Apparatus and method for constructing summary model that summarizes patent document, and patent document summarization apparatus using summary model

Abstract

A summary model construction apparatus for constructing a summary model that summarizes a patent document with respect to a certain topic, according to the present invention, comprises: a memory that stores a summary model construction program for constructing a summary model; and a processor that executes the summary model construction program stored in the memory, wherein the summary model construction program inputs a plurality of pieces of patent data into a transfer learning model through a plurality of prompts in which summary conditions for different summary topics are set, generates summary data regarding each summary topic by matching the patent data with output data of the transfer learning model that is based on the summary conditions, and constructs a summary model by transfer learning the summary data regarding each summary topic to the transfer learning model, the summary model is formed in a large language model (LLM) structure and, if at least one piece of section content of a patent document is input, summarizes and outputs input data on the basis of a certain summary topic, and the patent data includes at least one piece of section content from among a plurality of pieces of section content of a certain patent document.

Description

A device and method for constructing a summary model that summarizes patent documents, and a patent document summarization device using the summary model

[0001] The present invention relates to a device and method for constructing a summary model for summarizing patent documents, and a patent document summarization device for summarizing patent documents using the summary model.

[0002] Today, business survival and growth are determined by a balance between innovation, market exploitation, and the ability to absorb technology. In this context, patents have long served as a source of technological innovation information, serving as the foundation for trend analysis and strategy formulation.

[0003] As technology advances, patent applications increase, creating a vast amount of basic data for trend analysis and strategy formulation. However, patents are comprised of complex technical terms and contain a variety of information, requiring significant time and expense to extract the necessary technical information from a large number of patents.

[0004] Traditional technical analysis for trend analysis and strategy formulation relies on qualitative analysis. It suffers from several limitations, including the intrusion of subjective expert opinions, which incurs significant costs. To address these limitations, quantitative technical analysis has been proposed as an alternative.

[0005] Quantitative technical analysis requires users to define rules during the preprocessing stage to distinguish between the various types of technical information contained in patent documents. These rules must be structured differently depending on the required information. Quantitative technical analysis requires users to create new rules each time they analyze the required technical information, which incurs significant costs and limits scalability.

[0006] Therefore, technologies to solve these problems are required.

[0007] The present invention, in order to solve the above-mentioned problem, provides a device and method for constructing a summary model, which is formed in the structure of a large language model (LLM), and constructs a summary model that summarizes and outputs the input data based on a predetermined summary topic when the contents of at least one section of a patent document are input.

[0008] In addition, the technical task is to provide a patent document summarization device that summarizes patent documents using the summary model constructed as described above.

[0009] However, the technical tasks that this embodiment seeks to accomplish are not limited to the technical tasks described above, and other technical tasks may exist.

[0010] As a technical means for solving the above-described technical problem, a summary model construction device according to an embodiment of the present invention comprises: a memory in which a summary model construction program for constructing a summary model is stored; and a processor for executing the summary model construction program stored in the memory, wherein the summary model construction program inputs a plurality of patent data into a transfer learning model through a plurality of prompts in which summary conditions for different summary topics are set, matches the patent data with output data of the transfer learning model based on the summary conditions to generate summary data for each summary topic, and constructs a summary model by transfer learning the summary data for each summary topic to the transfer learning model, wherein the summary model is formed in the structure of a large language model (LLM), and when the contents of at least one section of a patent document are input, summarizes the input data based on a predetermined summary topic and outputs the summarized data, and the patent data includes the contents of at least one section among the contents of a plurality of sections of a predetermined patent document.

[0011] In addition, a method for constructing a summary model using a summary model constructing device according to an embodiment of the present invention includes the steps of: inputting a plurality of patent data into a transfer learning model through a plurality of prompts in which summary conditions for different summary topics are set; matching the patent data with output data of the transfer learning model based on the summary conditions to generate summary data for each summary topic; and constructing a summary model by transfer learning the summary data for each summary topic to the transfer learning model, wherein the summary model is formed in the structure of a large language model (LLM) and summarizes and outputs the input data based on a predetermined summary topic when input data including the contents of at least one section of a patent document is input, and the patent data includes the contents of at least one section among the contents of a plurality of sections of a predetermined patent document.

[0012] In addition, a patent document summarization device using a summary model according to one embodiment of the present invention includes a memory in which a summary program is stored; and a processor for executing the summary program stored in the memory, wherein the summary program receives a plurality of patent documents and at least one summary topic, inputs the plurality of patent documents into a summary model to generate summary data of each patent document based on the summary topic, and visualizes and outputs analysis data for the plurality of summary data analyzed according to analysis conditions, and the summary model is formed as a large language model (LLM) and, when input data including contents of at least one section of a patent document is input, summarizes and outputs the input data based on a predetermined summary topic, inputs a plurality of patent data into a transfer learning model through a plurality of prompts in which summary conditions for different summary topics are set, matches the patent data with output data of the transfer learning model based on the summary conditions, and generates summary data for each summary topic by transfer learning the summary data for each summary topic in the transfer learning model, and the patent data includes contents of at least one section among contents of a plurality of sections of a predetermined patent document.

[0013] According to the problem solving means of the present invention described above, by means of a summary model constructed through transfer learning, patent documents can be easily summarized based on the summary topic without the need to apply rules for distinguishing various types of technical information included in patent documents, thereby saving cost and time.

[0014] Figure 1 is a conceptual diagram schematically illustrating a summary model construction device according to one embodiment of the present invention.

[0015] Figure 2 is an example of prompts used in the summary model building process.

[0016] Figure 3 is a flowchart illustrating a summary model construction method according to one embodiment of the present invention.

[0017] FIG. 4 is a conceptual diagram schematically illustrating a patent document summary device according to one embodiment of the present invention.

[0018] Figures 5 and 6 are examples of visualizing analysis data for summary data.

[0019] Hereinafter, the present invention will be described in detail with reference to the attached drawings. However, the present invention can be implemented in various different forms and is not limited to the embodiments described herein. In addition, the attached drawings are only intended to facilitate easy understanding of the embodiments disclosed in the present specification, and the technical ideas disclosed in the present specification are not limited by the attached drawings. In order to clearly explain the present invention in the drawings, parts that are not related to the description are omitted, and the size, shape, and shape of each component shown in the drawings can be variously modified. The same / similar drawing reference numerals are assigned to the same / similar parts throughout the specification.

[0020] The suffixes "module" and "part" used in the following description for components are assigned or used interchangeably solely for the convenience of writing the specification, and do not in themselves have distinct meanings or roles. Furthermore, in describing the embodiments disclosed herein, detailed descriptions of related known technologies have been omitted if they are deemed to obscure the gist of the embodiments disclosed herein.

[0021] Throughout the specification, when a part is said to be "connected (connected, in contact with, or coupled)" to another part, this includes not only cases where it is "directly connected (connected, in contact with, or coupled)" but also cases where it is "indirectly connected (connected, in contact with, or coupled)" with another member in between. Furthermore, when a part is said to "include (have or provide)" a certain component, this does not mean that it excludes other components, but rather that it may "include (have or provide)" other components, unless otherwise specifically stated.

[0022] As used herein, ordinal terms such as "first," "second," etc., are used solely to distinguish one component from another and do not limit the order or relationship of the components. For example, the first component of the present invention may be referred to as the "second component," and similarly, the second component may also be referred to as the "first component."

[0023] Figure 1 is a conceptual diagram schematically illustrating a summary model construction device according to one embodiment of the present invention.

[0024] Referring to FIG. 1, a summary model construction device (100) according to one embodiment of the present invention will be described. The summary model construction device (100) is configured with a large language model (LLM) structure, and when the contents of at least one section of a patent document are input, the device constructs a summary model that summarizes and outputs the corresponding input data based on a predetermined summary topic. To perform such operations, the summary model construction device (100) includes a memory (110) and a processor (120).

[0025] The memory (110) stores a summary model construction program that constructs a summary model. The summary model construction program inputs multiple patent data into a transfer learning model through multiple prompts in which summary conditions for different summary topics are set, matches the patent data with output data of the transfer learning model based on the summary conditions to generate summary data for each summary topic, and transfer-learns the summary data for each summary topic to the transfer learning model to construct a summary model. Here, the patent data includes the contents of at least one section among the contents of multiple sections of a given patent document.

[0026] Meanwhile, the memory (110) should be interpreted as a general term for a non-volatile storage device that maintains stored information even when no power is supplied and a volatile storage device that requires power to maintain the stored information. The memory (110) may perform a function of temporarily or permanently storing data processed by the processor (120). The memory (110) may include a magnetic storage media or a flash storage media in addition to a volatile storage device that requires power to maintain the stored information, but the scope of the present invention is not limited thereto.

[0027] And, the processor (120) executes the summary model construction program stored in the memory (110), and when the contents of at least one section of the input patent document are input, the processor (120) trains a summary model that summarizes and outputs the input data based on a predetermined summary topic. In the present embodiment, the processor (120) may be implemented in the form of a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), etc., but the scope of the present invention is not limited thereto.

[0028] In addition, the processor (120) provides a function of executing a summary model construction program stored in the memory (110) and controlling the hardware of the summary model construction device (100) according to the execution of the program. That is, the processor (120) can perform hardware control functions such as a necessary file system, memory allocation, network, basic library, timer, device control (display, media, input device, 3D, etc.), and other utilities according to the execution of the program.

[0029] Next, the summary model building program describes the operation of building the summary model.

[0030] The summary model building program inputs multiple patent data sets into a transfer learning model via multiple prompts, each with a summary condition set for a different summary topic. The summary topic can be a topic to be summarized, such as a technical field or a technical component. The summary condition is set in the prompt and can be an example sentence of output data for the summary topic and input data. The patent data includes the contents of at least one section of a given patent document, such as the title of the invention, the technical field, the abstract, and the first claim.

[0031] The summary model is formed by the structure of a large language model (LLM), and when the contents of at least one section of a patent document are input, the input data is summarized and output based on a predetermined summary topic.

[0032] Figure 2 is an example of a prompt with summary conditions set to summarize a technical field. The prompt (10) illustrated in Figure 2 sets the technical field as the summary topic (11) and, when the title of the invention, summary, and claim 1 are input as summary conditions (12), the output data for this is set as an example sentence.

[0033] The summary model building program applies multiple patent data to the transfer learning model through multiple prompts with different summary topics and summary conditions, and generates summary data for each summary topic using the output data of the transfer learning model.

[0034] Here, the summary data is data that matches the output data of the transfer learning model with the patent data corresponding to the output data, and may be data that matches one or more of the contents of multiple sections included in the patent data with the output data of the transfer learning model.

[0035] For example, if the transfer learning model summarized multiple patent data by technical field through the prompt in FIG. 2, the summary data by technical field may be output data that matches the title, abstract, and first claim of the invention.

[0036] The summary model building program stores this summary data for each summary topic and builds a summary model by transfer learning multiple summary data for each summary topic to the transfer learning model.

[0037] The summary model building program can build a summary model using multiple transfer learning models. The summary model building program inputs multiple patent data through prompts, each with a different summary condition for a different summary topic, and generates summary data for each summary topic using the output data of each transfer learning model. Furthermore, the summary model building program can build a summary model by transfer-learning the summary data for each summary topic to one of the multiple transfer learning models.

[0038] Additionally, the summary model building program can build a summary model using a single transfer learning model. The summary model building program sequentially applies multiple prompts, each with summary conditions set for different summary topics, to the transfer learning model, and inputs multiple patent data into the transfer learning model through the prompts to perform learning on each summary topic. In this case, the transfer learning model for which learning has been completed can be set as the summary model. Furthermore, the summary model building program can receive summary data from another summary model building device (100) and build a summary model through transfer learning.

[0039] Transfer learning can utilize Low-Rank Adaptation (LORA), a learning parameter reduction technique commonly used for transfer learning of large-scale language models (LLMs). LORA is an algorithm that dramatically reduces the number of learning parameters in large-scale language models (LLMs) by decomposing the learning parameter matrix into low-dimensional data. It can also determine the hyperparameters (e.g., batch size, learning rate, epoch size) required for optimal learning to minimize the cross-entropy loss function.

[0040] Additionally, the style of the summary conditions set in the prompts for different summary topics can be set to the same style, unifying the style of the summary data and setting the style of the summary model's output data. The summary model trained in this way summarizes multiple patent documents written in different styles or perspectives using the same style. Therefore, the output data summarizing each patent document has a unified style, making it easy to utilize for future large-scale patent document analysis.

[0041] The communication module (130) may include a device including hardware and software required to transmit and receive signals such as control signals or data signals through wired or wireless connections with other network devices in order to perform data communication with external devices and signal data. The database (140) may store various data for the operation of a summary model construction program that constructs a summary model. For example, data required for the operation of or generated through the operation of a summary model construction program that constructs a summary model, such as summary data generated using output data output from a transfer learning model through a plurality of prompts in which each summary topic and summary condition is set, may be stored.

[0042] Figure 3 is a flowchart illustrating a summary model construction method according to one embodiment of the present invention.

[0043] Referring to FIGS. 1 and 3, a summary model construction method (S100) using a summary model construction device (100) according to an embodiment of the present invention is described. The summary model construction method (S100) is a method of constructing a summary model using a plurality of transfer learning models. The summary model construction device (100) inputs a plurality of patent data into a plurality of transfer learning models through a plurality of prompts in which summary conditions for different summary topics are set (step S110), and matches the patent data with output data of each transfer learning model based on the summary conditions to generate summary data for each summary topic (step S120). Then, a summary model is constructed by transfer-learning the summary data for each summary topic to one of the plurality of transfer learning models (step S130).

[0044] Here, the summary model is structured as a Large Language Model (LLM), and when input data containing the contents of at least one section of a patent document is input, the input data is summarized and output based on a predetermined summary topic. The patent data includes the contents of at least one section among multiple sections of a given patent document, and may include the contents of sections such as the title of the invention, technical field, abstract, and claim 1.

[0045] Afterwards, each step is explained in detail.

[0046] Referring to FIG. 2, a process (step S110) of inputting multiple patent data into a transfer learning model through multiple prompts is described.

[0047] The summary model building device (100) sets a plurality of prompts (10) each having a different summary topic (11) and a summary condition (12) for the summary topic (11) to each of a plurality of transfer learning models, and inputs a plurality of patent data into the transfer learning model through the set prompt (10). In the prompt (10) of Fig. 2, when a technical field is set as a summary topic (11), and the title of the invention, an abstract, and the first claim are input as input data as summary conditions (12), output data for this is set as an example sentence, and a plurality of patent data can be input into the transfer learning model through a prompt like Fig. 2.

[0048] Here, multiple transfer learning models may be stored in different summary model building devices (100) or may be stored in one summary model building device (100).

[0049] Next, the process of generating summary data for each summary topic (step S120) is described. The summary model construction device (100) generates summary data for each summary topic using the output data of the transfer learning model based on the summary topic and summary conditions set in the prompt. Here, the summary data is data that matches the output data of the transfer learning model with the patent data corresponding to the output data, and may be data that matches one or more of the contents of multiple sections included in the patent data with the output data of the transfer learning model.

[0050] For example, if the transfer learning model summarized multiple patent data by technical field through the prompt in FIG. 2, the summary data by technical field may be output data that matches the title, abstract, and first claim of the invention.

[0051] And, when explaining the process of constructing a summary model by transfer learning of such summary data (step S130), the summary model construction device (100) stores such summary data for each summary topic, and constructs a summary model by transfer learning a plurality of summary data for each summary topic to one of a plurality of transfer learning models.

[0052] When multiple transfer learning models are stored in different summary model building devices (100), summary data generated by one of the summary model building devices (100) is transmitted, and the summary model building device (100) receives summary data for each topic and performs transfer learning on the stored transfer learning model to build a summary model.

[0053] Additionally, the summary model building device (100) can also build a summary model using a single transfer learning model. Multiple prompts, each with summary conditions for different summary topics, are sequentially set in the transfer learning model, and multiple patent data can be input into the transfer learning model through the set prompts to perform learning. Finally, the transfer learning model can be set as a summary model.

[0054] FIG. 4 is a conceptual diagram schematically illustrating a patent document summary device according to one embodiment of the present invention.

[0055] Referring to FIG. 4, a patent document summarization device (200) according to one embodiment of the present invention will be described. The patent document summarization device (200) receives a plurality of patent documents and one or more summary topics, inputs the plurality of patent documents into a summary model, generates summary data for each patent document based on the summary topic, and visualizes and outputs analysis data for the plurality of summary data analyzed according to analysis conditions. To perform such operations, the patent document summarization device (200) includes a memory (210) and a processor (220).

[0056] The memory (210) stores a summary program that summarizes multiple patent documents based on a summary topic. The summary program receives multiple patent documents and one or more summary topics, inputs the multiple patent documents into a summary model, generates summary data for each patent document based on the summary topic, and outputs visualized analysis data for the multiple summary data analyzed according to analysis conditions.

[0057] Meanwhile, the memory (210) should be interpreted as a general term for a non-volatile storage device that maintains stored information even when no power is supplied and a volatile storage device that requires power to maintain the stored information. The memory (210) may perform a function of temporarily or permanently storing data processed by the processor (220). The memory (210) may include a magnetic storage media or a flash storage media in addition to a volatile storage device that requires power to maintain the stored information, but the scope of the present invention is not limited thereto.

[0058] In addition, the processor (220) executes a summary program stored in the memory (210) to summarize at least one patent document based on at least one summary topic. Specifically, the summary program receives a plurality of patent documents and at least one summary topic, inputs the plurality of patent documents into a summary model, and generates summary data for each patent document based on the summary topic.

[0059] Here, the summary model is formed by a large language model (LLM), and is constructed by inputting multiple patent data into a transfer learning model through multiple prompts in which summary conditions for different summary topics are set, matching the output data of the transfer learning model based on the summary conditions with the patent data to generate summary data for each summary topic, and transfer learning the summary data for each summary topic to the transfer learning model. The specific learning process of the summary model is the same as the method (S100) for constructing a summary model using the summary model construction device (100) described above, so a detailed description thereof is omitted.

[0060] Additionally, the processor (220) visualizes and outputs analysis data for a plurality of summary data analyzed according to analysis conditions. For example, if the analysis conditions are set to classify into similar technical fields, a plurality of patent documents can be visualized by classifying them into similar technical fields, as shown in FIG. 5.

[0061] As another example, if the analysis conditions are set to classify by function, multiple patent documents can be classified and visualized by function as shown in Fig. 6.

[0062] In this embodiment, the processor (220) may be implemented in the form of a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), etc., but the scope of the present invention is not limited thereto.

[0063] In addition, the processor (220) provides a function of executing a summary program stored in the memory (210) and controlling the hardware of the patent literature summary device (200) according to the execution of the program. That is, the processor (220) can perform hardware control functions such as a necessary file system, memory allocation, network, basic library, timer, device control (display, media, input device, 3D, etc.), and other utilities according to the execution of the program.

[0064] The communication module (230) may include a device including hardware and software required to transmit and receive signals, such as control signals or data signals, via wired or wireless connections with other network devices to perform data communication with external devices and signal data. The database (240) may store various data for the operation of the summary program. For example, data required for the operation of the summary model may be stored.

[0065] The present invention may also be implemented in the form of a non-transitory storage medium containing computer-executable instructions, such as program modules executed by a computer. Computer-readable media may be any available media that can be accessed by a computer, and include both volatile and nonvolatile media, removable and non-removable media. Computer-readable media may also include computer storage media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data.

[0066] Additionally, although the methods and systems of the present invention have been described with respect to specific embodiments, some or all of their components or operations may be implemented using a computer system having a general-purpose hardware architecture.

[0067] Those skilled in the art will appreciate that, based on the above description, the present invention can be readily modified into other specific forms without altering the technical spirit or essential characteristics thereof. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. The scope of the present invention is defined by the following claims, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be construed as being included within the scope of the present invention.

[0068] The scope of the present invention is indicated by the claims described below rather than the detailed description above, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present invention.

Claims

1. In a summary model construction device that constructs a summary model that summarizes patent literature on a given topic, A memory storing a summary model building program that builds a summary model; and Including a processor for executing a summary model building program stored in the above memory, The above summary model building program is, Input multiple patent data into a transfer learning model through multiple prompts with summary conditions set for different summary topics, match the output data of the transfer learning model based on the summary conditions with the patent data to generate summary data for each summary topic, and transfer learn the summary data for each summary topic to the transfer learning model to build a summary model. The above summary model is, It is formed with the structure of a large language model (LLM), and when the contents of at least one section of a patent document are input, the input data is summarized and output based on a predetermined summary topic. The above patent data is, A summary model construction device comprising the contents of at least one section among the contents of multiple sections of a given patent document.

2. In paragraph 1, The above summary model building program is, A summary model building device that inputs a plurality of patent data through a prompt in which summary conditions for different summary topics are set for each of a plurality of transfer learning models, and generates summary data for each summary topic using output data of each transfer learning model.

3. In paragraph 2, The above summary model building program is, A summary model construction device that performs transfer learning using the plurality of summary data on one of the plurality of transfer learning models.

4. In paragraph 1, The above summary model building program is, A summary model building device that sequentially inputs a plurality of patent data into a transfer learning model through a plurality of prompts in which summary conditions for different summary topics are set, thereby generating summary data for each summary topic.

5. In paragraph 4, The above summary model building program is, A summary model construction device that constructs the summary model by performing transfer learning using the plurality of summary data on the transfer learning model.

6. In paragraph 1, The above summary data is, A summary model building device comprising one or more of the contents of a plurality of sections of the patent data based on the summary conditions set in the above prompt and the output data of the transfer learning model.

7. A method for constructing a summary model using a summary model construction device, A step of inputting multiple patent data into a transfer learning model through multiple prompts with summary conditions set for different summary topics; A step of generating summary data for each summary topic by matching the output data of the transfer learning model based on the summary conditions and the patent data; and A step of constructing a summary model by transfer learning the summary data for each summary topic to the transfer learning model, The above summary model is, When input data including the contents of at least one section of a patent document is input and formed in the structure of a large language model (LLM), the input data is summarized and output based on a predetermined summary topic. The above patent data is, A method for constructing a summary model, the method comprising: including the contents of at least one section among the contents of multiple sections of a given patent document.

8. In paragraph 7, The steps for generating the above summary data are: For each of the multiple transfer learning models, the multiple patent data are input through a prompt with summary conditions set for different summary topics, and summary data for each summary topic is generated. The steps for building the above summary model are: A method for constructing a summary model, comprising: performing transfer learning using the plurality of summary data on any one of the plurality of transfer learning models.

9. In paragraph 7, The steps for generating the above summary data are: Multiple prompts with summary conditions set for different summary topics are sequentially applied to the transfer learning model, and the multiple patent data are input into the transfer learning model through the prompts to generate summary data for each summary topic. The steps for building the above summary model are: A method for constructing a summary model, the method comprising: constructing the summary model by performing transfer learning using summary data for each summary topic in the transfer learning model.

10. In paragraph 1, The above prompt is, The summary condition is set for any one of the above multiple summary topics, The above summary data is, A summary model construction device that matches output data of the transfer learning model with one or more of the contents of multiple sections of the patent data based on the summary conditions above.

11. In a patent literature summary device using a summary model that summarizes patent literature on a given topic, Memory where the summary program is stored; and Including a processor for executing a summary program stored in the above memory, The above summary program is, Receives multiple patent documents and one or more summary topics, inputs the multiple patent documents into a summary model to generate summary data of each patent document based on the summary topics, and outputs visualized analysis data for the multiple summary data analyzed according to analysis conditions. The above summary model is, It is formed as a large language model (LLM), and when the contents of at least one section of a patent document are input, the input data is summarized and output based on a predetermined summary topic. A plurality of patent data are input into a transfer learning model through multiple prompts with summary conditions set for different summary topics, and summary data for each summary topic is generated by matching the output data of the transfer learning model based on the summary conditions with the patent data, and the transfer learning is performed on the plurality of summary data to the transfer learning model. The above patent data is, A patent document summary device, which includes the contents of at least one section among the contents of multiple sections of a given patent document.

12. In paragraph 11, The above multiple summary data are, A patent literature summarization device, which is generated by inputting multiple patent data into multiple transfer learning models respectively through prompts with summary conditions set for different summary topics.

13. In paragraph 12, The above summary model is, A patent document summarization device constructed by transfer learning of the above plurality of summary data to one of the above plurality of transfer learning models.

14. In paragraph 11, The above multiple summary data are, A patent literature summarization device, which is generated by sequentially applying the plurality of patent data to the transfer learning model through a plurality of prompts in which summary conditions for different summary topics are set.

15. In paragraph 14, The above summary model is, A patent document summarization device constructed by transfer learning of the above-mentioned multiple summary data to the above-mentioned transfer learning model.

Images