Trademark risk management system and method thereof

Abstract

The present invention provides a trademark risk management system and method implemented by providing an electronic device operated by a user, the electronic device including a processor and a network interface controller, a server including an application, the processor connected to the server via the network interface controller, executing the application and used for category recommendation and risk management purposes, in particular for recreating text or graphics.

Description

Technical Field

[0001] The present invention relates to a trademark risk management system and method, and in particular, is used for recreating trademark texts or graphics.

Background Art

[0002] In both the conventional patent application processes in China and internationally, it is necessary to print paper documents and fill out multiple forms. Many documents are in paper form, which has been a great deal of trouble in terms of management and classification. It is not only environmentally unfriendly but also wastes a large amount of paper, and may lead to errors in the patent application process by management supervision or invalidation of patents, resulting in serious losses. In addition to the conventional document creation, if communication between personnel is required in the patent application process, due to their different professional backgrounds, language use, cultural differences and other unpredictable factors, there is a risk that information will be transmitted inaccurately or misunderstood. This leads to differences in understanding among applicants, offices, agents, and government agencies. Therefore, the applicant may not be able to achieve the originally intended results.

[0003] Also, patent applications are not only defensive weapons in patent infringement lawsuits but also aim to be symbols of a company's image. In fact, the creation of value by patent rights may not only generate license income by permitting others to implement the patent but also generate settlement income from the right to exclude infringement and the right to claim damages.

[0004] For enterprises, protecting the results of research and development through patent applications has conventionally been an essential and important part of the business process. Some companies seem to think that patent applications are the work of specialized patent law firms. Specialized patent law firms can file patent documents with accurate technical descriptions, complete content disclosure, and a wide scope of invention by inviting contractors to discuss the technical content of the patent applications.

[0005] In practice, patent applications are not only a collaborative task between a company and a patent firm, but also require communication between different departments within the company, particularly between the development department and the intellectual property department, regarding internal patent applications. This communication requires development engineers in the development department to provide relevant technical information. This information may include a specification outlining the basic background of the technology, defects or problems requiring improvement in the prior art, and characteristics of the new technology. Furthermore, it is necessary to pre-search the proposed technical content in patent databases of numerous countries to identify similar prior art, facilitating internal discussions of patent proposals. In many companies, the number of people in the development department far exceeds the number of people in the internal intellectual property department, which increases the workload of the intellectual property department.

[0006] However, despite many medium and large enterprises having appropriate systems for proposing relevant patents, in reality, communication between the development and intellectual property departments incurs significant time costs, ranging from days to months, due to specialized differences between them. While the development department is familiar with the technology, it typically cannot meet the requirements of the intellectual property department for patent proposals (patent disclosures). Conversely, when the intellectual property department discusses with the development department based on prior art search reports, it often fails to clearly and comprehensively explain the differences in a way that the development department can easily understand. Therefore, the progress of internal patent proposal (patent disclosure) discussions is extremely laborious and time-consuming. Creating patent disclosures, including drawings and claims, internally within a company typically takes at least several days to several weeks. Many companies outsource the drafting of patent specifications and filing of patent applications to third-party patent and trademark offices or law firms without patent publication, resulting in increased communication and understanding costs, such as developers having to communicate the technology again, delaying patent applications and impacting the company's technical rights.

[0007] Furthermore, for small and medium-sized enterprises (SMEs) and venture companies that lack an intellectual property department, patent searches are entirely entrusted to third-party patent and trademark offices and law firms. However, since there are no patent proposals (patent disclosures), this process is mainly conducted through presentations or chat. This model generally causes inventors to expend considerable effort and time on technical communication. The cost of communication and understanding in this process is astonishing compared to medium and large enterprises.

[0008] However, many search technologies and platforms currently available help reduce user time.

[0009] The patent drafting support system disclosed in Chinese application number CN201610297330.0 includes: a disclosure template creation module for creating a template for drafting a technical disclosure document containing multiple fields; a disclosure input module for inputting content corresponding to the multiple fields of the template in order to create a technical disclosure document; a content identification and extraction module for identifying and extracting content from each field of the technical disclosure document; a related search module connected to an external database server to retrieve related information from an external database, to select content extracted by the content identification and extraction module and associate it with the external database server; a copy storage module for copying and storing data retrieved from the external database; and a document creation module for creating a document in a predetermined format using the copied data.

[0010] However, the above Chinese application number CN201610297330.0 has several issues that need improvement. Its main feature is that it modularizes the text entered by the author into the appropriate fields, extracts keywords from the modularized text based on different fields, searches for each keyword, and finally creates search data for the author to refer to. This reduces the time spent manually entering keyword searches. However, the actual time saved for document planning, analysis and comparison, and communication between different departments within a company is quite limited, and it does not support graph generation, only applying to text.

[0011] For example, there is a system for preparing a patent specification, which includes a method mainly implemented as software or an application, as disclosed in Taiwan application number TW097119308. The present invention relates to a system capable of producing a written paper copy of a patent specification. The system described above includes a central processing means for executing a calculation process for preparing the patent specification; a data storage means connected to the central processing means and storing data for executing the patent specification preparation method; an input means connected to the central processing means and providing a user with an interface for inputting disclosure data of the relevant technology; a control means connected to the central processing means and controlling the process of preparing the patent specification data; an output means connected to the central processing means and used as an interface for outputting the patent specification data, and connectable to a printing means for printing the patent specification; and finally, a display means connected to the central processing means for displaying the outputted patent specification data.

[0012] However, the above Taiwan application number TW097119308 has several issues that need improvement. The user must enter the relevant text into the corresponding field. This method is similar to filling in blanks when creating the body of a patent specification, mainly in terms of using common terminology. However, the entered text must conform to pre-set specifications; otherwise, the resulting sentences may be grammatically incorrect, making it difficult for the average user to begin using it easily. Essentially, the user must spend a lot of time creating the document themselves, failing to solve the problem of communication documents disclosed before the patent application. Furthermore, it is limited to text and cannot assist in creating graphics.

[0013] As disclosed in U.S. Patent Application No. US17745671, in certain embodiments, the assistance system may assist a user in obtaining information or services. The assistance system allows the user to interact with it in a stateful, multi-round session using various input modes (e.g., audio, voice, text, graphics, video, gestures, motion, position, orientation, etc.) to receive assistance from the system. As an example, but not an limitation, the system may support single-mode input (e.g., voice only), multi-mode input (e.g., voice and text), hybrid / multi-mode input, or any combination thereof. The user input provided may be associated with a specific assistance task and may include, for example, a user request (e.g., a verbal request for information or the performance of a motion), interaction with an assistance application associated with the assistance system (e.g., selecting a UI element by touch or gesture), or any other appropriate type of user input that can be detected and understood by the assistance system (e.g., user motion detected by the user's client device). The assistance system may create and store a user profile containing personal and contextual information associated with the user. In some embodiments, the assistance system may use natural language understanding (NLU) to parse the user input. This analysis may achieve a more personalized and context-aware understanding based on the user's profile. The system may analyze entities associated with user input based on the analysis. In some embodiments, the assistance system may interact with different agents to obtain information or services associated with the analyzed entities. The assistance system may use natural language writing (NLG) to create responses to the user regarding information or services.

[0014] However, the aforementioned U.S. Patent Application No. US17745671 has several issues that need improvement. In the previous scenario, user input text is analyzed using natural language understanding and natural language creation to generate corresponding response actions or replies realized by a "voice assistant." However, this is applied only to general everyday life, and the language used or trained in the background data is completely different in terminology and grammar from the legal terminology of intellectual property (patents and trademarks). When the horizontal training mode is applied to the vertical industry of intellectual property, the accuracy is significantly reduced.

[0015] For example, as disclosed in the doctoral dissertation of Lee Chieh-sheng of the Institute of Computer Science and Engineering, National Taiwan University (Deep Learning Application in Patent Field 10.6342 / NTU202100999), there are problems with insufficient training scale and data volume for language models, which may lead to inaccurate training results. Furthermore, the training is not deeply focused on the patent language. This method generates summaries that heavily depend on the accuracy of the input text and derives the technical content from those summaries. This makes it easy for the text of the technical content to be biased. The comparison method used in the main text is text mapping, which has many advantages, such as capturing semantic information and contextual relationships of text, but it also has several disadvantages: (1) Dimensional disaster: Text mapping can cause high-dimensional vector representations for very large text data, which is challenging for computation and storage. (2) Difficulty in semantic identification: Text mapping may capture some semantic information, but it may not fully reflect the meaning of ambiguous or contextually related words. Semantic similarity between different texts can also be a problem. (3) Training data requirements: Text mapping methods typically require a large amount of labeled training data to learn vector representations of words and text. Obtaining large-scale, high-quality labeled data can be time-consuming and expensive. (4) Vocabulary maintenance: Text mapping methods map words to a vector space, but in real-world applications, they may encounter new vocabulary, misspellings, or other word variations. Vocabulary maintenance and updates may be necessary to ensure the accuracy of the model. Furthermore, this specification uses BERT and version 2.0 for training, but this has limitations. Communication issues between the inventor and patent employees remain unresolved.

[0016] Furthermore, current natural language processing (NLP) models such as Chat GPT® 3.5 and 4.0, developed by OpenAI, are autoregressive language models trained using reinforcement learning (RLHF) with human feedback. These are primarily used for processing customer service conversations, explaining stories, translation, grammar editing, poetry, lyrics, text compilation, and even software program creation. However, when inventors use GPT to create dialogue-based patent disclosures, they cannot directly create accurate patent industry content or relevant patent process diagrams. For inventors, it is only used as a chatbot to consult about patent regulations. Also, when using GPT to perform patent searches, only the correct patent certificate number is generated, and accurate case names and patent charts are not provided. See Figures 1 and 2. For patent employees, this may require repeated verification and can be time-consuming. Therefore, to solve this problem by assisting in the management of patent disclosures and patent searches, a deep learning model focused on the intellectual property vertical and aiming for industry accuracy is needed.

[0017] For example, similar to traditional patent drafting tools such as Patent Theory, PowerPatent, and Rowan, these tools help in creating patent documents, providing necessary support or editing functions with the assistance of professionals during the drafting process. They also contribute to maintaining consistency of terminology, such as reducing the error rate in the completed patent description, and even to suggesting reference documents or creating content. Meanwhile, the efficiency of drafting patent specifications has also improved.

[0018] However, the aforementioned conventional tools still have some shortcomings that require improvement. These tools were primarily developed for professionals in the intellectual property industry. In other words, the main target users of these tools are engineers with experience in drafting patent specifications. These tools only play a supplementary role. Therefore, developers at relatively inexperienced companies can hardly use these tools to help solve problems. Operating these tools is not intuitive for them, and they cannot create graphs. These tools cannot provide technical disclosures or create graphs for individuals or personnel who are not professionals, such as those from small and medium-sized enterprises.

[0019] From the above explanation, it is clear that improving the convenience of trademark selection for developers requires the modification or adaptation of known technologies. In order to avoid missing a timely opportunity for urgent trademark applications, the inventors conducted extensive research and ingenuity, and ultimately developed the system and method according to the present invention. [Overview of the project] [Problems that the invention aims to solve]

[0020] The present invention aims to provide a system and a method for implementing the same that solve problems in the prior art. [Means for solving the problem]

[0021] Therefore, in order to achieve the object of the present invention, the present invention provides a trademark risk management method realized by providing an electronic device operated by a user. This electronic device includes a processor and a network interface controller. There is a server that includes an application. The processor is connected to the server via the network interface controller and runs the application to perform category recommendation and risk management, and the method includes at least,

[0022] (S100) The user causes the input module to execute on the processor so as to input text content via the user interface of the electronic device, and

[0023] (S200) The processor executes the semantic analysis module in the application so as to perform semantic analysis on the text content, and

[0024] (S300) The semantic analysis module is further connected to a classification module, a search module, and a database module, classifies the text content analyzed by industrial techniques, performs a comparative search in the database module, and transmits the matching data to the intellectual property information disclosure module, and

[0025] (S400) The intellectual property information disclosure module analyzes and compiles the data and presents the information to the user via the user interface of the electronic device.

[0026] (S500) The analyzed and compiled data is also transmitted to a category recommendation module that classifies, compiles, ranks, and recommends the application intellectual property types to be displayed on the user interface.

[0027] (S600) The user selects a trademark type via the user interface, and

[0028] (S700) The user inputs a brand description via the user interface and presents it to the processor so as to execute the input module, and

[0029] (S800) By completing the login, the login module causes the processor to execute the semantic analysis module and the search module so as to create a recommended trademark category by matching with the analysis result in the database module, and

[0030] (S900) The processor includes the steps of performing a second comparative search in the database module based on the recommended trademark category and executing the search module to send the second search results to a search document creation module that creates a risk assessment report.

[0031] The process, after step (S900), further includes (S901) a step in which the risk management module recreates text or figures based on similar text or figures in the risk assessment report and conceptual information from the user received by the input module.

[0032] Another object of the present invention is to provide a system that solves problems in the prior art.

[0033] Therefore, to achieve the other objective, the present invention provides a trademark risk management system operated by a user via electronic equipment. The processor of the equipment is connected to a server via a network interface controller and runs applications for category recommendation and risk management. The system comprises at least,

[0034] An input module that receives text content entered by the user, marks it, converts it to a string, sends the string information, and records the input language of the string information in temporary memory.

[0035] A semantic analysis module that receives the aforementioned string information, analyzes it using a natural language database, marks it, creates a semantic analysis result, and transmits it,

[0036] A classification module analyzes the semantic analysis results of industry category codes, connects to a database module, and determines and creates at least one set of industry classification codes.

[0037] A search module that performs a comparative search in a database module based on at least one set of industry classification codes and creates data from the search results,

[0038] An intellectual property information disclosure module that receives the aforementioned data, performs further statistical analysis, and creates basic intellectual property information,

[0039] A recommendation module that receives the data and classifies and compiles the intellectual property rights in the data in order to create and apply recommended types of intellectual property rights,

[0040] The system includes a login module that allows the user to perform identity verification on the electronic device.

[0041] The input module receives a second trademark description entered by the user once a trademark is selected from the intellectual property types recommended by the user. The semantic analysis module receives and analyzes string information related to the brand description after identity verification by the login module, and sends the analysis results of the technical description. The search module uses the analysis results to perform a comparative search in the database module and sends the search results to the recommendation module to create recommended trademark categories. The search module performs a second comparative search in the database module based on the recommended trademark categories and sends the second search results to the search document creation module, which creates a risk assessment report.

[0042] Furthermore, it includes a risk management module that recreates text or graphics based on similar text or graphics in the risk assessment report and conceptual information received from the user via the input module.

[0043] The following detailed explanation is provided solely by specific embodiments, further illustrated with drawings. [Brief explanation of the drawing]

[0044] [Figure 1] This is a schematic diagram of the conventional technology. [Figure 2] This is another schematic diagram of the conventional technology. [Figure 3]A flowchart of the method according to the present invention is shown. [Figure 4] This is a schematic diagram of the system according to the present invention. [Figure 5] This is a schematic diagram of the system according to the present invention. [Figure 6] This is a schematic diagram of the system according to the present invention. [Figure 7] A schematic diagram of another embodiment of the system according to the present invention is shown. [Figure 8] A flowchart of the method according to the present invention is shown. [Figure 9] A flowchart of the method according to the present invention is shown. [Figure 10] A flowchart of the method according to the present invention is shown. [Modes for carrying out the invention]

[0045] The present invention provides a trademark risk management system and method. The method according to the present invention, referring to Figure 3, is performed by a user operating an electronic device. The processor of the electronic device is connected to a server via a network interface controller and runs an application for category recommendation and risk management. The method comprises at least:

[0046] (S100) The user instructs the processor to execute an input module to input text content via the user interface of the electronic device,

[0047] The (S200) processor performs the steps of executing the application's semantic analysis module to perform semantic analysis on text content,

[0048] The (S200) processor performs the steps of executing the application's semantic analysis module to perform semantic analysis on text content,

[0049] (S300) The semantic analysis module is further connected to a classification module, a search module, and a database module, and the steps include classifying the analyzed text content as industrial technology, searching the database, and transmitting the acquired data to the intellectual property information disclosure module.

[0050] (S400) The intellectual property information disclosure module performs statistical analysis of the data and presents the information to the user via the user interface of the electronic device,

[0051] (S500) The analyzed and statistically processed data is also sent to a category recommendation module that classifies and statistically analyzes the intellectual property types in the data, and displays and applies the recommended intellectual property types in the user interface.

[0052] (S600) The user selects a trademark type via the user interface,

[0053] (S700) The user instructs the processor to execute an input module to input a brand description using the user interface,

[0054] (S800) The user completes login via the login module, the processor executes the semantic analysis module, the search module performs a search in the database based on the analysis results, and the search results are sent to the recommendation module to create a recommended trademark category.

[0055] (S900) The processor includes the steps of executing the search module to perform a second search in the database based on the recommended trademark category, and sending the second search results to the document search module to create a risk assessment report.

[0056] Following step (S900), the risk management module further includes step (S901) of recreating text or shapes based on similar text or shapes in the risk assessment report and conceptual information from the user received by the input module.

[0057] For diagrams of the system according to the present invention, please refer to Figures 4 to 6.

[0058] The system according to the present invention is realized by providing an electronic device 100 operated by a user. The electronic device 100 includes a processor 101 and a network interface controller 102. A server 200 includes an application 201. The processor 101 is connected to the server 200 via the network interface controller 102 and runs the application 201 for category recommendation and risk management. The system includes at least a login module 300, an order processing module 301, an input module 302, a semantic analysis module 303, a classification module 304, a content learning module 305, a risk management module 306, a graphic creation means 309, a text creation means 3091, a document search module 310, a graphic search means 311, a text search means 3111, a database module 600, a natural language database 500, a recommendation module 802, a category recommendation module 700, and an intellectual property information disclosure module 703.

[0059] The input module, upon receiving the user's input of a brand description after they select "trademark" as the recommended intellectual property type in the recommended application, then receives the user's input again, and the login module completes identity verification. Subsequently, the semantic analysis module receives the string information related to the brand description, performs analysis and word segmentation, and creates and sends the analysis results for the technical description. The search module then compares the analysis results in the database module and sends the search results to the recommendation module to create a trademark category for the recommended application. Based on the trademark category of the recommended application, the search module performs another comparative search in the database module and sends the second set of search results to the search document creation module to create a risk assessment report.

[0060] The risk management module recreates text or shapes based on similar text or shapes from the risk assessment report and conceptual information received from the user via the input module.

[0061] Specifically, the user may connect to the server 200 via the network interface controller 102 using an electronic device 100 and run application 201. They may input text or website links. The input text may be a description of the company brand or a technical overview of the company brand's core products or services. They may input a URL link to the official website. The semantic analysis module 303 may perform semantic analysis on the website content and input text by extracting and reading the website content, and send the analysis results to the classification module 304.

[0062] The classification module 304 performs industry category code analysis on the semantic analysis results, connects to the database module, and determines and creates at least one set of trademark classification codes.

[0063] Subsequently, the classification module is connected to the database module 600 using at least one set of patent classification codes to perform comparative analysis. Information such as the number of intellectual property applications and categories related to the same or similar industries in the database module 600 is compared. This comparison generates data related to the search results.

[0064] The Intellectual Property Information Disclosure Module receives data, performs further statistical analysis, and creates basic intellectual property information. This disclosed basic information may include details such as technical value, technical risks, intellectual property time costs, and payment costs.

[0065] This invention provides a trademark risk management system and method that can automatically recreate trademark names and figures based on creative or conceptual input from an inventor. The system according to this invention, when an inventor publishes or inputs a creative idea, uses a natural language processing (NLP) algorithm to perform semantic analysis on artificial intelligence and converts the inventiveness of the input into a programming language for recreating the figure. Simultaneously, it performs figure search and comparison.

[0066] In one embodiment of the present invention, the important part is to describe the process by which a user completes intelligent recommendations and then proceeds with category recommendations, risk assessment, and risk management. However, in this embodiment, the user may also begin by receiving recommendations for intellectual property types before proceeding with category recommendations and risk management. For a fuller understanding, please refer to Figures 4 to 6. The system according to the present invention is implemented by electronic equipment 100, which includes a processor 101 and a network interface controller 102. A server 200 includes an application 201. The processor 101 is connected to the server 200 via the network interface controller 102 and runs the application 201 for category recommendations and risk management. This system includes at least a login module 300, an order processing module 301, an input module 302, a semantic analysis module 303, a classification module 304, a category recommendation module 700, a content learning module 305, a graphic learning means 3051, a risk management module 306, a text creation means 3091, a graphic creation means 309, a document search module 310, a text search means 3111, a graphic search means 311, a temporary memory 400, a database module 600, and a natural language database 500.

[0067] Server 200 may be a cloud server or a locally managed server architecture.

[0068] In this embodiment, the creator trains and runs the trademark item transformation model of this technology using a server having the following specifications: For the processor 101 (CPU), it is a high-performance multi-core processor with at least 16 cores, such as the AMD Ryzen® Threadripper® or Intel Xeon® series. Such a CPU selection is particularly important for processing large amounts of data and performing complex calculations. For memory (RAM), the memory size may be 64GB or more to accommodate the size of the corpus data and word vector model. Network interface controller 102: High-speed and reliable network connectivity hardware, used particularly in conjunction with cloud computing resources or when transmitting large amounts of data. For the GPU (Graphics Processing Unit), a highly efficient GPU such as the NVIDIA RTX30 series or Tesla® series is used to reduce model training time. In such cases, the training server used in the server 200 architecture uses the higher specifications described above for training purposes. However, once model training is complete and inference processing is performed, a server host with lower specifications may be used, either in the cloud or locally. The choice of server host specifications does not affect the technical features emphasized in this case, and any server host specification should fall within this technical scope.

[0069] The login module 300 allows the user to perform identity verification when operating the electronic device 100. It verifies the user's ID and other identification information.

[0070] The order processing module 301 provides users with functions such as member registration / login, creation and management of international patent orders, and input of international patent orders. It also provides functions for checking order status, checking the examination status of patents and trademarks, and checking accounts.

[0071] Specifically, users may register and log in as members of this system to manage patent applications and case orders on the platform. This includes tasks such as creating documents, receiving cases, submitting case copies, and notifying formal documents.

[0072] The user may create a new case order in the order processing module 301. Once the user selects the first target country, the system's temporary memory 400 is periodically updated with information from the case order.

[0073] The input module 302 is responsible for receiving explanatory text or figures, particularly their creative concepts, entered by the user. It converts the explanatory text into a string for marking and transmits the string information. The input language is recorded in the temporary memory 400.

[0074] Specific implementations may include text preprocessing, tag extraction, serialization, and tagging. Text preprocessing may further include tokenization, stop word removal, part-of-speech tagging, and part-of-speech wording. Tagging involves first extracting tags using predetermined rules or patterns, and then performing automated tag extraction using a machine learning model. Columnization involves grouping the extracted labels into a single column, such as "movie#action#drama". Tagging may be performed based on specific requirements and may include removing duplicate tags, classifying tags, and other operations. More specific methods may involve using NLTK, Stanford NLP, HanLP, machine learning frameworks, and natural language processing (NLP) toolkits such as TensorFlow®, PyTorch, and scikit-learn.

[0075] The semantic analysis module 303 receives string information and processes it for analysis and marking via the natural language database 500. It then creates and transmits the results of the semantic analysis.

[0076] The classification module 304 analyzes trademark category codes based on the results of semantic analysis. It connects to the database module 600 and determines and creates at least one set of trademark classification codes or trademark categories. This classification analysis may include internationally recognized trademark Nice classifications and / or trademark classifications specific to various countries, such as the Taiwanese trademark classification.

[0077] To perform classification analysis and create trademark categories for a single paragraph of explanatory text, the following methods can be considered.

[0078] Text preprocessing: First, the input explanatory text is preprocessed. This may involve removing punctuation, stop words, and meaningless characters, as well as performing lexical normalization (e.g., stemming or semanticization).

[0079] Feature extraction: This involves extracting meaningful features from preprocessed text. This can be achieved using common text feature representation methods such as bag-of-words or word embeddings. Bag-of-words represents text as word frequency or presence, while word embeddings map words to a continuous vector space.

[0080] Trademark classification model: Construct an industry classification model, which may be based on machine learning classifiers such as support vector machines, decision trees, or deep learning models, or on rule-based methods. Train the model to classify text into the corresponding industry categories.

[0081] Trademark Category Creation: Using a pre-trained classification model, the entered descriptive text is classified, and the corresponding trademark classification code is generated. The Nice Classification is an international standard used for trademark classification.

[0082] It is assumed that you have explanatory text.

[0083] This company specializes in the development of new materials and technologies for solar power generation. They design highly efficient solar cells that convert solar energy into electrical energy while reducing energy costs and environmental impact. These technologies have potential in the renewable energy sector and are widely applied to residential and industrial purposes.

[0084] Trademark classification analysis and trademark categories may be created by following the steps below.

[0085] Text preprocessing: Preprocess the explanatory text by removing punctuation, converting to lowercase, and eliminating stop words. For example, the above explanatory text would be preprocessed to read: "Company specializes in the development of new materials and technologies for solar power generation, high-efficiency solar cell design, solar energy conversion, energy cost reduction, environmental impact reduction, renewable energy field potential, housing industry purpose, and wide application."

[0086] Feature Extraction: Preprocessed text using a Bag-of-Words model is converted into a feature vector representation. Each word may be considered a feature, and the frequency of word occurrences in the explanatory text may be used to represent their importance. For example, the following feature vector representation can be obtained.

[0087] {"Company":1, "Specialization":1, "Research":1, "Development":1, "Solar":2, "Power Generation":1, "New":1, "Materials":1, "Technology":2, "Design":1, "High Efficiency":1, "Battery":1, "Conversion":1, "Power":1, "Reduction":1, "Energy":1, "Cost":1, "Environment":1, "Impact":1, "Renewable Energy":1, "Potential":1, "Widely":1, "Application":1, "Household":1, "Industrial":1, "Application":1}.

[0088] Trademark classification model: A trained classification model is constructed using marked training data with corresponding labels for model training, such as a support vector machine, recurrent neural network, or deep learning model like a convolutional neural network.

[0089] Category Creation: Pre-processed text feature vectors are input and predictions are made using a trained classification model. Based on the prediction results, corresponding trademark categories may be created. For example, the model may predict that the text in the specification belongs to the solar energy-related industry category, and the corresponding Nice classifications are categories 04, 09, 37, and 40, etc.

[0090] The category recommendation module 700 uses a computation model trained on a natural language model, a trademark classification table, and subcategories. It analyzes string information with ambiguous meanings (or inaccurate descriptions) into trademark category recommendation information and combines it with the semantic analysis module 303 and the classification module 304 to ultimately create recommended trademark categories for the application.

[0091] Natural language models can predict the next word or construct sentences that fit grammar and meaning based on known text data. This includes, but is not limited to, the following:

[0092] N-gram model: The N-gram model is a probabilistic language model that assumes the probability of a word appearing depends only on the previous N-1 words. For example, in a binary model (N=2), the probability of the next word is predicted based on the previous word.

[0093] Recurrent Neural Network (RNN) Models: RNNs are neural networks well-suited for handling sequence data, allowing them to capture time dependencies between words. In natural language processing, RNNs are typically used to build language models that treat each word as a time step.

[0094] Pre-trained language models (e.g., BERT): Pre-trained language models are models trained on large, unsupervised datasets to understand and construct natural language. BERT models, based on a converter architecture, are pre-trained on a wide range of text data and then fine-tuned for specific tasks (e.g., text classification, named entity recognition, etc.).

[0095] The trademark classification tables and subcategories are extracted from trademark databases in various countries. These classifications include all categories of goods and services listed, which are extracted and compiled into a list. This list is then used for training and calculations with natural language models.

[0096] The document search module 310 includes a text search means 3111, a graphic search means 311, a conversion means 317, and a graphic matching means 318. The graphic search means 311 receives a graphic input by the user and performs a search in the database module 600, and the text search means 3111 receives input text and performs a search in the same database module 600 to obtain the searched document.

[0097] Specifically, when the shape search means 311 receives an input shape, it first uses the conversion means 317 to convert the shape and adds a Vienna classification tag. At the same time, the shape is converted into a vector. Then, the shape matching means 318 compares it with the database module 600 and creates a search document.

[0098] We convert the shape into a vector, and assume, for example, that the shape is basically made up of pixels, of which the brightest pixel is considered to be 1 and the darkest pixel is considered to be 0, and therefore we do not consider the three primary colors. For a 3x3 pixel shape, the top left, left center, top center, left center, center center, right center... bottom right corner are (1, 0, 1, 0, 1, 0, 1, 0, 1). In this case, the shape is black-white-black-white-black-white-black-white-black. By considering the three primary colors, we can easily triple the length of this vector and convert it to R(1,0,1,0,1,0,1,0,1)+G(1,0,1,0,1,0,1)+B(1,0,1,0,1,0,1,0,1)=(1,0,

[0099] Specifically, during the matching process, similarity measures such as edit distance, cosine similarity, or Jaccard similarity may be calculated in different forms. Shapes with a similarity level exceeding a set threshold may be filtered, and this threshold may be adjusted as needed.

[0100] Edit distance is typically used to measure the similarity between columns. It quantifies the minimum number of operands (insertions, deletions, substitutions) required to transform one string into the other. However, the same applies to the similarity between approximation vectors.

[0101] The following is a general procedure for calculating vector similarity using edit distance.

[0102] Vector serialization: Converts each vector into a sequence. For example, a digital vector can be directly converted into a digital sequence. A feature vector can also be a combination of the names and values ​​of each feature into a column sequence.

[0103] Calculating the edit distance: The edit distance between two sequences is calculated using an edit distance algorithm (e.g., Levenshtein distance).

[0104] Normalization: To facilitate comparison, edit distances are normalized so that they range from 0 to 1. A common normalization method is to divide by the maximum sequence length.

[0105] Similarity: Indicates the similarity between two vectors (1 - normalized edit distance). Generally, the closer the value is to 1, the more similar the vectors are.

[0106] For example, let's assume we have two vectors: vector 1 = [1, 2, 3, 4] and vector 2 = [2, 3, 1, 4].

[0107] Serialization: Vector 1 -> "1234", Vector 2 -> "2314"

[0108] Edit distance: 2 (Requires two operations: swapping "1" and "3")

[0109] Normalization: 2 / 4=0.5

[0110] Similarity: 1-0.5=0.5

[0111] Therefore, the similarity between these two vectors is 0.5, indicating a certain degree of similarity.

[0112] Cosine similarity is a method for measuring the directional similarity between vectors and is used to estimate the similarity between them. The principle is to calculate the cosine value of the angle between two vectors. The closer the angle is to 0, the more similar the vectors are, and the closer the cosine value is to 1.

[0113] The following is a general procedure for calculating vector similarity using cosine similarity.

[0114] Vector preprocessing: Ensure that the two vectors have the same dimension and normalize the vectors to eliminate the effect of differences in vector lengths (for example, by setting their lengths to 1).

[0115] Dot product calculation: Calculate the dot product between two vectors, multiply the corresponding elements in pairs, and then sum them up. The dot product reflects the directional correlation between the two vectors.

[0116] Calculating vector length: This involves calculating the length of each vector, finding the sum of the squares of the corresponding elements, and then taking the square root. The vector length reflects its overall magnitude.

[0117] Calculating cosine similarity: To obtain the cosine similarity between two vectors, divide the dot product by the product of the lengths of the two vectors.

[0118] Similarity Analysis: Cosine similarity ranges from -1 to 1. A cosine similarity close to 1 indicates that the two vectors have very similar directions and are highly similar. A cosine similarity of 0 means that the two vectors are orthogonal and not similar at all. A cosine similarity close to -1 indicates that the two vectors have opposite directions and are not similar in height.

[0119] Assume there are two vectors: vector 1 = [2, 3, 4] and vector 2 = [4, 6, 8].

[0120] Vector preprocessing: Since these two vectors have the same dimension, no further processing is required.

[0121] Calculation of the dot product: 24 + 36 + 4 * 8 = 62

[0122] Calculation of vector length: sqrt(2^2+3^2+4^2)=sqrt(29), sqrt(4^2+6^2+8^2)=sqrt(100).

[0123] Calculation of cosine similarity: 62 / (sqrt(29)*sqrt(100))≈0.906

[0124] Therefore, the cosine similarity between these two vectors is close to 0.9, indicating a very high level of similarity in direction.

[0125] The Jaccard similarity measure is a method for measuring the similarity between vectors and is used to estimate the similarity between them. The principle is to calculate the ratio of common elements between two vectors. A high ratio indicates high similarity and results in a high Jaccard similarity score.

[0126] The following is a general procedure for calculating vector similarity using the Jaccard similarity index.

[0127] Vector preprocessing: Ensure that the two vectors have the same dimension. Normalize the vectors to eliminate the effect of differences in vector lengths (for example, by making the sum of their elements equal to 1).

[0128] Finding the intersection: Count the number of common elements between two vectors.

[0129] Calculating the union: Count the total number of elements in two vectors.

[0130] Calculating Jaccard similarity: To obtain the Jaccard similarity between two vectors, divide their intersection by their union.

[0131] Similarity Analysis: The Jaccard similarity score ranges from 0 to 1. A Jaccard similarity score close to 1 indicates that the two vectors are identical and have a high degree of similarity. A Jaccard similarity score of 0 means that the two vectors are not perfectly similar.

[0132] Assume there are two vectors: vector 1 = [1, 2, 3, 4] and vector 2 = [1, 3, 4, 5].

[0133] Vector preprocessing: Since these two vectors have the same dimension, no further processing is required.

[0134] Finding the intersection: 1 + 3 + 4 = 8

[0135] Calculation of the union: 1+2+3+4+5=15

[0136] Jaccard similarity calculation: 8 / 15 ≈ 0.533.

[0137] Therefore, the Jaccard similarity between the two vectors is close to 0.5, indicating a certain degree of similarity.

[0138] The text search means 3111 receives semantic analysis results from the semantic analysis module and performs matching analysis, such as trademark name similarity matching analysis, in the database module 600.

[0139] The search document creation module 310 combines matching results from the graphic search means 311 and the text search means 3111 to create a search document, further classify and filter cases with a similarity higher than a predetermined risk threshold, and create a risk assessment report.

[0140] The content learning module 305 is a large-scale language model that learns different trademark content from the database module 600 for different trademark categories. The graphic learning means 3051 further learns trademark graphic designs used for different trademark categories, such as graphic designs, text designs, and unique words.

[0141] The risk management module 306 creates text and / or graphics based on learning from the content learning module 305 and past cases matched by the graphic search means 311 and the text search means 3111.

[0142] Database module 600 provides multiple databases necessary for searching for trademark pre-registration cases in multiple countries within the system, including official trademark databases from various countries, trademark category databases from multiple countries, and Nice category databases.

[0143] The graphic creation means 309 receives user input regarding trademark names, graphic changes, or concepts via the input module 302. The graphic creation means analyzes the concept and combines it with a semantic analysis module to translate it into a graphic creation language. For the concept, it creates a corresponding graphic code, compiles it, and creates a graphic that matches the concept. The graphic search means 311 compares the similarity between the recreated graphic and past examples that have been previously searched during the process of recreating the graphic, and ensures that the similarity between the recreated graphic and past examples is below a predetermined threshold.

[0144] The text creation means 3091 receives user input regarding changes to trademark names or figures or concepts via the input module 302. The text creation means 3091 combines with a semantic analysis module to analyze the concepts and with a content learning module to create text that is the trademark name. Simultaneously, the text search means 3111 compares the similarity of the recreated text with previously searched past cases so that the similarity of the recreated text with past cases is below a predetermined threshold.

[0145] The risk management module 306 combines recreated concepts to mitigate application risks. After a risk assessment report is generated, the user may choose whether or not to allow the system to participate in risk management. In other words, the user may decide whether or not to allow the system to recreate text or graphics to avoid text or graphics from past cases.

[0146] Specifically, translating conceptual ideas into a graphical language involves several steps. First, the translation module needs to analyze the extracted conceptual ideas in order to understand their meaning. Then, the translation module needs to create a graphical language creation manual based on the grammatical rules of the language. More specifically, the following steps are used to create a graphical language, including technical idea identification, technical feature identification, and technical relationship identification.

[0147] Technical Concept Identification: First, the translation module identifies the technical concepts within the technical information. These technical concepts are fundamental elements in the graphic creation language.

[0148] Technical Feature Identification: Next, the translation module identifies the technical features in the technical information. Technical features are used to explain the technical concepts.

[0149] Technical Relationship Identification: Finally, the translation module identifies technical relationships in the technical information. Technical relationships describe the connections between technical concepts.

[0150] In practice, the semantic analysis module 303 may perform keyword extraction, which is a natural language processing technique aimed at automatically extracting important keywords or phrases from text. Methods for keyword extraction may include, but are not limited to, statistical methods, frequency-based methods, text vectorization methods, or machine learning methods.

[0151] Frequency-based methods determine word importance based on the frequency of words in a text. Common methods include TF-IDF (Term Frequency-Inverse Document Frequency) and word frequency. TF-IDF considers both the frequency of a word in a text and its importance within the overall document set, while word frequency only considers the frequency of a word in a text.

[0152] Text statistical methods analyze the distribution and relevance of words in text, relying on statistical models. Common methods include mutual information, pointwise mutual information, and chi-squared tests. These methods typically require establishing a statistical model between words and text and calculating word importance based on that model.

[0153] Text vectorization methods convert text into vector representations and then calculate word importance using vector space models. Common methods include bag-of-words models, word embeddings, and text vectorization techniques such as TF-IDF vectorization.

[0154] Keyword expansion is another feature of keyword extraction. It receives the generated keywords, finds similar or synonymous words, expands them, and connects them to a synonym library to create additional synonyms.

[0155] The system further includes a language detection module 320 for determining whether the language of a string entered by the user matches the official language of a first target country. If the language detection module 320 determines that the string information does not match the official language of the first target country, the translation module 321 translates the string information. After creating a search document, the translation module 321 translates the language of the search document back into the language of the original string information.

[0156] The system also includes a case processing module 312 that executes system case orders and prepares them to conform to trademark application documents. The system further includes a multi-country translation module 324 that can perform multi-country translation of trademark application documents and can also integrate system case orders with third-party electronic payment services.

[0157] The case processing module 312 further includes an application form creation means 323. The application form creation means 323 either extracts user login authentication ID information and integrates it into application data, or the user may directly input the application data received by the input module 302 into the fields using the electronic device 100. Subsequently, the application data is created as an application form using an application format template.

[0158] After the application form is created, the user can operate the electronic device 100 to submit the application online or manage the case online using the case processing module 312. Submission is completed by the user directly submitting the application to the authority or by selecting the submission function, after which the submission function is handled by the law firm of the providing system, taking into account different application document formats and file formats of different countries.

[0159] The system further includes a multi-country conversion module 324. After the application form has been prepared, the user may use the multi-country conversion module 324 to select a second target country in order to file a trademark application in a different country.

[0160] When a second target country is selected by the user, the language detection module 320 first determines whether the language of the string information matches the official language of the second target country. If they do not match, the translation module 321 translates the trademark name into the official language of the second target country. After translating the application form into the official language of the second target country, it then adjusts the format.

[0161] Alternatively, the language detection module 320 may first determine whether the language of the application form matches the official language of the second target country. If they match, the module 321 adjusts the instruction file and the application form as they are, according to the official file format of the second target country. If they do not match, the translation module 321 translates the application form into the official language of the second target country and then adjusts the format.

[0162] In other embodiments of the present invention, see Figures 4 and 7. Figure 7 shows another schematic diagram of a system according to the present invention, which includes a data receiving module 330, a semantic analysis module 303, a keyword extraction module 331, a search module 332, a classification analysis module 333, a text creation module 334, a content learning module 305, a keyword expansion module 335, a document classification module 336, an automatic writing module 337, and a search report creation module 338.

[0163] The data receiving module 330 may include a login module 300, an order processing module 301, and an input module 302. The login module 300 allows the user to verify their identity using the electronic device 100, verifies the user's identity (logged-in user), and confirms their identification information. The order processing module 301 provides the user with functions such as member registration / login, creation and management of multinational trademark order, and introduction of multinational trademark order, as well as functions such as order inquiry, trademark examination status confirmation, and account inquiry. The input module 302 receives explanatory text or figures entered by the user, converts the explanatory text into a string for marking processing, transmits the string information, and records the input language of the string information in the temporary memory 400.

[0164] The semantic analysis module 303 receives string information, analyzes it using the natural language database 500, marks it, creates a semantic analysis result, and transmits it.

[0165] The keyword extraction module 331 performs word segmentation and keyword extraction based on the input content to create multiple keywords.

[0166] Specifically, keyword extraction is a natural language processing technique for automatically extracting important keywords and phrases from text. Methods may include, but are not limited to, statistical methods, frequency-based methods, text vectorization methods, or machine learning methods.

[0167] Frequency-based methods (e.g., TF-IDF (Term Frequency-Inverse Document Frequency) and Term Frequency (TF)) determine word importance based on the frequency of words in a text. TF-IDF considers both word frequency in the text and importance within the overall document set, while TF considers only word frequency in the text.

[0168] Statistical methods in text analysis involve statistical models that analyze the distribution and relationships of words in a text. Common methods include mutual information, pointwise mutual information, and chi-squared tests. These methods typically require establishing a statistical model between words and text and calculating the importance of words based on that model.

[0169] Text vectorization methods convert text into vector representations and calculate word importance using a vector space model (VSM). Common methods include Bag-of-words (BoW) models, word embeddings, and text vectorization methods such as TF-IDF vectorization.

[0170] Machine learning methods use machine learning algorithms to train models that learn word importance from text. Common methods include text classification, text clustering, and keyword extraction models. These methods require model training using marked text data.

[0171] The method described above may also be implemented by creating a script in a programming language such as Python®.

[0172] The search module 332 includes a text search means 3111, a graphic search means 311, a conversion means 317, and a graphic matching means 318. The graphic search means 311 receives a graphic input by the user and searches the database module 600 to find matching trademark cases. When the graphic search means 311 receives an input graphic, it first converts the graphic using the conversion means 317, adds a Vienna classification tag, and converts the graphic into a vector. Then, the graphic matching means 318 performs a comparison based on the database module 600 and searches for matching trademark cases. It also extracts past cases with a similarity score higher than a certain risk threshold.

[0173] The classification analysis module 333 may analyze the industry category codes of the input text using the classification module 304. It connects to the database module 600 and determines and creates at least one set of trademark classification codes, which may also be trademark categories.

[0174] The text creation module 334 may include the risk management module 306 and the search document creation module 310, and may further include text creation means 3091 and graphic creation means 309. The search document creation module 310 may be part of the search module 332 described above.

[0175] The risk management module 306 receives conceptual ideas and, based on the semantic analysis results combined with the content learning module 305, generates text and / or figures.

[0176] The keyword expansion module 335 receives and expands keywords created by the keyword extraction module 331, and means searching for similar words or synonyms by generating additional synonyms by comparing them with a glossary.

[0177] The document classification module 336 may classify all case files created by the user, for example, by applying specific classification labels. The created cases are classified into disclosure documents or search documents, and patents or trademarks, etc.

[0178] Furthermore, the document classification module 336 may classify similar figures retrieved by the figure search means 311 in the database module 600. Similar figures are classified based on these similarity levels. For example, figures with a similarity score higher than the first risk threshold are classified into the high-risk region, figures between the first and second risk thresholds are classified into the medium-risk region, and figures lower than the second risk threshold are classified into the low-risk region.

[0179] The automatic writing module 337 may be integrated with the risk management module 306 and the content learning module 305 to automatically write and create application documents.

[0180] The search report creation module 338 may instead be replaced by the aforementioned search document creation module 310, disclosure document creation means 307, claim content creation means 308, or graphic creation means 309.

[0181] For an example flowchart of the execution method by the system, see Figures 8 to 10. The method involves a user operating an electronic device 100, the processor 101 of the electronic device 100 being connected to a server 200 via a network interface controller, and executing an application 201 for recommended categories and risks, comprising at least:

[0182] (1) The user logs in using an electronic device 100, undergoes identity verification, and confirms the user's ID and user data.

[0183] (2) The user creates an incident order using the electronic device 100 and selects a first target country, and the information within the incident ranking is periodically updated in the temporary memory 400,

[0184] (3) The user inputs explanatory text or figures using the electronic device 100, converts the explanatory text into a string, marks it, and creates string information. The input language of the string information is recorded in the temporary memory 400.

[0185] (4) The semantic analysis module performs semantic analysis on string information, and the classification module includes the step of classifying the string information into trademark categories.

[0186] In step (4), further,

[0187] (411) The steps include: analyzing and splitting string information using a semantic analysis module to create semantic analysis results;

[0188] (412) A step of creating at least one set of trademark classification codes based on the semantic analysis results using the classification module 304,

[0189] (413) The category recommendation module 700 includes the step of combining it with the semantic analysis module 303 and the classification module 304 to analyze string information having an ambiguous meaning (or an inaccurate description) into trademark category recommendation information and finally create a recommended trademark category.

[0190] In step (3), further,

[0191] (31) The language determination module includes the step of determining whether the input language of the string information is the same as the official language of the first target country,

[0192] (32) If so, proceed with the semantic analysis step,

[0193] (33) Otherwise, the process includes the steps of first translating the string information into the official language of a first target country using a translation module, and then performing semantic analysis. After step (413), the created recommended trademark categories are translated into the input language of the string information.

[0194] After step (413), further,

[0195] (421) The graphic search means and text search means 3111 receive text and / or graphics entered by the user, perform a search in the database module 600, compare and search for past trademark examples, create a search document, and further filter past examples with high similarity.

[0196] (422) The shape search means receives an input shape, first converts it using a conversion means, and adds a Wien classification label to convert the shape into a vector,

[0197] (423) The process includes the steps of: (423) The figure matching means compares the input figure with past examples in the database module, selecting the figure whose similarity score exceeds a predetermined value as a past example of a trademark; and the text search means 3111 performs text matching in the same manner.

[0198] (5) Based on the learning of the content learning module 305, the risk management module 306 creates text and / or figures based on past cases matched by the figure search means 311 and the text search means 3111 in the database module 600.

[0199] Furthermore,

[0200] (51) The user inputs a trademark name or a conceptual idea of ​​a figure using the input module 302, the figure creation means 309 combines with a semantic analysis module to analyze the conceptual idea and translate it into a figure creation language, creates a corresponding figure code for the conceptual idea, then creates a figure corresponding to the conceptual idea using the compilation means, and the figure search means 311 compares the created figure with past examples that have been searched in the past during the figure creation process, ensuring that the similarity of the created figure is less than a predetermined value.

[0201] (52) The text creation means 3091 is combined with a semantic analysis module to analyze a conceptual idea and create text that is a trademark name, and further includes the step of using the text search means 3111 to compare the created text with past cases that have been searched in the past and ensuring that the similarity of the created text is less than a predetermined value.

[0202] (6) The application document creation means either extracts the user's login authentication identification data as application data, or allows the user to create an application document by directly entering the application data into the fields.

[0203] After creation, the case order may be completed, or the patent may be filed for application.

[0204] Also, after step (6), the following:

[0205] (7) The user takes the step of selecting a second target country,

[0206] (71) The language determination module includes the step of determining whether the language of the application form is the same as the official language of the second target country,

[0207] (72) Otherwise, the process involves first translating the application document into the official language of the second target country using a translation module, and then adjusting the format of the translated normative document and the application document using an application document preparation means.

[0208] (73) If so, the format is adjusted as is. This includes the step of completing the conversion of the international patent application after the format has been adjusted and submitting it to an official in the second subject country by a law firm.

[0209] Once the formatting adjustments are complete, the conversion of the international patent application is finished, and the application is submitted within the system. Then, it is handed over to a law firm for formal submission to the authorities of the second target country.

[0210] Finally, the technical features and feasible technical effects of the present invention are summarized below.

[0211] First, the trademark risk management system and method according to the present invention solve common difficulties faced in the field of intellectual property, particularly those who lack the concept of protecting their own technologies through intellectual property rights, by providing intelligent recommendations.

[0212] Next, according to the trademark risk management system and method of the present invention, more time and cost are saved by providing a rapid recommendation, risk system, and method to those who have difficulty deciding which category to file an application in, even after determining the type of intellectual property.

[0213] Thirdly, the trademark risk management system and method according to the present invention solve the problems faced by small and medium-sized enterprises and venture companies that lack an intellectual property department, and by reducing the effort and time required for brand dissemination, as well as the costs of dissemination and understanding during that time, unproposed scenarios can be avoided.

[0214] Fourth, the trademark risk management system and method according to the present invention provides a complete system and implementation method from the initial stages of a technology brand, recommends intellectual property types, and ultimately creates recommended categories and risks. This significantly reduces the professional hurdles for individuals to obtain intellectual property protection for their specialized technologies.

Claims

1. A trademark risk management method in which an electronic device is operated by a user, the processor of the electronic device is connected to a server via a network interface controller, and an application program is executed to perform category recommendation and risk management, (S100) The user inputs explanatory text or figures via the user interface of the electronic device, and the processor executes an input module to accept the explanatory text or figures. (S200) The processor performs the steps of executing a semantic analysis module in the application program to analyze the explanatory text, (S300) The semantic analysis module is further connected to a classification module, a search module, and a database module, classifies the explanatory text using industry technology, performs a matching search in the database module, and transmits the matching data to the intellectual property information disclosure module. (S400) The intellectual property information disclosure module analyzes the matching data and compiles it into intellectual property information, and presents the intellectual property information to the user via the user interface of the electronic device, (S500) The analyzed and compiled data is also classified and compiled by intellectual property type in the data, and the recommended intellectual property types are displayed in ranking order on the user interface and sent to a category recommendation module for application. (S600) The user selects a trademark type via the user interface, (S700) The user inputs a brand description via the user interface, and the processor executes an input module to input the brand description. (S800) The user completes the login process using the login module, the processor executes the semantic analysis module, the search module performs a matching search against the analysis results in the database module, and sends the search results to the recommendation module to create recommended application trademark categories. The (S900) processor executes the search module to perform a second matching search in the database module based on the recommended application trademark category, and the second search results are sent to the search document creation module to create a risk assessment report. Trademark risk management methods including those mentioned above.

2. After step (S900), further, (S901) The trademark risk management method according to claim 1, comprising the step of recreating trademark text or figures based on similar trademark text or figures in a risk assessment report and conceptual information from a user's brainstorming session received by an input module.

3. A trademark risk management system for receiving clients that accept users by operating electronic devices, The processor of the aforementioned electronic device is connected to a server via a network interface controller and executes application programs for category recommendation and risk management, and at least, An input module that receives explanatory text or figures entered by the user, converts the explanatory text into a string, performs marking processing, transmits the string information, and records the input language of the string information in temporary memory. A semantic analysis module that receives the aforementioned string information, analyzes and segments the string information using a natural language database, creates and transmits semantic analysis results, A classification module that analyzes the industry category classification codes used in the semantic analysis results, connects to a database module, and determines and creates at least one set of industry classification codes, A search module that performs a matching search in the database module based on the aforementioned at least one set of industry classification codes and creates data of the matching search results, An intellectual property information disclosure module that receives the aforementioned data and further statistically analyzes the aforementioned data to create basic intellectual property information, A recommendation module that further receives the aforementioned data, classifies the intellectual property types in the aforementioned data, and creates a set of intellectual property types for which applications are recommended, A login module in which the user operates the electronic device to authenticate the ID. Includes, A trademark risk management system in which the input module receives a brand description re-entered by the user when the user selects a trademark from the intellectual property types for which they recommend filing an application, and the login module completes user authentication; the semantic analysis module receives string information related to the brand description, analyzes and splits it to create and transmit the analysis results of the technical description; the search module performs a matching search against the analysis results in the database module and sends the search results to the recommendation module to create trademark categories for which applications are recommended; and the search module performs a second matching search in the database module based on the recommended application trademark categories and sends the second search results to the search document creation module to create a risk assessment report.

4. The trademark risk management system according to claim 3, further comprising a risk management module that recreates trademark text or figures based on similar trademark text or figures in the risk assessment report and conceptual information from the user's brainstorming received by the input module.

5. A trademark risk management system for receiving clients that accept users by operating electronic devices, The processor of the aforementioned electronic device is connected to a server via a network interface controller and executes application programs for category recommendation and risk management, and at least, A login module in which the user operates the electronic device to authenticate the ID, After the user selects the first target country, an order processing module creates a new case order and periodically updates the information in the case order in the system's temporary memory. An input module that receives text or shapes entered by the user of the instruction manual, converts the text of the instruction manual into a string for marking processing, transmits the string information, and records the input language of the string information in temporary memory. A semantic analysis module that receives string information, analyzes and segments it using a natural language database, creates semantic analysis results, and transmits them. A classification module analyzes the trademark category classification code based on the semantic analysis results, connects to a database module, and determines and creates at least one set of trademark classification codes. A category recommendation module is a calculation model that combines a natural language model, a trademark classification table, and detailed training to analyze string information with ambiguous meanings or inaccurate descriptions into trademark category recommendation information, and ultimately create recommended application trademark categories, by combining it with a semantic analysis module and a classification module. A search document creation module including a text search means that receives the input text, searches for past cases in the database module, and ranks the similarity to create a risk assessment report; a graphic search means that receives the input graphic and searches for past cases in the database module; a conversion means; and a graphic comparison means. A content learning module, which is a large language model, further includes a mode learning means for learning corresponding trademark content from the database module for different trademark categories, A risk management module further comprising text creation means and pattern creation means, which recreates text and / or figures based on learning of the content learning module and past cases matched by the figure search means and text search means in the database module, Includes, The pattern creation means, after receiving a trademark name or a brainstorming concept of a figure from the user via the input module, combines with the semantic analysis module to analyze the brainstorming concept and translate it into a pattern creation language, creates a pattern code corresponding to the brainstorming concept using the pattern creation language, then creates a recreated figure corresponding to the brainstorming concept using the compiler means, and the figure search means, during the recreation of the figure, compares the similarity between the recreated figure and past examples that have been searched in the past so that the similarity between the recreated figure and past examples is less than a predetermined value. The text creation means combines with the semantic analysis module to analyze the brainstorming concept after the user inputs the trademark name and the brainstorming concept of the figure via the input module, and then combines with the content learning module to recreate the text; and simultaneously, the text search means compares the similarity between the recreated text and the past examples that have been searched in the past, such that the similarity between the recreated text and the past examples is within the range of the predetermined value. Trademark risk management system.

6. The trademark risk management system according to claim 5, wherein the figure search means, upon receiving the input figure, first converts the figure into a vector representation using the conversion means, and then performs a matching search in the database module using the figure comparison means to find past cases.

7. The trademark risk management system according to claim 5, further comprising a language determination module that determines whether the input language of the string information is the same as the official language of the first target country.

8. The trademark risk management system according to claim 7, wherein the language determination module determines that the input language of the string information is different from the official language of the first target country, and the translation module translates the string information, and the translation module also translates the language of the trademark design in the final search document into the input language of the string information.

9. The system further includes a case processing module that includes means for extracting user identification information authenticated through login authentication and incorporating it into application data, or for creating application data by having the user directly input application data into fields using an electronic device, receiving the application data via an input module, and applying a format template. The trademark risk management system according to claim 5, wherein the user's case order is completed after the case processing module has created the application form.

10. The trademark risk management system according to claim 5, wherein the conversion means converts the figure from pixels to vectors.

11. The trademark risk management system according to claim 10, wherein the figure comparison means calculates the similarity using edit distance, cosine similarity, or Jaccard similarity during the comparison process, and filters out figures whose similarity exceeds a predetermined value.

12. The trademark risk management system according to claim 5, further comprising a keyword extraction module that extracts keywords from the input content, creates a plurality of keywords, and trains the model using a machine learning algorithm with the keywords.

13. The trademark risk management system according to claim 8, further comprising a multi-language translation module, wherein after a second target country is selected by the user, the language determination module first determines whether the input language of the string information is the same as the official language of the second target country, and if not, translates the trademark name into the official language of the second target country before being formatted by the translation module.

14. (1) The user logs into an electronic device and authenticates their ID and identification information, (2) When the user creates an incident order using an electronic device and selects a first target country, the information in the incident order is periodically updated in temporary memory, (3) The user inputs explanatory text or figures via the input module of the electronic device, performs a marking process to convert the explanatory text into a string, creates string information, and records the input language of the string information in temporary memory. (4) The semantic analysis module performs semantic analysis on the string information, and the classification module performs trademark classification on the string information. Includes, In step 4, further, (411) The semantic analysis module includes the steps of analyzing and splitting string information to create semantic analysis results, (412) The classification module includes the step of classifying string information based on semantic analysis results so as to create at least one set of trademark classification codes, (413) The category recommendation module includes the steps of analyzing string information into trademark category recommendation information and combining it with the semantic analysis module and the classification module to ultimately create the application trademark category to recommend, (421) The graphic search means and the text search means include the steps of receiving text and / or graphics entered by the user, performing a search in a database module to compare and search past examples of trademarks, creating a search file, and further filtering past examples whose similarity is higher than the risk value, In step 421, further, (5) The risk management module includes the step of recreating text and / or figures based on past cases matched by the figure search means and text search means in the database module, based on the learning of the content learning module. Trademark risk management methods.

15. In step (3), further, (31) The language determination module includes the step of determining whether the input language of the string information is the same as the official language of the first target country, (32) If so, the step of performing semantic analysis as is, (33) Otherwise, the process includes the steps of first translating the string information into the official language of the first target country using a translation module, and then performing semantic analysis, The trademark risk management method according to claim 14, wherein in step (413), the created recommended application trademark category is translated again into the input language of the string information.

16. After step (421), further, (422) The shape search means, after receiving an input shape, includes the step of converting the input shape into a vector representation, (423) The figure comparison means then compares figures in the database module, filters out figures whose similarity exceeds a predetermined value, and the filtered figures are considered to be past examples of trademarks. A trademark risk management method according to claim 14, including the following:

17. In step (5), further, (51) The figure creation means, after a user inputs a trademark name or a brainstorming concept of a figure via an input module, analyzes the brainstorming concept and combines it with a semantic analysis module to translate it into a figure creation language, the figure creation language creates a figure code corresponding to the brainstorming concept, the figure code is then compiled by a compiler means to create a recreated figure corresponding to the brainstorming concept, and the figure search means, in the figure recreation process, compares the recreated figure with past examples that have been searched in the past to ensure that the recreated figure and the past examples have a similarity lower than a predetermined value, (52) The text creation means analyzes the concept of brainstorming, then combines it with the semantic analysis module to recreate the text based on the content learning module, and the text creation means further compares the recreated text with past cases that have been searched in the past to ensure that the recreated text and the past cases have a similarity lower than a predetermined value, A trademark risk management method according to claim 14, including the following:

Images