stored in a cloud database and labeled with corresponding patent report IDs. These reports are then displayed on the user report details page for users to view and download. The specific process includes: The generated patent reports are stored in a structured format in a cloud database, and each report is assigned a unique patent report ID for indexing and association. Users can access specific patent report content through the patent report ID on the report details page. The front-end interface is designed based on the Vue framework, using a responsive layout and beautiful HTML templates to display data. Internally, it uses the Vue ECharts chart rendering component, and then requests the report details data from the server by calling the Chrome kernel through PyPpeteer. After rendering, it is exported as a PDF patent report file, so that the generated PDF patent report file can almost 100% restore the loading effect of the original HTML page, providing users with a good experience.

[0029] Example 2 Please refer to Figure 1 and Figure 2 Specifically: a patent report intelligent generation system based on an LLM large model, including a semantic recognition module, a query confirmation module, a technical analysis module, a ranking analysis module, a regional analysis module, and a report display module; The semantic recognition module is used to perform semantic recognition on the natural semantic query statement input by the current user, and combined with the preset initial query period, to filter out a number of patent documents within the initial query period, while recording the publication time of the corresponding patent documents, and counting the number of patent documents in each unit year by year. The query confirmation module is used to perform numerical model analysis on the changing trend of the number of patent documents in each unit year, obtain the patent change value of the corresponding unit year, and obtain the average patent change value by combining statistical averaging algorithm. Based on the comparative analysis process, the number of marked patent expansion years is counted, and the proportion of expansion years in the initial query period is calculated. The query period confirmation analysis is performed on the proportion of expansion years in the initial query period to obtain the effective query time of the corresponding level. The technology analysis module is used to extract the patent change value and the number of patent documents in each unit year within the valid query duration based on the confirmed valid query duration of the corresponding class. After feature extraction, it analyzes the fluctuation level of patent changes within the valid query duration and generates a patent report on technology development trends. The ranking analysis module is used to filter patent documents within the initial query period, extract valid period patent documents, and download the text information of the corresponding valid period patent documents. After feature extraction, it analyzes the concentration of patent applications filed by each applicant in different technology subcategories within the valid query period, and generates an applicant ranking information patent report accordingly. The regional analysis module is used to analyze the patent competition level of each province within the valid query period based on the text information of the corresponding valid period patent documents after feature extraction, and generate a regional competition distribution image information patent report accordingly. The report display module stores the generated patent report content in a cloud database and marks the corresponding patent report ID. The report content is then displayed on the user report details page for users to view and download.

[0030] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent generation of patent reports based on an LLM large model, characterized in that: Includes the following steps: S1. Perform semantic recognition on the natural semantic query statement input by the current user, and in combination with the preset initial query period, filter out a number of patent documents within the initial query period, record the publication time of the corresponding patent documents, and count the number of patent documents in each unit year by year. S2. Perform numerical model analysis on the changing trend of the number of patent documents in each unit year to obtain the patent change value of the corresponding unit year. Combined with the statistical averaging algorithm, obtain the average patent change value. Based on the comparative analysis process, count the number of marked patent expansion years and calculate the proportion of expansion years in the initial query period. Perform query period confirmation analysis on the proportion of expansion years in the initial query period to obtain the effective query time of the corresponding level. S3. Based on the confirmed effective query duration of the corresponding class, extract the patent change value and the number of patent documents in each unit year within the effective query duration. After feature extraction, analyze the fluctuation level of patent changes within the effective query duration and generate a patent report on technology development trend information accordingly. S4. Filter the patent documents within the initial query period, extract the valid period patent documents, and download the text information of the corresponding valid period patent documents. After feature extraction, analyze the concentration of patent applications filed by each applicant in different technology subcategories within the valid query period, and generate an applicant ranking information patent report accordingly. S5. Based on the text information of the corresponding valid period patent documents, after feature extraction, analyze the patent competition level of each province within the valid query period, and generate a regional competition distribution image information patent report accordingly. S6. Store the generated patent report content in the cloud database and mark the corresponding patent report ID. Display the information on the user report details page for users to view and download.

2. The intelligent patent report generation method based on an LLM large model according to claim 1, characterized in that: The specific steps in S1 include: S11. Based on the natural semantic query statement entered by the current user on the front-end interface, after semantic recognition by the LLM large model, extract the main information of the current user's query statement, which includes technical field keywords. The semantic recognition process includes: identifying the distribution of verbs, noun phrases, and adjective structures in the natural semantic query statement structure, removing irrelevant modifiers, and extracting the main information of the query statement; S12. Using the technical field keywords in the main information of the query statement entered by the current user, a comprehensive query is performed in the patent resource database. The preset historical query duration is used as the initial query period. Several patent documents containing the corresponding technical field keywords are selected within the initial query period, and the publication time of the corresponding patent documents is recorded. Based on the selected initial query period containing several patent documents containing keywords of the corresponding technical field, and combined with the publication time of the corresponding patent documents, the publication time of each patent document is classified by year, and the number of patent documents in each year is counted.

3. The method for intelligent generation of patent reports based on an LLM large model according to claim 2, characterized in that: The specific steps in S2 include: S21. Perform a numerical model analysis on the changing trend of the number of patent documents in each unit year. Use the unit year as the horizontal axis and the number of patent documents in the corresponding unit year as the vertical axis. Based on this, establish a two-dimensional coordinate system and draw a line graph of the patent change trend. Count the total number of inflection points of the change line in the two-dimensional coordinate system of the patent change trend. Using the inflection point of the change line in the two-dimensional coordinate system of patent change trend as the dividing point, the patent change trend line graph is divided into line segments to obtain the unit change line segments for each unit year. By calculating the slope value formed between the unit change line segment for each unit year and the horizontal axis of the two-dimensional coordinate system, it is recorded as the patent change value for the corresponding unit year. Combined with the statistical mean calculation algorithm, the average patent change value is obtained. S22. Compare and analyze the patent change value of each unit year with the average patent change. If the patent change value of the corresponding unit year exceeds the average patent change, mark the corresponding unit year as a patent expansion year and count the number of patent expansion years. Conversely, mark the corresponding unit year as a patent contraction year. The total number of units per year is determined based on the preset historical query duration. The ratio of the total number of units per year to the number of patent expansion years is calculated to obtain the proportion of expansion years within the initial query period.

4. The intelligent patent report generation method based on an LLM large model according to claim 3, characterized in that: The specific steps in S2 also include: S23. Based on the proportion of expansion years in the initial query period, and pre-setting gradient comparison intervals A-range, B-range and C-range of expansion year proportion, the expansion year proportion in the initial query period is compared with the pre-set gradient comparison intervals A-range, B-range and C-range of expansion year proportion by performing interval substitution analysis, thereby obtaining the first-level query period confirmation instruction, the second-level query period confirmation instruction and the third-level query period confirmation instruction respectively. The specific process of interval substitution analysis includes: If the proportion of expansion years in the initial query period is within the preset gradient comparison range A-range, a first-level query period confirmation instruction will be generated. If the proportion of expansion years in the initial query period is within the preset gradient comparison range B-range, a second-level query period confirmation instruction will be generated. If the proportion of expansion years in the initial query period is within the preset gradient comparison range C-range, a third-level query period confirmation instruction will be generated. Based on the generated query cycle confirmation instruction of the corresponding level, the query cycle of the natural semantic query statement entered by the current user on the front-end interface is determined as the first-order effective query duration t1, the second-order effective query duration t2, and the third-order effective query duration t3, respectively, where t1 > t2 > t3.

5. The intelligent patent report generation method based on an LLM large model according to claim 4, characterized in that: The specific steps of S3 include: S31. Based on the confirmed valid query duration for the corresponding class, extract the patent change value and the number of patent documents for each unit year within the valid query duration to construct patent change data information. S32. Perform feature identification on the patent change data information, extract the minimum patent change value, the maximum patent change value, the minimum number of patent documents, and the maximum number of patent documents respectively, and correlate the four data items. After dimensionless processing, analyze the fluctuation level of patent changes within the effective query time and determine the patent change fluctuation value. S33. Compare and analyze the patent change fluctuation value with the pre-set fluctuation threshold. If the patent change fluctuation value exceeds the fluctuation threshold, issue a first-level patent technology development prompt message. The prompt message is: the patent technology development trend is good within the effective query period. If the patent change fluctuation value does not exceed the fluctuation threshold, issue a second-level patent technology development prompt message. The prompt message is: the patent technology development trend is stable within the effective query period. Based on the patent technology development prompts issued at the corresponding level, a patent report on technology development trends is generated.

6. The method for intelligent generation of patent reports based on an LLM large model according to claim 1, characterized in that: The specific steps of S4 include: S41. Based on the confirmed valid query duration of the corresponding class, filter each patent document within the initial query period, extract the patent documents whose publication time is within the valid query duration, and mark them as valid period patent documents. At the same time, download the text information of the corresponding valid period patent documents and send the text information of each valid period patent document to the cloud database for storage. The cloud database is also used to store the patent status determination table, wherein the text information includes the applicant's name, application address, claims, description, legal status text information and technical subclassification. S42. Perform feature recognition on the text information of each valid period patent document stored in the cloud database, extract the applicant name and technology subclass from the text information of each valid period patent document within the valid query time, construct patent technology classification data information, and count the number of patents applied for by each applicant in each technology subclass within the valid query time in the patent technology classification data information, and obtain the total number of patents applied for in each technology subclass within the valid query time by combining statistical summation algorithm.

7. The method for intelligent generation of patent reports based on an LLM large model according to claim 6, characterized in that: The specific steps in S4 also include: S43. Correlate the number of patents filed by each applicant in each technology subclass with the total number of patents filed in each technology subclass, analyze the concentration of patents filed by each applicant in different technology subclasses within the effective query period, and determine the concentration value of each applicant's applicant entity, specifically as follows: In the formula, This represents the concentration value of the applicant's entities for the i-th applicant. This represents the number of patents filed by the i-th applicant in the j-th technology subclass. This represents the total number of patents applied for in the j-th technology subclass, where j = 1, 2, 3, ..., n, and n represents the number of technology subclasses. S44. Based on the numerical value of each applicant's subject concentration value, the corresponding applicants are arranged in descending order to obtain the applicant subject ranking sequence. The applicants ranked in the top 5% of the applicant subject ranking sequence are selected as priority display applicants. The text information of the corresponding valid period patent documents applied for by the priority display applicants is extracted to generate an applicant ranking information patent report.

8. The intelligent patent report generation method based on an LLM large model according to claim 1, characterized in that: The specific steps of S5 include: S51. Perform feature recognition on the text information of each valid period patent document stored in the cloud database, extract the application address and legal status text information from the text information of each valid period patent document within the valid query period, compare the legal status text information of each valid period patent document with the patent status determination table stored in the cloud database, output the patent status type of each valid period patent document, and combine it with the application address of the corresponding valid period patent document to count the overall number of patents and the number of authorized patents in each province, and construct regional patent status data information. Among them, the patent status types include authorized status, substantive examination status, invalid status, and expired status. S52. Based on the ratio of the total number of patents to the number of authorized patents in each province in the regional patent status data, obtain the percentage of authorized patents in each province. Add up the total number of patents in each province to calculate the total number of patents in all provinces. Based on the ratio of the total number of patents in all provinces to the total number of patents in each province, obtain the percentage of patents in each province.

9. The method for intelligent generation of patent reports based on an LLM large model according to claim 8, characterized in that: The specific steps in S5 also include: S53. Correlate the percentage of patents held by each province with the overall percentage of patents held by the corresponding province, analyze the patent competition level of each province within the effective query period, and determine the regional competition value of each province. S54. Based on the obtained regional competition values ​​of each province, and combined with the statistical averaging algorithm, obtain the regional competition average; compare and analyze the regional competition values ​​of each province with the regional competition average. If the regional competition value of the corresponding province is greater than the regional competition average, the corresponding province is marked as a highly competitive province and marked with a red mark on the corresponding province map. If the regional competition value of the corresponding province is equal to the regional competition average, the corresponding province is marked as a moderately competitive province and marked with a yellow mark on the corresponding province map. If the regional competition value of the corresponding province is less than the regional competition average, the corresponding province is marked as a low-competition province and marked with a green mark on the corresponding province map. Based on the corresponding color markings applied to each province, a patent report on the regional competition distribution image information is generated.

10. A patent report intelligent generation system based on an LLM large model, used to implement the patent report intelligent generation method based on an LLM large model as described in any one of claims 1 to 9, characterized in that: It includes a semantic recognition module, a query confirmation module, a technical analysis module, a ranking analysis module, a regional analysis module, and a report display module; The semantic recognition module is used to perform semantic recognition on the natural semantic query statement input by the current user, and combined with the preset initial query period, to filter out a number of patent documents within the initial query period, while recording the publication time of the corresponding patent documents, and counting the number of patent documents in each unit year by year. The query confirmation module is used to perform numerical model analysis on the changing trend of the number of patent documents in each unit year, obtain the patent change value of the corresponding unit year, and obtain the average patent change value by combining statistical averaging algorithm. Based on the comparative analysis process, the number of marked patent expansion years is counted, and the proportion of expansion years in the initial query period is calculated. The query period confirmation analysis is performed on the proportion of expansion years in the initial query period to obtain the effective query time of the corresponding level. The technology analysis module is used to extract the patent change value and the number of patent documents in each unit year within the valid query duration based on the confirmed valid query duration of the corresponding class. After feature extraction, it analyzes the fluctuation level of patent changes within the valid query duration and generates a patent report on technology development trends. The ranking analysis module is used to filter patent documents within the initial query period, extract valid period patent documents, and download the text information of the corresponding valid period patent documents. After feature extraction, it analyzes the concentration of patent applications filed by each applicant in different technology subcategories within the valid query period, and generates an applicant ranking information patent report accordingly. The regional analysis module is used to analyze the patent competition level of each province within the valid query period based on the text information of the corresponding valid period patent documents after feature extraction, and generate a regional competition distribution image information patent report accordingly. The report display module stores the generated patent reports in a cloud database and marks them with corresponding patent report IDs. These reports are then displayed on the user report details page for users to view and download.

Images