User-customized patent document automatic classification method, device and system based on machine learning

Abstract

The method may include the steps of learning patent documents stored in a patent database through machine learning technology, establishing a first classification benchmark based on the similarity between the patent documents, and generating a basic classification model for classifying the patent documents based on the first classification benchmark; generating a patent pool including multiple patent documents based on a user's search and / or selection input; receiving the user's classification input, and classifying the multiple patent documents included in the patent pool based on the classification input; predicting a second classification benchmark for the user to classify the multiple patent documents based on the similarity between the classified patent documents; and classifying the remaining patent documents in the patent pool that have not yet been classified based on the basic classification model and the second classification benchmark.

Description

Technical Field

[0001] This specification proposes a method for learning a user's patent document classification pattern and constructing a classification model, and automatically classifying patent documents using the constructed classification model, and a system / device therefor. Background Art

[0002] Recently, machine learning, which can be used in various fields, has made great progress. As a field of artificial intelligence, this type of machine learning is divided into supervised learning, unsupervised learning, reinforcement learning, etc., and various algorithms have been developed, such as artificial neural network, boosting, decision tree, support vector machine, random forests, etc.

[0003] Artificial neural networks include input layers, hidden layers, and output layers. They are divided into shallow neural networks and deep neural networks according to the method of stacking hidden layers. In addition, convolutional neural networks ('ConvNet'), recurrent neural networks, deep belief networks, deep Q-networks, etc. have been developed as more improved artificial neural networks in deep neural networks. The machine learning methods of these artificial neural networks (including deep neural networks) are generally referred to as deep learning. In other words, deep learning is to learn data using information input and output layers similar to brain neurons, and refers to machine learning that attempts high-level abstraction through a combination of multiple nonlinear transformation techniques. Summary of the invention

[0004] Technical issues

[0005] As various businesses / projects based on patents have become increasingly active recently, there are many operations to classify patents by technology. Patent offices in various countries classify technology by assigning technology classification codes (IPC, CPC, etc.) to patent documents. However, these technology classification codes are assigned arbitrarily and often do not match the technology disclosed in actual patent documents. Not only is the degree of matching with actual industries and products low, but it is also not actually used for actual classification operations because it does not match the criteria that users want to classify. As a result, users need to personally understand all the patents to be classified and perform classification operations, which requires a lot of time, cost, and labor.

[0006] In addition, there are cases where patent holders (individuals / companies) set the duration of patents by technology based on their own patent strategies. In this case, the patent holder needs to calculate the average duration of each technology of the terminated patent documents by himself, and needs to classify the pending patent documents by technology and apply the previously calculated duration of each technology. This not only consumes a lot of time and cost / manpower, but also makes it very difficult to manage the duration of patents held in large quantities.

[0007] Technical Solution

[0008] According to one embodiment of the present invention, it may include: a step of learning patent documents stored in a patent database through machine learning technology, establishing a first classification benchmark based on the similarity between the patent documents, and generating a basic classification model for classifying the patent documents based on the first classification benchmark; a step of generating a patent pool including multiple patent documents based on a user's search and / or selection input; a step of receiving the user's classification input, and classifying the multiple patent documents included in the patent pool based on the classification input; a step of predicting a second classification benchmark for the user to classify the multiple patent documents based on the similarity between the classified patent documents; and a step of classifying the remaining patent documents in the patent pool that have not yet been classified based on the basic classification model and the second classification benchmark.

[0009] Technical Effects

[0010] According to one embodiment of the present invention, patent documents are automatically classified after learning the classification criteria of the user and predicting the classification criteria desired by the user, thereby having the effect of significantly saving the time / effort / cost / inconvenience consumed by the user in classifying patent documents.

[0011] Furthermore, according to one embodiment of the present invention, the termination patterns of patent rights once held by the user are learned by technology, the durations of various technologies expected by the user are predicted by technology, and the termination dates of currently pending patent rights are predicted, thereby enabling the user to more easily and efficiently manage the durations of currently held patents.

[0012] In addition, there are various effects of the embodiments of the present invention, which will be described below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 An example diagram of a patent document search result screen of a patent search engine illustrating an embodiment of the present invention;

[0014] Figure 2 A schematic diagram of a sliding view screen illustrating an embodiment of the present invention;

[0015] Figure 3 A flowchart illustrating a patent document classification method of a patent document classification system according to an embodiment of the present invention;

[0016] Figure 4 A schematic diagram illustrating a patent document classification method based on user classification input according to an embodiment of the present invention;

[0017] Figure 5 A schematic diagram illustrating a folder setting UI according to an embodiment of the present invention;

[0018] Figure 6 A schematic diagram illustrating a method for representing representative information of patent documents when setting a folder according to an embodiment of the present invention;

[0019] Figure 7 A schematic diagram illustrating a patent document classification method based on user classification input according to an embodiment of the present invention;

[0020] Figure 8 A schematic diagram of a classification history window illustrating an embodiment of the present invention;

[0021] Fig. 9 A schematic diagram illustrating a multi-folder storage method according to an embodiment of the present invention;

[0022] Fig.10 A flowchart illustrating an embodiment of predicting the expiration date of a patent document according to one embodiment of the present invention;

[0023] Fig.11 A schematic diagram of a patent termination recommendation list illustrating an embodiment of the present invention;

[0024] Fig.12 A block diagram of a patent document classification system showing an embodiment of the present invention;

[0025] Fig.13 The following is a block diagram showing a patent document classification system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The technology described below can be subjected to various changes and can have various embodiments, and specific embodiments are illustrated in the figures and described in detail. However, the purpose is not to limit the technology described below to specific implementations, and it should be understood that all changes, equivalents and even substitutes included in the ideas and technical scope of the technology described below are included.

[0027] The terms first, second, A, B, etc. may be used to describe a variety of components, but these components are not limited to the terms, and the purpose of use is only to distinguish one component from other components. For example, the first component may be named as the second component without exceeding the scope of the technology described below, and similarly, the second component may be named as the first component. The term "and / or" includes a combination of multiple related recorded items or any one of multiple related recorded items.

[0028] The singular expressions used in this specification should be understood to include the plural expressions unless otherwise clearly explained in the text. Terms such as "including" are used to indicate the existence of the described features, numbers, steps, actions, constituent elements, parts or their combinations, and do not exclude the existence or additional possibility of one or more other features or numbers, steps, constituent elements, parts or their combinations.

[0029] Before describing the drawings in detail, it is stated that the division of components in this specification is based on the main functions of each component. That is, two or more components to be described below can be combined into one component, or one component can be divided into two or more components according to more detailed functions. Moreover, each component to be described below can further perform part or all of the functions of other components in addition to its own main functions, and part of the main functions of each component can be performed exclusively by other components.

[0030] Furthermore, in the case of an execution method or an action method, the various processes constituting the method may be executed in a different order than the order explicitly described in the text if the specific order is not explicitly described in the text. That is, the various processes may be executed in the same order as the order explicitly described, may be executed substantially simultaneously, or may be executed in the reverse order.

[0031] The present invention is based on a web page / server that provides a patent search engine (or a computer program / application that implements the patent search function through such a web page / server). Therefore, the embodiments described in this specification can be executed by a web server, an application server and / or a client device, and the system included therein can be collectively referred to as a 'patent document classification system'.

[0032] For the sake of convenience, the following description is based on the subject of the implementation of the embodiment being a "patent search engine" or a "patent document classification system", but it can be replaced by a web server, an application server and / or a client device for description. In addition, for the sake of convenience, the following description is based on a patent search engine provided through a web page, but it is not limited to this and can also be applied to a patent search engine provided through an application, which is obvious.

[0033] First, after introducing the basic functions and graphical user interface (GUI) of the patent search engine that is the basis of the present invention, the automatic classification method of patent documents is described in detail.

[0034] Figure 1 The following is a schematic diagram of a patent document search result screen of a patent search engine, illustrating an embodiment of the present invention.

[0035] The patent search engine of the present invention can provide various information about patents retrieved according to the user's search input on one screen.

[0036] For example, Figure 1 As shown, the patent search engine is a patent document search result, and can provide the number of patent documents retrieved by country / status 130 and a list of all or by country retrieved patent documents 140. In particular, when providing the patent document list 140, the patent search engine can also provide bibliographic information about each patent document (country of application, application / grant number, type, application date, current status (application / publication / grant, etc.), invention name, etc.).

[0037] Furthermore, the patent search engine may provide a running key label 120 for the user to perform various additional operations on the retrieved patent documents. Figure 1 As shown, the patent search engine may provide a slide view 110, an operation key tab 120 including operation keys for work storage, image engine, download, and list output operations.

[0038] Among them, the sliding view 110 provides a function of compiling a catalog by briefly composing information representing each patent document (e.g., abstract drawing, abstract information, independent claims, etc.) into one screen so that the user can more easily understand the retrieved patent documents. Through this function, the user can more easily understand the patent documents. Figure 2 This is explained in more detail.

[0039] Work storage is used to provide a function of storing patent documents in a specific folder, which can be used for the patent document classification task proposed in this specification. Image engine is used to provide a function of collecting patent documents selected by the user and generating charts or calculating statistics according to the user's desired benchmark. Download is used to provide a function of downloading patent documents, and list output is used to provide a function of outputting a list of retrieved patent documents, abstract information, etc.

[0040] Furthermore, the patent search engine can provide execution keys for instructing various additional operations using patent documents, without being limited to the above-described embodiment.

[0041] Figure 2 A schematic diagram of a sliding view screen illustrating an embodiment of the present invention.

[0042] See also Figure 2 The sliding view screen can be roughly composed of a highlight window 210, a project window 220, an abstract illustration window 230, a patent document representative content window 250, and a patent list window 240.

[0043] The highlight window 210 provides a function of highlighting a specific keyword / search term input by the user in each patent document. The item window 220 is an item window that can manage (generate and edit, etc.) various upper / lower folders generated by the user. The user can classify the patent documents by storing them in the desired folders, and manage the patent documents by folders.

[0044] The abstract drawing window 230 is used to display the abstract drawing of the target patent document, and the patent document representative content window 250 performs the function of briefly providing representative information required to understand the target patent document. The user can easily understand the characteristics and technical content of the target patent document through the abstract drawing window 230 and the patent document representative content window 250.

[0045] The patent list window 240 is a window for providing a list of patent documents selected by the user to be viewed through the sliding view. Through this patent list window 230, the user can understand which patent documents are contained in the current pool, how many are contained, and can also know what patent document the current user is viewing.

[0046] The above is a brief description of the patent search engine that is the basis of the present invention. The embodiments of the present invention described below can be executed / provided based on such a patent search engine, and in particular, the patent classification system and method can be provided as a function in the sliding view screen. However, it is obvious that the present invention is not limited to the above-mentioned patent search engine, and can also be applied to various patent search engines / web pages in the same / similar manner.

[0047] Figure 3The flowchart is a method for classifying patent documents in a patent document classification system according to an embodiment of the present invention.

[0048] See also Figure 3 First, the patent document classification system can learn part or all of the patent documents stored in the patent database through machine learning technology. In addition, the patent document classification system can establish a first classification benchmark based on the similarity between the patent documents learned based on the machine learning technology, and generate a basic classification model for classifying the patent documents according to the established first classification benchmark (S301).

[0049] The patent document classification system can use a variety of machine learning techniques to learn patent documents. As an embodiment of the machine learning technology, recurrent neural network (RNNs) technology, deep neural network (DNNs) and / or convolutional neural network (CNNs) can be used. RNNs (Recurrent Neural Networks) technology is an artificial neural network that recognizes patterns from data with a certain arrangement (sequence or time series data) form, such as words, genetic genes, handwriting, voice signals, sensor-sensed data, text, stock prices, etc. In RNNs technology, the current input data and the past input data are considered at the same time, and the arrangement information is stored in the hidden layer of RNNs. The information stored in the hidden layer can be used again when the input of the information to be stored is received after a certain period of time. DNNs represents a learning method with more than two such hidden layers. In DNNs, the computer can derive the best dividing line by repeatedly executing the process of autonomously making a classification table and distorting the space and distinguishing the data. CNNs technology is an algorithm that applies this DNNs, which will be described in more detail in step S306 later.

[0050] Before learning patent documents, the patent document classification system can use a translation engine to translate the patent documents into a common language (e.g., Korean or English), so that the patent documents can be learned in a unified language. However, the patent document classification system can also use a translation engine to translate into up to two languages ​​and learn patent documents.

[0051] When judging the similarity between patent documents, the patent document classification system can make judgments based on the patent documents of the same family of each patent document, patent classification codes (for example, IPC, CPC, etc.), existing patent documents, patent documents cited during the examination stage, patent documents cited during the examination of other patent documents, keywords (extracted by text mining) and / or text data, etc. At this time, algorithms such as Word Emdedding and Word2vec can be used to vectorize the text data.

[0052] In order to calculate the similarity between patent documents, the patent document classification system may use cosine coefficient, Jaccard coefficient, Dice coefficient, Euclidean distance, vector inner product, etc. According to the embodiment, there may be a situation where a specific similarity measurement method needs to be used according to the classification (or clustering) technology. For example, the Ward technique of measuring the distance between documents using the Euclidean distance and the group average technique of calculating the similarity by the inner product are equivalent to the above-mentioned technology. When different similarity measurement methods are used, the patent documents (or components) belonging to each classification (or cluster) may be different.

[0053] Afterwards, the patent document classification system may generate a patent pool (Pool) including a plurality of patent documents according to the user's search and / or selection input (S302). The plurality of patent documents included in the patent pool may become the classification target documents of the user as described below.

[0054] Afterwards, the patent document classification system may receive the user's classification input (S303), and classify the multiple patent documents included in the patent pool according to the classification input (S304). The patent document classification system may generate at least one folder according to the user's folder generation input, and store the patent documents of the patent pool in each folder (or move them to the folder) according to the user's classification input for classification. The classification input is various user inputs for the user to classify the patent documents, and there may be multiple embodiments. As one embodiment, the user may simply perform the classification operation by performing an action / touch input of dragging and dropping the patent document to the folder side. As another embodiment, the classification input may be equivalent to the user input of the immediate storage key provided by the patent document classification system. Among them, the immediate storage key may be equivalent to the UI provided by the patent document classification system that enables the user to immediately store the patent document in the folder required by the user for classification. The immediate storage key may be provided for each patent document. As another embodiment, the classification input is an input for classifying the target patent document in a toggling manner, for example, it may be equivalent to the user input of O or X. The user may select / input O or X for the target patent document to classify into two categories (for example, O or X). The category items (O or X) are set directly by the user or provided as preset items by the web page / application server. Figure 4 and Figure 7 An embodiment of categorical input is described in more detail.

[0055] Afterwards, the patent document classification system can predict the second classification criterion for the user to classify the patent documents based on the similarity between the classified patent documents (S305). The prediction method of the second classification criterion based on similarity can be performed in the same / similar manner as the method of establishing the first classification criterion based on similarity.

[0056] Afterwards, the patent document classification system may classify the remaining patent documents in the patent pool that have not yet been classified according to the basic classification model and the second classification benchmark (S306).

[0057] More specifically, as an embodiment, the patent document classification system may combine the predicted second classification benchmark with the basic classification model to construct a user-customized classification model. The user-customized classification model can be constructed based on transfer learning. Transfer learning can mainly use a pre-learned model (pre-trained model), which refers to a model that has been learned with large-sized data for the purpose of solving similar problems. In particular, the pre-learned model used for transfer learning in this specification has a convolutional neural network (CNNs) structure. This CNN is composed of the following two parts.

[0058] -Convolutional base: The part where convolutional layers and pooling layers are stacked into multiple layers. The goal of convolutional base is to effectively extract features from data.

[0059] -Classifier: It is mainly composed of fully connected layers. A fully connected layer is a layer in which all neurons in all layers are fully connected to the output nodes of the previous layer. The ultimate goal of the classifier is to learn the extracted features well to classify the data into corresponding categories (image classification).

[0060] The patent document classification system can use these two parts to build a user-customized classification model. In particular, a strategy of fixing the Convolutional base and learning a new classifier can be used to build a user-customized classification model. This strategy is effective when the computing power is insufficient or the data set is too small, and / or the problem I want to solve is very similar to the data set that has been learned by the pre-learning model.

[0061] When it is applied to the present embodiment, the patent document classification system can fix the basic classification model as the convolutional base and then learn the second classification benchmark as the classifier to construct a user-customized classification model. The so-called fixed basic classification model can be interpreted as fixing the features / benchmarks for pre-classifying the patent documents of the patent pool into the number of folders generated by the user according to the first classification benchmark of the basic classification model. Therefore, the pre-classification features / benchmarks based on the first classification benchmark are fixed, and the classification features / benchmarks based on the second classification benchmark are newly learned as the classifier to construct a user-customized classification model.

[0062] As another embodiment, the patent document classification system may not learn the patent documents classified by the user as in the embodiment, but only use the basic model to classify the patent documents. In this case, the patent document classification system can calculate the technical distance between the patent documents classified by the user according to the basic classification model, and predict the second classification benchmark according to the calculated technical distance (that is, no learning is required). The patent document classification system can classify the remaining patent documents according to the second classification benchmark predicted based on the basic classification model as above. For example, when the user classifies some patent documents in the patent pool into four categories, the patent document classification system can extract the features of the patent groups belonging to the four categories according to the basic classification model, and can automatically classify the remaining patent documents according to the features of each patent group in the basic classification model extracted in this way. According to this embodiment, the second classification benchmark can be finally interpreted as a form of grouping the classification of the first classification benchmark according to the user classification features to reduce the types / numbers of all classifications. For example, assuming that there are 1000 technical classifications that can be classified according to the first classification benchmark, the second classification benchmark can be established to cluster / group the classifications with high similarity according to the user classification mode / benchmark, so as to divide the 1000 technical classifications into 50 technical classifications.

[0063] Furthermore, the patent document classification system may output an input key (e.g., a classification start key) for instructing automatic classification of the remaining patent documents. When receiving the user input to the input key, the patent document classification system may classify the remaining patent documents using the user-customized classification model. More specifically, when receiving the user input to the input key, the patent document classification system may classify the remaining patent documents using the user-customized classification model and automatically store them in folders that match the classification characteristics, thereby classifying the remaining patent documents.

[0064] That is, this flowchart can be applied as the following embodiments.

[0065] The user can search for 1,000 patents and set them as a patent pool to generate four folders. The patent document classification system can use the basic classification model to pre-classify the 1,000 patents into four folders. The user can classify and store the first 100 patents out of the 1,000 patents into four folders, and the patent document classification system can predict the user's classification benchmark / features from such user classification information. The patent document classification system can generate a model that adds the user's classification benchmark / features to the pre-classification benchmark / features (i.e., the first classification benchmark), thereby classifying the remaining 900 patents into four folders for storage.

[0066] In addition, although not shown in this flowchart, the patent document classification system can calculate the accuracy of automatic classification (first accuracy) in real time and output an indicator representing the first accuracy in real time.

[0067] As an embodiment, the first accuracy can be calculated based on the number of patent documents included in the patent pool, the number of categories set / specified by the user, and / or the number of patent documents that have been classified by the user to date. More specifically, in some cases, the patent document classification system will learn the basic classification model and pre-calculate / derive the second accuracy based on the total number of classification object patent documents (first factor), the number of categories (second factor), and / or the number of patent documents included in each category (third factor) as data. For example, after learning the basic classification model, the patent document classification system can pre-calculate / derive the statistical data that the classification accuracy is 80% when the number of classification object patent documents is 1,000, the number of categories is three, and the number of patent documents included in each category is 100. In this case, the patent document classification system can import the user's classification pattern / benchmark into this statistical data as a factor to derive the first accuracy. That is, the first accuracy may be equivalent to the second accuracy pre-calculated when the number of multiple patent documents included in the patent pool by the user, the number of classifications based on the second classification benchmark, and / or the number of patent documents included in each classification are applied to the statistical data in sequence as the first to third factors. Therefore, if the user includes 1,000 patent documents in the patent pool, and 100 patent documents are classified in three folders respectively, according to the above example, the derived user's classification accuracy (i.e., the first accuracy) is 80%. Furthermore, the patent document classification system can further consider the probability of the closeness of the distance between the patent documents included in each classification by the user (indirectly verifying whether the user has reasonably placed them according to the classification) to derive / display the first accuracy.

[0068] When based on the above, the first to third factors used to derive the maximum accuracy can be fixed within the statistical data derived based on the basic classification model. In this case, when the patent document classification system determines that the derived first accuracy is lower than the maximum accuracy, a prompt window for guiding the user to further classify can be output. For example, when the patent documents in a patent pool including 1,000 classification object patent documents are classified into three categories, the number of patent documents that should be classified at least in each folder in order to derive the maximum accuracy of 80% may have been set to 100. Therefore, in order to enable the user to make each folder include at least 100 patent documents in order to achieve the maximum accuracy of 80%, the patent document classification system can output a prompt window for guiding this. For example, the patent document classification system can output a prompt window such as 'Since the current classification accuracy is 60%, please classify more patents.' or 'The classification accuracy of the XX folder is low, so please classify more patents into the XX folder.'

[0069] As another example, the first accuracy may be equivalent to the degree of matching between the result classified according to the first classification reference and the result classified according to the second classification reference.

[0070] The patent document classification system can calculate the accuracy of automatic classification by folder and / or patent pool, and can output indicators representing the accuracy by folder and / or patent pool.

[0071] Figure 4 The diagram is a schematic diagram illustrating a patent document classification method based on user classification input according to an embodiment of the present invention.

[0072] As described above, the user can simply perform the classification work by dragging and dropping the patent documents to the folder to be classified.

[0073] According to the composition of the sliding view screen, for example, Figure 4 As shown, the patent document classification system can output an icon 450 that can drag and drop the patent document into a folder in the patent document representative content window. The folder 420 generated by the user for classification can be provided in the project window 410 of the sliding view, and the user can drag and drop 430 the icon 450 to the side of the desired folder 410 to perform classification and storage in the folder 410.

[0074] In addition, the patent document classification system may provide a setting UI 440 so that the user can directly set the level and / or technical theme / keyword for each folder 410, which will be described later. Figure 5 and Figure 6 This is explained.

[0075] Figure 5 FIG. 1 is a diagram illustrating a folder setting UI according to an embodiment of the present invention. Figure 6 2 is a schematic diagram illustrating a method for displaying representative information of patent documents when setting a folder according to an embodiment of the present invention.

[0076] The patent document classification system may provide a setting UI 520 for the user to directly input additional information by folder. Figure 5 As shown, the patent document classification system can output a setting UI 520 that can set the level and / or keyword / theme of each folder 510. After selecting the folder 510 to be set, the user can directly set the importance level and / or keyword / theme of the selected folder 510 to more easily manage patent documents by folder 510.

[0077] Regarding patent documents belonging to / stored in a folder in which additional information is input / set using the setting UI 520, the set / input additional information may be displayed and provided to the user as representative information.

[0078] For example, Figure 5 and Figure 6 As shown in FIG. 6 , when the level of folder B is set to A and the keyword / subject is set to #camera, the level of all patent documents that can be stored in folder B is set to A and the keyword / subject is set to #camera as representative information 610. Therefore, the user does not need to set the level / keyword / subject, etc. for each patent document, but can use the folder setting UI 520 to set the level / keyword / subject for each folder, so the user can manage patent documents more easily by folder.

[0079] Figure 7 The diagram is a schematic diagram illustrating a patent document classification method based on user classification input according to an embodiment of the present invention.

[0080] As described above, the user can perform a user input on the immediate storage key 710 output together with the patent document to simply perform the classification work.

[0081] According to the sliding view screen composition, for example, Figure 7 As shown, the patent document classification system may output an immediate storage key 710 in the patent document representative content window that can execute the function of immediately storing the patent document in a folder. The user can touch / select (720) the immediate storage key to immediately store the patent document in a folder. If there are multiple folders that can be stored, the patent document classification system may provide a selection UI (730) that can select a folder that can be stored when receiving user input on the immediate storage key. The user can select a folder in which the target patent document is to be stored among the folders displayed by the input selection UI (730), thereby immediately storing the target patent document in the folder.

[0082] Figure 8 A schematic diagram of a classification history window according to an embodiment of the present invention.

[0083] See also Figure 8 The patent document classification system can record the user's recent classification history and provide the user with a window (ie, classification history window) 810 that displays the record information.

[0084] In the classification history window 810, the patent documents recently classified by the user can be cataloged in the order of classification, and simple classification information about each patent document (for example, the application / grant number of the classified patent document, the folder in which it is stored, etc.) can also be provided. In addition, the classification history window 810 can provide a 'cancel operation' icon 820 for each classified patent document. The 'cancel operation' icon 820 is equivalent to an icon for canceling the classification operation of the target patent document. Therefore, when receiving a user input to the cancel operation icon 820, the patent document classification system can cancel the classification operation of the target patent document for which the cancel operation input is received, and the target patent document will return to the state before being stored in / belonging to the folder (i.e., before classification).

[0085] Fig. 9 The diagram is a schematic diagram illustrating a multi-folder storage method according to an embodiment of the present invention.

[0086] The patent document classification system supports multi-folder storage, and as a result, one patent document can be stored in multiple folders at the same time. For example, the target patent document can be stored in folders A and B at the same time.

[0087] The patent document classification system may provide an icon 910 for displaying information of a folder storing / classifying the target patent document. Fig. 9 As shown, when the target patent document is stored / classified in folders A and B, the patent document classification system can provide a folder icon 910 indicating the number of folders 2, and when the user touches / selects the folder icon 910, list information (folders A and B) 920 of the folders storing the current target patent document can be provided.

[0088] As the patent document classification system supports multi-folder storage, each patent document can be assigned a level and / or keyword / topic that is at most equivalent to the number of all folders currently generated (because the level and / or keyword / topic can be set by folder).

[0089] Although not shown in this figure, the patent document classification system that has completed automatic classification can provide a reclassification UI for each folder so that the user can directly confirm and modify the classification content.

[0090] The reclassification UI may provide a reclassification object display key for the user to reclassify the patent documents classified into each folder, a reclassification key for instructing the reclassification object patent documents to be reclassified, and the like. Therefore, the user may use the 'reclassification object display key' to display the object patent documents that are not suitable for classification into the folder, and may use the 'reclassification key' to delete the patent documents displayed by the reclassification object display key from the folder or move them to other folders. The reclassification object may be displayed as a variety of embodiments, for example, it may be displayed using a label or color provided by the web page / application server (or directly set by the user). In this case, the user may select and input (for example, touch / click input) the label or color provided by the web page / application server with each patent document as the object to specify the patent document to be reclassified.

[0091] The patent document classification system can also digitize the content reclassified by the user as above, and learn the reclassified data based on the basic classification model to evolve autonomously.

[0092] The above introduces a method for automatically classifying patent documents by building a user-customized classification model based on machine learning (or deep learning). The above method of combining user classification patterns / habits as features / benchmarks with existing models to build new models can be applied to a variety of user-customized patent management embodiments. For example, the method can also be used to predict / recommend the expiration date of patent rights, the number of claims to be maintained for each patent, the possibility of rights transfer (licensing), the patent maintenance method for each inventor, and the patent maintenance method based on patent evaluation.

[0093] For example, when the method of the present invention is applied to predicting the expiration date of patent rights, the patent document classification system can classify the patents held by the user by technology (in this case, the basic classification model can be used), and can learn the patent expiration pattern according to the classified technical field and generate a user-customized model. Through this, the patent document classification system can automatically recommend the predicted expiration date of the patent by technical field in the user's desired period according to the user's existing patent management (termination, etc.) pattern / features.

[0094] For example, when the Korea Life Science Research Institute has a characteristic of maintaining stem cell-related patents for more than 15 years without terminating them, and compound or manufacturing process patents have a characteristic / pattern of terminating within five years, the patent document classification system can grasp this termination characteristic / pattern and recommend that the patents in the field of stem cell technology currently held by the Korea Life Science Research Institute be maintained for more than 15 years, and guide / recommend that the compound and manufacturing process patents be terminated within 5 years.

[0095] See later Fig.10 This embodiment is described in more detail.

[0096] Fig.10 The flowchart is a flowchart illustrating an embodiment of predicting the expiration date of a patent document according to one embodiment of the present invention.

[0097] This embodiment is written in order to facilitate the description. Figure 3 The embodiment is executed after the embodiment of, but is not limited to, it can also be used as an independent Figure 3 It is obvious that the embodiments of the present invention are carried out by the embodiments of the present invention.

[0098] First, the patent document classification system can calculate the first average period from the application date or the authorization date to the expiration date of the patent documents classified by the basic classification model according to the first classification standard for the patent documents included in the patent database (Data base; DB) (S1001).

[0099] Afterwards, the patent document classification system may classify the multiple documents included in the patent pool according to the first or second classification criteria (S1002), and calculate the second average period from the application date or authorization date to the expiration date of the multiple patent documents according to the classification (S1003). Among them, the multiple documents included in the patent pool may be equivalent to (including) patent documents that the user currently holds or has held but has currently terminated. Through this step, the patent document classification system can predict the average patent expiration period / mode / characteristics of the user for various technologies.

[0100] Finally, the patent document classification system may predict the expiration date of a specific patent document by considering the first and second average terms (S1004), wherein the specific patent document may correspond to (include) a patent currently held by the user but not expired and / or a patent document specified according to the user's input.

[0101] In order to predict the expiration date, the patent document classification system may first classify the specific patent document according to the first and / or second classification criteria. After that, the patent document classification system may combine the first and / or second average terms according to the classification of the specific patent document based on the first and / or second classification criteria in a preset ratio to calculate the expiration date of the specific patent document. For example, it may be assumed that the first specific patent document belongs to the 'biological' classification according to the first classification criteria, and the average duration of the 'biological' classification is five years in the entire patent database. Also, it may be assumed that the first specific patent document belongs to the 'biological rhythm' classification according to the second classification criteria, and the average duration of the 'biological rhythm' classification is 10 years within the user's terminated patent. When the preset ratio between the first and second classification criteria is set to 3:7, the term of 5*0.3+10*0.7=8.5 may be derived as an evaluation index. That is, it may be predicted that the expected duration of the first specific patent document is 8 years and 6 months.

[0102] The preset ratio can be freely set / changed according to various embodiments. From the perspective of a user-customized model, the ratio of the second classification standard can be set to be higher than the first classification standard.

[0103] Fig.11 A schematic diagram of a patent termination recommendation list illustrating an embodiment of the present invention.

[0104] See also Fig.11 The patent document classification system may catalog the predicted expiration dates of specific patent documents into an evaluation index 1120 for inclusion in the patent expiration recommendation list 1110 for output.

[0105] The time point for outputting the patent termination recommendation list 1110 can be set from the termination prediction time point to before a preset time point (for example, 6 months or 1 year ago), which can be freely set by the user.

[0106] The user can use this patent expiration recommendation list 1110 to predict the expiration date of the currently held patent rights, and can determine in advance whether to maintain or terminate the rights to effectively manage the duration of each patent right.

[0107] Fig.12 The following is a block diagram showing a patent document classification system according to an embodiment of the present invention.

[0108] See also Fig.12 ,The patent document classification system can be roughly composed of a web server (or application server) and a client device.

[0109] The web server (or application server) 1200 may include a processor 1210, a memory unit 1220 and / or a communication unit 1230, and the client device 1300 may include a processor 1240, a sensor unit 1250, a communication unit 1260 and / or a display unit 1270. At least one of the components may be excluded or a new component may be included according to the embodiment. In addition, each unit may have a hardware circuit / device, etc. in order to perform the functions described below, and may be implemented as one chip or multiple chips.

[0110] The memory unit 1220 can store various digital data such as video, audio, photo, video, computer program, application, etc. The memory unit 1220 can be equivalent to various digital data storage spaces such as flash memory, hard disk drive (HDD), solid state drive (SSD), etc. In particular, the memory unit 1220 of the web server (or application server) 1200 can be equivalent to a patent database, which can store various patent documents from all over the world.

[0111] The communication units 1230 and 1260 can perform communication and send and receive various data using at least one communication protocol. In particular, user input to the client device 1300 can be transmitted to the web server (or application server) 1200 through the communication units 1230 and 1260, and various output / feedback data / information of the web server (or application server) 1200 can also be transmitted to the client device 1300 through the communication unit 1260.

[0112] The sensor unit 1250 is a general term for the above-mentioned various sensing mechanisms, which can sense various inputs of the user and / or the environment of the client device 1300. In particular, the sensor unit 1250 can sense various inputs of the user (for example, touch / action / selection / retrieval input, etc.) and send the sensing results to the processor. As an embodiment, the sensor unit 1250 may include at least any one of various sensing mechanisms such as a gravity sensor, a geomagnetic sensor, a motion sensor, a gyroscope sensor, an acceleration sensor, an infrared sensor, an inclination sensor, a brightness sensor, an altitude sensor, an olfactory sensor, a temperature sensor, a depth sensor, a pressure sensor, a bending sensor, an audio sensor, a video sensor, a GPS (Global Positioning System) sensor, a touch sensor, and a grip sensor. The above-mentioned sensors can be included in the device as another element, or integrated into at least one element included in the device.

[0113] The display unit 1270 can display various images / videos. In particular, the display unit 1270 of the client device 1300 can display various web pages / UIs provided by the web server, various application screens / UIs provided by the application server, and the like.

[0114] The processors 1210 and 1240 can execute the various embodiments proposed in this specification by controlling at least one other constituent unit. Therefore, the execution subject of the embodiments proposed in this specification can be replaced by a processor for description. The processors 1210 and 1240 can run various programs / applications stored in the memory unit 1220 and process the data / information inside the server / device. The processors 1210 and 1240 can be configured to include at least one of a central processing unit (CPU), a microprocessor unit (MPU), a microcontroller unit (MCU), an application processor (AP), an AP (Application Processor), or a processor of a known form in the technical field of the present invention. The processor can execute the operation of at least one application or program for executing the method of the embodiment of the present invention.

[0115] Fig.13 The following is a block diagram showing a patent document classification system according to an embodiment of the present invention.

[0116] The components in this figure can be used Fig.12 For example, the loading unit can be implemented by a processor, a memory unit, and a folder generating unit, the patent document display unit and the folder display unit can be implemented by a processor and a display unit, and the storage unit can be implemented by a memory unit.

[0117] The loading unit 1310 can perform the function of loading at least one patent document requested by the user. The loading unit 1310 is connected to a storage space storing various data, commands and / or information, and can store one or more application programs for obtaining the patent classification service providing method of the embodiment of the present invention. The loading unit 1310 can also store various information received through the communication unit or input by the user.

[0118] The storage unit 1350 can temporarily or non-temporarily store data transmitted from an external device, etc. The storage unit 1350 can be configured to include a non-volatile memory such as a read-only memory (ROM), an erasable programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a flash memory, a hard disk, a removable disk, or any form of computer-readable storage medium known in the technical field to which the present invention belongs.

[0119] The folder generating unit 1320 may generate folders for storing or classifying patent documents.

[0120] The patent document display unit 1330 plays a role of displaying at least one of the distinguishing items in the patent document in the information area, and the folder display unit 1340 plays a role of displaying the folder in the folder area. It is preferable that the information area and the folder area are separate areas.

[0121] Furthermore, the storage unit 1350 sets at least a portion of the information area as a draggable area and at least a portion of the folder area as a draggable area, and stores the patent document in the folder according to the user's drag and drop input.

[0122] Embodiments of the present invention may be implemented by various mechanisms, such as hardware, firmware, software, or a combination thereof. In the case of hardware implementation, an embodiment of the present invention may be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, etc.

[0123] Furthermore, in the case of implementation by firmware or software, an embodiment of the present invention can be implemented as a module, step, function, etc. that can execute the functions or actions described above, and stored in a storage medium that can be read by various computer mechanisms. Among them, the storage medium may include one or a combination of program commands, data files, data structures, etc. The program commands stored in the storage medium may be specially designed and constructed for the present invention, or they are well known to computer software technicians. For example, storage media examples include magnetic media (Magnetic Media) such as disks, floppy disks and tapes, optical recording media (optical media) such as CD-ROM (Compact Disk Read Only Memory), DVD (Digital Video Disk), magneto-optical media (Magneto-Optical Media) such as floppy disks (Floptical Disk), and hardware devices for storing and executing program commands of special components such as ROM, RAM, and flash memory. Examples of program commands include not only mechanical codes obtained by compilers, but also high-level language codes executed by computers using interpreters, etc. Such hardware devices can be configured to work with more than one software module in order to perform the work of the present invention. The opposite is true.

[0124] Furthermore, the device or interrupt of the present invention may be driven by commands that cause one or more processors to execute the functions and processes described above. For example, such commands may include interpreted commands such as script commands such as JavaScript or ECMAScript commands, or other commands stored in executable code or computer-readable media. Furthermore, the device of the present invention may be implemented as a distributed type such as a server farm distributed over a network, or may also be implemented in a computer device.

[0125] Furthermore, the computer program (also known as a program, software, software application, script or code) carried on the apparatus of the present invention and running the method of the present invention can also be written in any form of a programming language including a compiled or interpreted language or a priori or procedural language, and can include independent programs or modules, components, subroutines or other units suitable for use in a computer environment to be deployed in any form. A computer program file does not necessarily correspond to a file in a file system. A program may be stored in a file provided to a requested program, or in multiple interactive files (e.g., one or more modules, files storing a portion of a lower program or code), or in a portion of another program or data file (e.g., one or more scripts stored in a markup language document). A computer program may be deployed to be located on a web page or distributed across multiple web pages and run on multiple computers or one computer interconnected by a communication network.

[0126] Although the figures are described separately for the sake of convenience, it is also possible to design a new embodiment by combining the embodiments described in the figures. Furthermore, the configuration and method of the embodiments of the present invention described above are not applicable in a limiting manner. In order to be able to make various modifications to the above embodiments, all or part of the configuration of each embodiment can be selectively combined.

[0127] Furthermore, although preferred embodiments have been shown and described above, this specification is not limited to the above-mentioned specific embodiments, and a person skilled in the art may perform various modified implementations without exceeding the scope of the subject matter described in the claims, and these modified embodiments should not be understood as departing from the technical ideas or prospects of this specification.

[0128] Industrial Applicability

[0129] The present invention can be applied to various patent search engine technology fields and / or machine learning technology fields.

Claims

1. A patent document classification method is a patent document classification method based on classification input by a user, comprising: The steps of learning patent documents stored in a patent database by machine learning technology, establishing a first classification benchmark based on the similarity between the patent documents, and generating a basic classification model for classifying the patent documents based on the first classification benchmark; A step of generating a patent pool including a plurality of patent documents according to a user's search and / or selection input; A step of receiving a classification input from the user, and classifying a plurality of patent documents included in the patent pool according to the classification input from the user; A step of predicting a second classification criterion for classifying the plurality of patent documents by the user based on the similarity between the classified patent documents; as well as The step of classifying the remaining unclassified patent documents in the patent pool according to the basic classification model and the second classification benchmark, The step of classifying the remaining unclassified patent documents in the patent pool according to the basic classification model and the second classification benchmark includes: The step of combining the second classification benchmark with the basic classification model to construct a user-customized classification model; and The step of using the user-customized classification model to classify the remaining patent documents in the patent pool that have not yet been classified; The patent document classification method further includes: The step of deriving a first accuracy of automatic classification in real time and outputting an indicator representing the first accuracy in real time, The first accuracy is calculated based on the number of patent documents classified according to the second classification standard and / or the number of patent documents that the user has completed classification so far.

2. According to the patent document classification method described in claim 1, the similarity between the patent documents is determined based on the same family patent documents, patent classification codes, existing patent documents, cited patent documents, cited patent documents, keywords and / or text data of each patent document.

3. According to the patent document classification method according to claim 1, the machine learning technology includes recurrent neural network, deep neural network and / or convolutional neural network technology.

4. According to the patent document classification method of claim 1, the step of classifying the plurality of patent documents contained in the patent pool comprises: A step of generating at least one folder; as well as storing at least one patent document from among the plurality of patent documents included in the patent pool into the at least one folder for classification according to the classification input of the user, The user's classification input is an input of dragging and dropping the at least one patent document into the folder.

5. The patent document classification method according to claim 4, further comprising: The step of outputting an input key for indicating automatic classification of the remaining patent documents in the patent pool that have not yet been classified, When the user's classification input to the input key is received, a step of classifying the remaining patent documents using the user-customized classification model is performed.

6. According to the patent document classification method of claim 5, the step of classifying the remaining patent documents comprises: The step of classifying the remaining patent documents using the user-customized classification model to automatically store them in the at least one folder.

7. The patent document classification method according to claim 1, further comprising: A step of calculating, by classification, a first average period from the application date or the grant date to the expiration date of patent documents classified according to the first classification standard; The step of classifying the plurality of documents included in the patent pool according to the first classification standard or the second classification standard, and calculating the second average period from the application date or the authorization date to the expiration date of the plurality of patent documents according to the classification; and The step of predicting the expiration date of a specific patent document taking into account the first average date and the second average date.

8. According to the patent document classification method according to claim 7, the multiple documents contained in the patent pool include patents that the user currently holds or has held but is currently terminated.

9. The patent document classification method according to claim 8, wherein the step of predicting the expiration date of the specific patent document comprises: A step of classifying the specific patent document according to the first classification standard and / or the second classification standard; as well as The step of combining the first average term and / or the second average term at a preset ratio to calculate the expiration date of the specific patent document based on the classification of the specific patent document according to the first classification standard and / or the second classification standard.

10. The patent document classification method according to claim 9, wherein the specific patent document corresponds to a patent currently held by the user and not terminated and / or a patent document specified based on a classification input by the user.

11. The patent document classification method according to claim 10, further comprising: The step of cataloguing and outputting the predicted expiration dates of the specific patent documents as evaluation indices.

12. The patent document classification method according to claim 1, wherein the step of learning the patent document by machine learning technology comprises: The step of translating the patent document into a language using a translation engine and studying it in a unified language.

13. The patent document classification method according to claim 4, wherein when the at least one folder is generated, the method further comprises: a step of outputting an immediate storage key, wherein the immediate storage key is used to enable the user to immediately store each patent document into the at least one folder for classification, The user's classification input is an input to the immediate storage key.

14. The patent document classification method according to claim 1, wherein In the case where the second accuracy based on the total number of patent documents to be classified, i.e. the first factor, the number of categories, i.e. the second factor, and / or the number of patent documents included in each category, i.e. the third factor, has been calculated in advance based on the learning of the basic classification model, The first accuracy is equivalent to the second accuracy calculated when the number of the multiple patent documents included in the patent pool by the user, the number classified according to the second classification standard and / or the number of patent documents included in each of the classifications are applied in sequence as the first factor, the second factor and / or the third factor.

15. The patent document classification method according to claim 14, wherein the step of outputting an indicator representing the first accuracy in real time comprises: The step of calculating the first accuracy according to the folder and / or the patent pool, and outputting an indicator representing the first accuracy in real time according to the folder.

16. The patent document classification method according to claim 15, wherein when the first to third factors for deriving the maximum accuracy in the basic classification model have been specified, and when the first accuracy is lower than the maximum accuracy, a prompt window is output to guide the user to further classify.

17. According to the patent document classification method according to claim 1, the user-customized classification model is constructed based on transfer learning technology.

18. The patent document classification method according to claim 17, wherein: Regarding the user-customized classification model, after the basic classification model is fixed as the convolution basis according to the transfer learning technology, the second classification benchmark is newly learned as a classifier to construct the user-customized classification model.

19. The patent document classification method according to claim 4, further comprising: A step of outputting a setting UI for setting a user level and / or setting a keyword for each of the folders.

20. The patent document classification method according to claim 1, wherein: The step of predicting a second classification criterion for classifying the plurality of patent documents by the user according to the similarity between the classified patent documents comprises: The step of calculating the technical distance between the classified patent documents according to the basic classification model and predicting the second classification criterion according to the technical distance, The steps of classifying the remaining patent documents in the patent pool that have not yet been classified include: The step of classifying the remaining patent documents in the patent pool that have not yet been classified according to the second classification benchmark predicted by the basic classification model.

21. A web page server for classifying patent documents according to classification input by a user, comprising: A memory unit for storing data; a communication unit for performing communication using at least one communication protocol; as well as a processor, configured to control the memory unit and the communication unit, Wherein, the processor is used for: By learning patent documents stored in a patent database through machine learning technology, a first classification benchmark is established based on the similarity between the patent documents, and a basic classification model for classifying the patent documents is generated based on the first classification benchmark. Generate a patent pool including multiple patent documents based on the user's search and / or selection input, receiving a classification input from the user, and classifying a plurality of patent documents included in the patent pool according to the classification input from the user, predicting a second classification criterion for classifying the plurality of patent documents by the user based on the similarity between the classified patent documents, Classify the remaining patent documents in the patent pool that have not been classified according to the basic classification model and the second classification benchmark, The step of classifying the remaining unclassified patent documents in the patent pool according to the basic classification model and the second classification benchmark includes: The step of combining the second classification benchmark with the basic classification model to construct a user-customized classification model; and The step of using the user-customized classification model to classify the remaining patent documents in the patent pool that have not been classified; wherein the patent document classification method also includes: The step of deriving a first accuracy of automatic classification in real time and outputting an indicator representing the first accuracy in real time, The first accuracy is calculated based on the number of patent documents classified according to the second classification standard and / or the number of patent documents that the user has completed classification so far.

Images