KR101269441B1 - Apparatus and method for assessing patent infringement risks based on semantic patent claim analysis - Google Patents

Abstract

An apparatus and method for determining patent infringement based on analysis of semantic patent claims are disclosed. A data collection unit for extracting information about the claims of the patent, a vector construction unit for converting the information about the claims into a hierarchical subject-action-object (SAO) vector, and a tree in the transformed hierarchical SAO vector The patent infringement determination device based on the analysis of a patent infringement based on a semantic-based patent claim analysis including a infringement determination unit that determines a possibility of patent infringement by applying a matching algorithm and measuring the similarity with a target technology, provides a basis for determining whether a patent infringement is incurred. By considering the structural similarity and the content similarity of the claims at the same time, it is possible to derive a more accurate evaluation results.

Description

Apparatus and method for assessing patent infringement risks based on semantic patent claim analysis}

The present invention relates to a service system, and more particularly, to an apparatus and method for determining patent infringement based on semantic-based patent claim analysis.

Recently, patent disputes have been increasing due to patent infringement as patent troll companies, which exercise intellectual property rights only, do not intend to commercialize patents. Patent disputes not only incur primary monetary costs, such as enormous patent fees and litigation fees, but also administrative and time costs required to respond to patent disputes.They also stop sales and delays in the sale of related products or services, reduce market share, and reduce stock prices. This could lead to secondary financial losses. For example, in 1991, Kodak lost $ 920 million in damages following a patent lawsuit against Polaroid, and more than $ 3 billion in economic losses, including closing $ 1.5 billion and firing 700 people. Due to bankruptcy. In Korea, Samsung Electronics is required to pay royalties of $ 400 million by 2012 and $ 290 million by 2010 to Interdigit, a leading patent monster company.

As the risk of patent infringement increases, the strategic importance of monitoring the possibility of patent infringement is increasing. Countries around the world are making major efforts to revise major international trade laws, allowing companies to retaliate against patent infringers through trade sanctions, while promoting the risk of patent infringement and fostering patent information support services. At the company level, efforts to monitor patent activity of other companies and to purchase infringing patents in advance or to build a patent portfolio are also increasing.

In this method, a technique for evaluating the risk of patent infringement has been mainly used by experts to read related patents. However, this process is a time-consuming, labor-intensive task, and the difficulty is increasing as the number of patents grows exponentially and the complexity of the technology increases automated methods that can assist professionals in assessing the risk of patent infringement. The demand for the system is increasing.

In recent years, there has been an increase in attempts to provide automated methods to support expert decision making in the patent infringement risk assessment process, driven by increased access and availability of patent databases and advances in computer algorithms. Conventional techniques utilize text mining, quantitative bibliographic analysis, data mining, information visualization techniques, etc. to claim point maps, patent claim maps, keyword-based patent maps, semantic-based patent maps, and patent topics. Patent analysis and visualization techniques such as maps have been proposed. These techniques can be used to analyze a large amount of patent documents and to select patents that require expert perusal, but have the following problems in terms of data, methodology, and practicality in providing useful information for evaluating patent infringement risk.

First, there is a lack of in-depth analysis of claims in terms of data. Determination of patent infringement is based on the claims that legally define the scope of patent rights, so analysis of claims is essential to provide useful information for evaluating patent infringement risk. Prior arts, however, do not focus on the analysis of claims, and do not take into account the semi-structured nature of claims, which are made up of citations of independent and dependent claims, providing useful information for assessing patent infringement risk. There is a limit to doing so.

Second, in terms of methodology, there is a lack of specialized approaches to assessing patent infringement risk. Prior arts focus on general-purpose patent analysis, including statistical analysis of bibliographic information, citation relationship analysis, and similarity-based typing, and because there is no specialized approach to comparing patents and measuring similarity at the claims level, There is a limit to providing accurate information for evaluating risks of infringement.

Finally, there is a lack of verification of utility and performance in terms of practicality. In the case of a method of measuring similarity between patents based on a keyword vector using a text mining technique, the performance depends largely on the keyword, and the selection of the keyword may vary according to industry, technology, and analysts. Therefore, in the prior art based on the keyword vector, robustness verification is required for the keyword in order to obtain validity and provide practical utility in the field. However, in the prior art, little consideration is given to this.

The above-described background technology is technical information that the inventor holds for the derivation of the present invention or acquired in the process of deriving the present invention, and can not necessarily be a known technology disclosed to the general public prior to the filing of the present invention.

The present invention provides an automated patent collection, processing, analysis, and patent infringement risk assessment method, the patent infringement determination device based on the semantic-based patent claims analysis that can reduce the human, time, and economic costs required for patent information research and analysis And a method thereof.

In addition, the present invention, when evaluating the risk of patent infringement, considers the structural similarity and the content similarity of the claims, which are the basis of the determination of patent infringement, at the same time, the meaning-based patent claim analysis that can derive more accurate evaluation results It is an object of the present invention to provide a patent infringement determination apparatus and method thereof.

In addition, the present invention can be used as a tool for evaluating the risk of patent infringement in the early stages of R & D, thereby preventing the possibility of encroachment on patent disputes in advance, and reducing the cost of R & D and effective resource utilization. Analysis of semantic-based patent claims that can be used as a tool to assist in patent infringement and violations, patent purchase in preparation for patent disputes, and establishment of patent portfolio construction strategy by deriving patents and technologies with high potential for disputes It is to provide a patent infringement determination apparatus and a method based on the.

Another object of the present invention is to provide an apparatus and method for determining patent infringement based on semantic-based patent claim analysis, which can primarily select patents requiring expert perusal in evaluating a patent infringement risk.

The technical problems other than the present invention can be easily understood from the following description.

According to an aspect of the invention, the data collection unit for extracting information on the claims of the patent, the vector construction unit for converting the information on the claims to a hierarchical SAO (subject-action-object) vector and the An apparatus for determining patent infringement based on a semantic-based patent claim analysis including an infringement determination unit that applies a tree matching algorithm to a transformed hierarchical SAO vector and measures similarity with a target technology to determine a possibility of patent infringement is provided.

Here, the vector construction unit, a keyword extraction unit for extracting SAO-related keywords from the claims, a SAO set generation unit for generating a SAO _j (j = 1 to n) set by combining the extracted keywords and the claim When the claim included in the range includes the generated SAO _j 1, and does not include the generated SAO _j may include a SAO vector generator for generating a SAO vector by allocating 0.

In addition, the vector construction unit, by generating the nodes (nodes) corresponding to the plurality of claims included in the claims, and connecting the generated nodes by a link (link) according to the citation relationship between the corresponding claims (tree) It may further include a tree structure generating unit for constructing.

In addition, the vector construction unit, the node for assigning the SAO vector of the claim corresponding to the node included in the generated tree structure, and set the level of the node corresponding to the citation level of the plurality of claims included in the claims The setting unit may further include.

In addition, the infringement determination unit may determine the possibility of patent infringement by applying a tree matching algorithm to the transformed hierarchical SAO vector and measuring the similarity between the claims between the first patent and the second patent.

Here, the infringement determination unit may include a node matching unit for measuring a first similarity level and matching each other in the order of high first similarity level targeting the claims having the same level of the node, wherein the first similarity level is as follows. Cosine similarity index can be measured.

Where X is the SAO vector of the claims of the first patent and Y is the SAO vector of the claims of the second patent.

The node matching unit may add a dummy node to the second patent when there is no claim corresponding to the level of the node of the first patent among the claims of the second patent.

Here, the SAO vector of the dummy node may be a zero vector.

In addition, the infringement determination unit, whether the similarity between the i-th element by measuring the second similarity, which is defined as follows based on the matched nodes based on one independent term or one independent term and the dependent terms citing it It may further include a similarity measuring unit for determining.

Where L is the number of levels contained in the claim element, N _j is the number of nodes at the j th level, w (j) is the similarity weight for the j th level, and w (N _j ) is the number of nodes located at level j. Similarity weight, the similarity between the corresponding node (L _j ) is the cosine similarity index value between the matching nodes.

Also, w (N _j ) may be the inverse of the number of nodes located at level j (1 / N _j ), and w (j) may be specified as follows.

Here, the infringement determination unit may further include a patent infringement determination unit that determines a possibility of patent infringement between the first patent and the second patent by deriving a third similarity using the second similarity as follows. Can be.

Where n is the number of elements included in the claims.

According to another aspect of the present invention, a data collection unit for extracting information on the claims of the patent, SAO-related keywords are extracted from the claims, combining the extracted keywords SAO _j (j = 1 to n ) to generate a set, and if the claims included in the claims includes the generated SAO _j 1, by assigning a 0 if it does not contain the generated SAO _j generate SAO vector, the following claims Constructs a citation relationship of a plurality of claims included in a tree structure, allocates a SAO vector of a claim corresponding to a node included in the generated tree structure, and assigns a citation level of the plurality of claims included in the claims. The vector constructing unit which sets the level of the node correspondingly and the first similarity are measured in the order of the first similarity in order of the same claim with the same level of the node. And determine similarity between elements by measuring a second similarity based on the matched nodes based on one independent term or one independent term and dependent terms citing the same. An apparatus for determining patent infringement based on a semantic-based patent claim analysis including an infringement determination unit for deriving similarity and determining a possibility of patent infringement between a first patent and a second patent is provided.

Further, according to another aspect of the present invention, in the method of determining a patent infringement by the patent infringement determination device based on the semantic-based patent claim analysis, extracting information on the claims of the patent, the claim And converting the information about the hierarchical SAO vector into a hierarchical SAO vector, and applying a tree matching algorithm to the transformed hierarchical SAO vector and measuring similarity with the target technology to determine the possibility of patent infringement. Based on the patent infringement determination method is presented.

The information extracting step may include extracting SAO-related keywords from the claims, generating SAO _j (j = 1 to n) sets by combining the extracted keywords, and including the claims. If the claim does not include the first, the generated SAO _j if it contains the generated SAO _j may further include the step of generating a SAO vector by assigning a zero.

The vector converting may include: constructing a citation relationship of a plurality of claims included in the claims in a tree structure, and allocating SAO vectors of claims corresponding to nodes included in the generated tree structure; The method may further include setting a level of the node corresponding to the level of citation of the plurality of claims included in the claims.

Here, in the patent infringement determination step, the first level of similarity is measured with respect to the claim of the same level of the node to match each other in the order of high first similarity, one independent claim or one independent claim and the dependent By determining the similarity between elements by measuring a second similarity based on the matched nodes using the terms as elements, the third similarity is derived by using the second similarity to infringe the patent between the first patent and the second patent. Can be judged.

Here, the first similarity may be measured by the following cosine similarity index.

Where X is the SAO vector of the claims of the first patent and Y is the SAO vector of the claims of the second patent.

In addition, the second similarity may be measured by the following equation.

Where L is the number of levels contained in the claim element, N _j is the number of nodes at the j th level, w (j) is the similarity weight for the j th level, and w (N _j ) is the number of nodes located at level j. Similarity weight, the similarity between the corresponding node (L _j ) is the cosine similarity index value between the matching nodes.

In addition, the third similarity may be measured by the following equation.

Where n is the number of elements included in the claims.

According to another aspect of the present invention, a program of instructions that can be executed by a digital processing apparatus is tangibly implemented and read by the digital processing apparatus to perform the patent infringement determination method based on the semantic-based patent claim analysis described above. A record carrier is provided which records the program.

Other aspects, features, and advantages will become apparent from the following drawings, claims, and detailed description of the invention.

Patent infringement determination device and method based on semantic-based patent claim analysis according to the present invention provides automated patent collection, processing, analysis and patent infringement risk assessment method, the human, time, economic cost required for patent information investigation and analysis There is an effect to reduce the.

In addition, the patent infringement determination device and method based on the semantic-based patent claim analysis according to the present invention, when evaluating the risk of patent infringement, simultaneously consider the structural similarity and the content similarity of the claims that are the basis of the determination of patent infringement As a result, a more accurate evaluation result can be obtained.

In addition, the patent infringement determination apparatus and method based on the semantic-based patent claim analysis according to the present invention can be utilized as a tool for evaluating the risk of patent infringement in the early stages of research and development, thereby preliminary to the possibility of being involved in a patent dispute It is possible to cut off R & D costs and effectively utilize resources, and to derive patents and technologies owned by companies and patents with high potential for disputes, so that patent infringement and violations can be monitored, patent purchases for patent disputes, and patent portfolio construction It can be used as a tool to support strategy establishment.

In addition, the present invention has an effect that can be primarily selected for patents that need to be carefully read in the process of evaluating the risk of patent infringement.

1A is a block diagram illustrating a patent infringement determination apparatus based on analysis of semantic-based patent claims according to an embodiment of the present invention.
Figure 1b is a block diagram of a vector construction unit according to an embodiment of the present invention.
Figure 1c is a block diagram of an infringement determination unit according to an embodiment of the present invention.
1D is a flowchart of a method for determining patent infringement based on analysis of semantic-based patent claims in accordance with an embodiment of the present invention.
2 is a diagram illustrating a major patent infringement analysis method.
3 illustrates a hierarchical SAO vector of claims according to an embodiment of the present invention.
4 is a diagram illustrating a procedure for comparing similarities for determining patent infringement according to an embodiment of the present invention.
FIG. 5 illustrates four patent sets for verifying a patent infringement determination method based on semantic based patent claim analysis according to an embodiment of the present invention. FIG.
6 illustrates a hierarchical SAO vector according to an embodiment of the present invention.
FIG. 7 illustrates a keyword similarity matrix of site 1 in accordance with an embodiment of the present invention. FIG.
8 illustrates a keyword similarity matrix of set 2 according to an embodiment of the present invention.
9 illustrates a keyword similarity matrix of set 3 according to an embodiment of the present invention.
10 is a view summarizing the results of the t test of the method for determining patent infringement based on analysis of semantic-based patent claims according to an embodiment of the present invention.
11 to 13 is a view comparing the hit ratio of the patent infringement determination method based on the semantic-based patent claim analysis according to an embodiment of the present invention with the prior art.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It is to be understood, however, that the invention is not to be limited to the specific embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

Terms including ordinal numbers such as first and second may be used to describe various components, but the components are not limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as the second component, and similarly, the second component may also be referred to as the first component. And / or < / RTI > includes any combination of a plurality of related listed items or any of a plurality of related listed items.

When a component is referred to as being "connected" or "connected" to another component, it may be directly connected to or connected to that other component, but it may be understood that other components may be present in between. Should be. On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this specification, the terms "comprises" or "having" and the like refer to the presence of stated features, integers, steps, operations, elements, components, or combinations thereof, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

The term " part, "" module," " means, "or the like, which is described in the specification, refers to a unit for processing at least one function or operation and may be implemented by hardware or software or a combination of hardware and software .

In the following description of the present invention with reference to the accompanying drawings, the same components are denoted by the same reference numerals regardless of the reference numerals, and redundant explanations thereof will be omitted. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail.

Figure 1a is a block diagram of a patent infringement determination device based on the analysis of semantic-based patent claims according to an embodiment of the present invention. Referring to FIG. 1A, the patent infringement determination apparatus 100 may include a data collection unit 110, a vector constructing unit 120, an infringement determination unit 130, and a control unit 140.

In this embodiment, in evaluating the risk of infringement of a patent, it is possible to derive a more accurate evaluation result by simultaneously considering the structural similarity and the content similarity of the claims which are the basis of the determination of patent infringement. That is, the present embodiment provides a method of simultaneously expressing the structure and contents of claims, which are the basis for determining patent infringement, through hierarchical SAO vectors, and tree matching for measuring similarity between hierarchical SAO vectors. By developing and applying the algorithm, it is possible to more accurately derive the risk of patent infringement.

The present invention is characterized by comparing a patent and a target technology, for example, a first patent and a second patent, a patent, and a specific product to determine whether an infringement occurs. When comparing a patent with a specific product, the components of the specific product expressed in the claims may be compared with the patent. In the following description, a patent and a patent are compared with each other to focus on an embodiment for determining whether an infringement occurs.

The patent infringement determination apparatus 100 may be a device that a user can operate using an input device connected to the device, such as a local server or a computer terminal, or a device that can be accessed and operated by a user on a wired or wireless Internet. have. Therefore, the patent infringement determination apparatus 100 may be provided locally in a company or home, or may be connected to a wired or wireless Internet to be accessible to external users.

The data collection unit 110 extracts information about the claims of the patent. Here, the data collection unit 110 may extract information about the claims from a predetermined patent database or extract information about the claims from a patent document.

The patent database may be a database that stores various information such as patent documents and bibliographic information, for example, application date, application number, applicant, registration date, information about claims, invention cited, invention cited, and the like. Information about a claim includes a claim number, a claim content, and information indicating the claim (for example, phrases such as "claim", "claim n", "We claim", "What is claimed is", etc.). can do.

The data collection unit 110 corresponds to a technical field, a specific patent number, a predetermined search expression, a specific patent classification, etc. received from a user, and a patent searched by a patent number, a specific patent number, and a predetermined search expression corresponding to the corresponding technical field. Numbers, patent numbers corresponding to specific patent classifications (IPC, USPC, F-Term, ECLA, etc.) can be extracted, and information on claims corresponding to the corresponding patent numbers can be extracted.

In addition, when the data collection unit 110 extracts information about the claims from the patent document, the data collection unit 110 may extract the information about the claims using a data parsing technique. Can be. The parsing technique may have various methods, for example, a method of constructing a parsing tree by performing a parsing operation of parsing tokens resulting from lexical analysis according to a grammar. The parsing technique according to the present invention may track information indicating that the claims are mentioned in the patent document in the patent document, and extract information about the claims described in the structure associated with the word.

The vector constructing unit 120 converts the information about the claims into a hierarchical SAO vector. This embodiment can convert the claims into a hierarchical SAO vector using semantic structural text analysis. The hierarchical SAO vector expresses the citation relationship between the patent claims constituting the claims in a tree structure and the contents of each patent claim in an SAO vector.

A hierarchical SAO vector is described as follows. A claim is made up of one or more independent claims and dependent claims that define or specify an independent or other dependent claim, and the citation relationship between the claims can be distinguished by a particular phrase ("in claim n", etc.). The hierarchical SAO vector is composed of nodes corresponding to the patent claims, and the nodes have a tree structure connected by links according to the citation relationship between the patent claims (see FIG. 3).

A node corresponding to the patent is created at level 0, a node corresponding to the independent claim is created at level 1, and a link is made with a node located at level 0. A node corresponding to a dependent claim is generated at a lower level of a claim referred to by the dependent claim, and a link is made with a node corresponding to the referenced claim. The vector constructing unit 120 according to the present exemplary embodiment may store node information corresponding to the claims and link information for combining them with each other in a separate storage unit.

Each node of the hierarchical SAO vector (except for nodes located at level 0) has a SAO vector representing the contents of the corresponding patent claim, and the SAO vector can be generated by the following procedure. That is, the vector constructing unit 120 extracts SAO-related keywords from the claims, derives a SAO set (SAO _j : j = 1 to n) based on the extracted keywords, and if the patent claim includes SAO _j , SAO vectors can be generated by assigning 1, or 0 if not included.

To this end, as illustrated in FIG. 1B, the vector constructing unit 120 combines the keyword extracting unit 121 for extracting SAO-related keywords from the claims, and the extracted keywords, SAO _j (j = 1 to n). and SAO set generation unit 123 to generate a set, SAO assigning a 0 if it does not include the first, generated SAO _j if it contains SAO _j of the claims are created including the claims to generate SAO vector The vector generator 125 may be included.

In addition, the vector constructing unit 120 may further include a tree structure generating unit 127 for constructing a citation relationship of the plurality of claims included in the claims in a tree structure. The tree structure generation unit 127 may store node information corresponding to the above-described claims and link information combining them.

In addition, the vector constructing unit 120 may further include a node setting unit 129 that sets the level of the node corresponding to the citation levels of the plurality of claims included in the claims. As described above, the node setting unit 129 sets levels of each node according to patents, independent claims, and dependent claims so that the claims form one tree structure. Here, the citation level of the claims may be a level indicating the dependency of the claims. For example, a level 1 refers to an independent term, a level 2 refers to a dependent term directly referencing an independent term (first dependent term), a level 3 refers to a dependent term directly referencing a first dependent term (second dependent term), or the like. The level can be determined accordingly.

Infringement determination unit 130 determines the possibility of patent infringement by applying a tree matching algorithm to the transformed hierarchical SAO vector and measuring the similarity with the target technology. In other words, the infringement determination unit 130 performs a claim-by-claim comparison based on a tree matching algorithm on a hierarchical SAO vector to measure the similarity between claims and patents. Evaluate the potential for infringement.

Here, the tree matching algorithm may include three steps: matching claims and adjusting structure, measuring component similarities, and deriving final similarity. In the present specification, the similarity used to match a patent claim may be referred to as a first similarity, an element similarity measured for determining similarity, a second similarity, and a similarity calculated for determining a final patent infringement possibility as a third similarity. .

As illustrated in FIG. 1C, the infringement determination unit 130 may further include a node matching unit configured to measure the first similarity with respect to the claims having the same level of nodes and match each other in the order of high first similarity. Here, the first similarity may be measured based on various criteria. For example, the first similarity may be a cosine similarity index as described later in detail.

In addition, the infringement determination unit 130 may further include a similarity measuring unit for determining whether or not similarity by measuring the second similarity using the matched claim as an element. The second similarity is calculated using the level of the element, the number of nodes, the weight, the first similarity, and the like, and will be described in detail below.

In addition, the infringement determination unit 130 may further include a patent infringement determination unit determining a possibility of patent infringement between the first patent and the second patent by deriving a third similarity using the second similarity. Here, the third similarity is calculated by reflecting the number of claims in the second similarity, which will be described in detail below.

The control unit 140 controls each function unit so that the functions of the above-described data collection unit 110, the vector construction unit 120, and the infringement determination unit 130 can be efficiently operated.

1D is a flowchart of a method for determining patent infringement based on semantic based patent claim analysis according to an embodiment of the present invention. Each step to be described below may be performed by each functional unit of the patent infringement determination apparatus 100 as a subject.

In step S110, the data collection unit 110 extracts information on the claims of the patent from the patent database or the patent document. Here, the patent database is stored online by storing various information such as patent documents and bibliographic information, for example, application date, application number, applicant, registration date, information about claims, invention cited, invention cited, etc. It can be a database accessible by. In addition, in the case of extracting information on a claim from a patent document, the patent document may be a patent document capable of text extraction.

In step S120, the vector constructing unit 120 converts the information about the claims into a hierarchical SAO vector. As described above, the hierarchical SAO vector represents a citation relationship between the patent claims constituting the claims as a tree structure, expresses the contents of each patent claim as an SAO vector, and the SAO vector has a dimension set in advance according to the situation. It can be a vector.

In operation S130, the infringement determination unit 130 determines the possibility of patent infringement by applying a tree matching algorithm to the transformed hierarchical SAO vector and measuring similarity with the target technology. Here, the tree matching algorithm may include three steps of patent claim matching and restructuring, element similarity measurement and final similarity derivation, and the similarity measured in each step may be similarity measured by different equations.

According to an embodiment of the present invention, a method of analyzing a semantic patent claim is provided to support comparison of claims between patents. Recent advances in software technology have helped linguists to improve traditional text analysis methods.

One such method, semantic-based text analysis, can examine the correlation of words and phrases in text at various levels from individual sentence clauses, and can suggest a semantic-based structure that better understands the meaning of a document.

In particular, the term 'meaning-based' herein is used to refer to the use of a method that can identify dependencies between various patent claim elements and between patents in order to determine patent infringement and obtain more relevant information indicative of its risk. Can be.

The method according to an embodiment of the present invention is based on a tree matching algorithm that enables comparison between claims and hierarchical SAO vectors representing structured dependency and unstructured text information between claims, which is a preliminary step in predicting infringement. Experts can select patents for manual reading.

The proposed method according to an embodiment of the present invention may include a data collection and preliminary preprocessing step, a hierarchical SAO vector generation step and an inter-claim patent comparison step using a tree matching algorithm. The first step is data collection and preliminary preprocessing. Once the technical field to be investigated is selected, related patents can be collected in an electronic document format.

Claims specifying the protection scope of a patented invention are intermingled with other information, such as a summary, a description of the invention, and citation relationships, which are removed by parsing the patent document after selection.

Second, patent claims are converted into hierarchical SAO vectors using Java programs for structural text mining developed in-house. Finally, the possibility of patent infringement is investigated by comparing the claims between patents using the similarity index developed from the tree matching algorithm in consideration of keyword generation and semantic-based relationships of the claims.

Hierarchical SAO vectors and tree matching algorithms developed in accordance with embodiments of the present invention may reinforce the method according to the prior art by overcoming data and methodological limitations respectively. These methods are validated by statistically inspecting different keyword lists and comparing their performance with keyword-based techniques according to the prior art.

New technology databases and intelligent computing algorithms can mean that patent engineering and management are no longer dependent on technology and management professionals. Delphi's decision-making and consensus methods are of limited use, and in recent years, attempts have been made to use intelligent computing techniques to provide productive and well-organized information to researchers and industry professionals.

Compared to the method according to the prior art, the semantic-based patent claim analysis method according to an embodiment of the present invention has the difference that each claim is considered. This can be particularly useful in patent invalidation, patent litigation and R & D exploration, and can be a starting point for developing more general models.

Hereinafter, the present invention will be described in order of applying and concluding the embodiment of the present invention with respect to DNA chip technology after explaining the general background of the present invention and the proposed technique.

Patent infringement is a general term related to the unauthorized production, use, or transfer of a patent invention, although it refers to various definitions and scopes in literature, and is related to various situations such as patent disputes, civil litigation, criminal litigation, and patent invalid litigation. .

Determination of patent infringement concentrates on the analysis of claims, which define the scope of protection afforded by technical terms and play an important role in the checklist used to determine patent infringement.

The United States Patent and Trademark Office (USPTO) has recorded about 6 million patents since 1976, with an increase of about 150,000 per year. These numbers require time and labor that is practically impossible to peruse by hand, a common way of judging violations.

As a result, several attempts have been made to reduce the manual effort and systematize it, one of which is the automatic selection of patents that need to be read by hand.

Patents include analysis items that can be divided into two broad categories: structured items and unstructured items. Structured items are, for example, patent numbers, filing dates, inventors and applicants (assignees), and the like, and are represented in a consistent sense and form, and unstructured items are, for example, detailed descriptions and claims of the invention, It is expressed in text and can have different structures and formats.

The main techniques for determining patent infringement can be divided into two categories according to the data type and the methods used, as shown in FIG. 2.

The first group used text mining techniques to transform unstructured items into structural items, and analyzed the patent infringement risk by comparing patent keyword vectors to find similarity between patents.

Yoon, Yoon & Park (2002) (On the development and application of a selforganizing feature map-based patent map.R & D management, 32 (4), 291-300.) Claim point maps using a selforganizing feature map (SOFM). Suggested. Keyword vectors were derived from patent claims, clustered and visualized by SOFM algorithms, and allowed the claim point map to place key technologies related to patents on a two-dimensional display.

Huang, Ke & Yang (2008) (Structure clustering for Chinese patent documents.Expert Systems with Applications, 34 (4), 2290-2297.) Worked similarly on Chinese patent documents, We propose a structural SOFM that reflects the implicit structure.

Lee, Yoon & Park (2009) (An approach to discovering new technology opportunities: Keyword-based patent map approach.Technovation, 29 (6/7), 481-497) instead of SOFM to cluster and visualize keyword vectors of patents. Principal component analysis (PCA) was used and Bergman et al. (2008) (Evaluating the risk of patent infringement by means of semantic patent analysis: The case of DNA chips.R & D Management, 38 (5), 550-562.) An analysis was proposed, and the concept was extended by integrating it with the SAO analysis, and showed that the similarity between patents can be calculated from the same SAO number.

Similar technical efforts extract structured information (Nishida & Takamatsu, 1982) (Structure-information extraction from patent-claim sentences.Information Processing and Management, 18 (1), 1-13.), And categorize patents ( Kim & Choi, 2007) (Patent document categorization based on semantic structural information, Information Processing and Management, 43 (5), 1200-1215.), Searching for patents (Kang, Na, Kim, & Lee, 2007) (Cluster- based patent retrieval.Information Processing and Management, 43 (5), 1173-1182.)

The second technique statistically analyzes the relationship between structured patent items (quotations, classifications, etc.) in order to obtain information that may indicate the possibility of patent infringement, where the probability of patent infringement increases as the patent cites more prior patents.

Sternitzke, Bartkowski, & Schramm (2008) (Visualizing patent statistics by means of social network analysis tools.World Patent Information, 30 (2), 115-131.) Applies social network analysis to patent citation data to help Rohm (Nichia The patent claims cited more than 80 patents in the technical network, especially those held by Nichia. Kasravi & Risov (2009) (Multivariate patent similarity detection. Proceedings of Hawaii International Conference on System Sciences, 42 (2), 1132-1139.) Utilizes several patent items, such as citations, classifications, and keywords, among the text elements. Monitor trends in technology, find similar patents and track technology history.

Shin & Park (2005) (Generation and application of patent claim map: text mining and network analysis.Journal of Intellectual Property Rights, 10, 198-205.) Utilizes social network analysis to map patent claims based on citation and classification data. Where a node represents a patent number, a classification or an applicant (assignee) and represents a similarity judged by an expert.

All these methods may be useful in some situations, but have the following limitations, so the present invention is presented to overcome these limitations.

In the prior art, the claim data at the heart of the litigation process is insufficiently considered (data limit). Some studies (Yoon, Yoon, & Park, 2002; Shin & Park, 2005; Huang, Ke, & Yang, 2008) have used them to examine patent infringement, but have semi-structured claims. Data (unstructured elements are combined in the same structural way as dependencies) is still not taken into account.

Most of the prior arts were inadequately applied versions of conventional statistical methods and could support patent analysis for general purposes, but were not focused on infringement (methodological limitations). They are often too complex to understand and operate, and they don't handle data well.

The performance of the method according to the prior art is hardly validated. Text mining performance may vary according to keyword definitions, and prior art methods need to be more robust to be externally verified and useful to industry professionals (utility issues).

Hereinafter, the basic concept of a semantic-based patent claim analysis model according to an embodiment of the present invention will be described. As shown in FIG. 1D, this embodiment utilizes various methods, such as data parsing techniques, structural text mining, and tree matching algorithms, to determine the potential for patent infringement.

While many conventional analytical methods and complex algorithms can actually use the wrong concept and be used incorrectly, this embodiment can overcome this by including the following three steps. That is, the data collection and preprocessing step, the step of transforming a patent claim into a hierarchical SAO vector capable of representing textual information as well as the dependent semantic relationship of the claim, and the similarity index based on the tree matching algorithm, The determining may be included in this embodiment.

Although the requirements for describing patent claims vary from country to country, this embodiment is based on data in the USPTO database. This is because the United States is the world's largest market and the most representative, because patents in other countries are often filed in the United States at the same time. The US database is well-organized in terms of search terms and reliability, and historically provided electronically up to 1975, with 50 titles linked to the entire patent document on screen.

In the United States, a method called peripheral claiming is mainly used, which means that the claims define the periphery of the invention and the entire scope of the patent, which is very important in patent litigation.

These claims may be divided into components connected to one another according to structure or function. Independent claims are generally self-supporting claims that do not refer to other claims, and dependent claims generally express specific features of the invention and are claims claiming one or more other independent or dependent claims. The citation relationship between different claims can be an important factor in accurately analyzing patent analysis and legal protection.

As Internet services evolve, it is easier to access patent databases electronically. Patent documents of the selected field are collected from the USPTO database corresponding to the relevant search conditions. At this stage, some patent data is unstructured, i.e. represented only in text, and needs to be processed so that claims can be separated from unnecessary elements such as abstracts, detailed descriptions of the invention, and citation information. Here, a claim may be placed between reference information and a detailed description and generally starts with phrases such as “claims”, “We claim” or “What is claimed is”.

Since the independent claims and the dependent claims are categorized into quoted phrases, such as, for example, "the method of claim n wherein", "according to claim n", and so on, the patent claims are not only textual information, It can be transformed into a hierarchical SAO vector representing semantic (ie, dependency) relationships between claims.

Referring to FIG. 3, a tree structure is shown in which nodes and links respectively correspond to patent claims and relationships. In particular, the patent number becomes the root node, the node of the first level represents the independent claim, and the node of the subsequent level represents the dependent claim.

The keyword vector can be generated for each node by structural text mining of the claims through the four-step keyword extraction process as follows. The method may include a supplementary word removal step, a stem word specification step, a statistical analysis step, and a keyword vector generation step. As shown in Fig. 3, 1 is assigned to the associated vector field corresponding to the portion of the claim containing the particular keyword.

Patent claim elements (each claim in the claims) are represented by a structural hierarchical SAO vector, as shown in FIG. It can be expressed in a variety of forms to represent dependent relationships, for example, from a simple tree with root and first level nodes, with multiple child nodes at the first and subsequent levels. More complex forms can be expressed.

The final step is to compare the claims with each other to identify possible patent infringement using these vectors. Referring to Figure 4, a tree matching algorithm is used to measure the similarity between the claims of the two patents, including the following three substeps. The sub-steps may include claim matching and restructuring steps, element similarity measurement steps, and final similarity derivation steps, and the proposed algorithm may be implemented with the following pseudo code.

0: CoSim (nodeA, nodeB) {

1: return DotProduct (nodeA, nodeB) / (| NodeA || NodeB |)

2:} // Measuring the cosine similarity between two nodes

3: Multi_Sim_Level (patA, patB, simBase) {

4: maxNodeCount = MAX (nodeCount (patA), nodeCount (patB));

5: return Multi_Sim (patA, patB, simBase) / maxNodeCount;

6:}

7: Multi_Sim (setA, setB, simBase) {

8: foreach (ai in SetA) {

9: foreach (bj in SetB) {

10: nodeSim = CoSim (ai, bj) * simBase;

11: if (HasChild (ai) or HasChild (bj)) {

12: nodeSim * = 0.5;

13: nodeSim + = Multi_Sim_Level (chidSet (ai), childSet (bj), simBase * 1/2);

14:}

15: patSim = nodeSim + Multi_Sim (SetA-ai, SetB-bj, SimBase)

16:}

17:}

18: return pat Sim;

19:}

Sub step 1 S110 is a claim matching and restructuring step. Different dependencies between the claims produce different hierarchical SAO vectors, which need to be adjusted to compare the claims. The independent claim, that is, the claim of the first level, is matched first, and the sequential dependent claims are matched with each other for similarity, and then the elements of the following tree structure are matched. According to this embodiment, the following cosine similarity index is adopted to measure the similarity between unstructured documents.

(1)

(One)

Where X is the SAO vector of the claims of the first patent and Y is the SAO vector of the claims of the second patent. SAO _j set is composed of (subject, action, object). Here, subject is defined as a keyword related to the subject, action is a keyword related to the action, and object is a keyword related to the purpose of the action. For example, in English, the subject may be an array of biopolymers, an action is a form, and an object is a microarray. The SAO set may be (array of biopolymers, form, microarray). The keyword extraction unit 121 extracts keywords related to subject, action, and object from each claim of the claims. If L keywords for subject, M keywords for action, and N keywords for object are extracted, the SAO set generator can generate a total of LxMxN SAO sets based on these combinations. have.

Referring to FIG. 4A, a method of using hierarchical SAO vectors of patents A and B as input data is illustrated. Referring to FIG. 4B, first level nodes 1, 5, and 6 of patent A are compared with nodes 1 and 4 of patent B in terms of similarity.

The most similar matches are A1: B1, A6: B4, and since node B of patent B has no further claims at that level, node 5 of patent A matches dummy node 7 of patent B added to the hierarchical SAO vector. Here, node 7 is a keyword vector {0, 0,... , 0}. This same process can be applied at the following levels until all claims between the two patents match each other.

Sub-step 2 (S120) is an element similarity measuring step. Referring to FIG. 4C, it is shown how the similarity of the corresponding claim elements is measured. Because each claim element and each level of each element represents a different aspect of the invention, the model according to an embodiment of the invention reflects the similarity and differences between nodes at each level.

The nodes of the first level contain the most general content, and their similarity is given a weight of 0.5. Nodes of the second level are weighted with a similarity weight of 0.25, including specified elements of the contents of the nodes of the first level, and weighted in the same way for nodes of the next level. Here, the weight assigned may be one example, and other numerical weights may be applied to the present invention. In order for the similarity level to be from 0 to a maximum of 1, the node weight is given the same weight as the previous level node for the last level. In addition, according to the elements at the same level, the similarity of the i th claim element may be derived as follows.

(2)

(2)

Where L is the number of levels contained in the claim element and N _j is the number of nodes at the j th level. w (j) is the similarity weight for the j th level and w (N _j ) is the similarity weight of the node located at level j. The similarity between corresponding nodes (L _j ) is the cosine similarity index value between matched nodes.

The similarity measurement method may vary depending on the analysis situation, and when the analysis method according to an embodiment of the present invention is used to determine the validity of a claim, a similarity index may be generated based on the same number of claims, and a cosine such as Equation (1) may be used. Indicators can also be used to determine overall similarity.

In addition, w (j) can be determined as follows.

(3)

(3)

Sub step 3 (S130) is a final similarity derivation step. The likelihood of a patent infringing another patent is derived by summing the similarity of the claim elements. The summation here again may depend on the situation of infringement.

The final similarity indicator may simply be the total number of identical claims. For example, the present embodiment may use a weighted summation method to derive the final similarity by weighting the inverse of the number of nodes of the first level. In FIG. 4 (d), since each patent has three first level nodes (including dummy node 7), the weight is 1/3.

The final similarity index, which is the third similarity, may be defined as follows.

(4)

(4)

Here, n may be the number of claims (elements) included in the claims. It may be determined that the greater the value of the similarity index, the greater the similarity between the two technologies.

The similarity derived according to the proposed method is influenced not only by the claim level but also by the occurrence of keywords. The higher the higher level claims of the two patents are similar, the higher the final similarity indicator will be.

In order to explain the analysis method according to an embodiment of the present invention, Affymetrix Inc. and Synteni Inc. We present a comparative analysis of the DNA chip patents in question.

(1) technical scope, (2) increasing trend in the number of patent infringement occurrences in this field, (3) well documented litigation results, (4) appropriateness of the number of patents to describe embodiments of the present invention, and ( 5) Considering the importance of technology for the development of next-generation products for various industries, this comparative analysis seems appropriate.

Ninety-nine patent documents were extracted from the US Patent and Trademark Database regarding DNA chips, which were separated by parsing by claim and title. The compiled data consists of four sets as shown in FIG.

Set 1 includes Affymetrix Inc. (US-5807522) and Synteni Inc. (US-5800992, US-5143853, US-5445934) consisting of four key patents relating to infringement litigation.

Set 2 constitutes 11 extended core patents with the addition of seven patents directly related to the core patent of Set 1. Of the additional patents, US-5695940, US-5525464 and US-5700637 are related to two other litigations of Affymetrix Inc., US-5763263 and US-5830645 are very similar to Synteni's US-5807522, and US -5424186 and US-5510270 are part of Affymetrix's family patent.

Set 3 consists of a total of 50 patents, including 39 more patents with similar subject matter extracted from subjects similar to Set 1's core patent. Set 4 consists of a total of 99 patents, further including 49 patents randomly extracted for comparison purposes. Such data include Bergmann et al. (2008) (Evaluating the risk of patent infringement by means of semantic patent analysis: The case of DNA chips.R & D Management, 38 (5), 550-562.).

As mentioned above, the organized context will determine how to define the keyword list. For the purpose of verifying this embodiment by applying it to a real situation, an expert and a text mining program were used to extract three keyword lists from each of sets 1, 2, and 3 patents.

The set 1 list best fits the core patent subject, the set 2 list contains terms of closely related patents, and the set 3 list contains general terms found in patents on the same subject. Based on this list, patent claims are automatically converted into hierarchical SAO vectors by a Java-structured text mining program developed by the inventors. Referring to FIG. 6, a sample of keyword vectors derived from patent US-5807522 is shown.

99 patents were compared in pairs, and because the number of pairs that can be generated is very large (99.98 / 2 = 4,851), the comparison is complex and it is practically impossible to compare them both structurally and processally. We have developed a Java-based program to automatically calculate the similarity between pairs.

This is shown as a 99 * 99 similarity matrix, which does not show the entirety due to lack of space, but FIGS. 7, 8, and 9 illustrate portions of three similarity matrices based on keyword lists extracted from sets 1, 2, and 3, respectively. Indicates.

The similarity between the claims was measured by the method according to the embodiment of the present invention and the conventional keyword vector-based method. In the conventional keyword vector-based technique, keywords are extracted based on TF-IDF (Term Frequency-Inverse Dopcument Frequency) without distinguishing between subject, action, and object in extracting keywords from the claims. In addition, in constructing keyword vectors representing claims, the entire claims are regarded as one unit without consideration of individual claims.

In order to control the difference by the similarity index, the conventional keyword vector based method also measures the similarity between keyword vectors using the cosine similarity index. To compare the performance of the two techniques, see Affymetrix Inc. Several analyzes of patents were performed.

First, in terms of ranking similarities for keyword lists extracted from sets 1, 2 and 3, the method according to an embodiment of the present invention provides a total of 10, 8 and 5 from a set 2 group directly related to the core patent. Patents were identified, which was above the respective critical mean values of 0.078, 0.014 and 0.001, whereas the prior art technique identified only 8, 6 and 5 patents each.

Second, in terms of significance, FIG. 10 shows the results of a two-tailed t test (independent sample t test), in which sample sizes are not equal. That is, the result of comparing the average similarity between the patents belonging to patent US-5807522 and set 2 and the average similarity between patents belonging to patent US-5807522 and set 3 only through an independent sample t test. According to this, the method according to the embodiment of the present invention shows a significant difference between the similarity between the core patent and the patents directly related thereto (Set 1 and 2) and the patent of similar subject matter (Set 3), The keyword vector based technique shows a significant difference only in the set 1 keyword list.

Here, the purpose of performing the t test is that the method according to the embodiment of the present invention has a similarity between the patents in set 2 and patent US-5807522 compared to the conventional keyword vector-based technique. It is statistically determined how much higher the similarity with -5807522 is. The purpose of this study is to determine whether there is a significant difference in the similarity means of the two groups.

The t test is a statistical method for testing whether two groups have the same mean. There are an independent sample t test and a corresponding sample t test. In other words, the independent sample t test considers two groups as independent, calculates the mean of each sample, and then uses the difference of the sample means to test. If the two groups are not independent, then a corresponding sample t test can yield more accurate test results. For example, if group 1 is the heart rate of 20 patients taking existing medication, and group 2 is the heart rate of 20 "same" patients taking new medication, the independent sample t-test calculates the average heart rate of the two groups, respectively. The mean of group 1 and the mean of group 2 are then tested. The corresponding sample t-test calculates the heart rate for each of the 20 patients (heart rate at the time of taking the existing drug—heart rate at the time of taking the new drug) and then calculates the mean for the difference to test whether the mean is equal to or different from zero. Therefore, in order to perform the corresponding sample t test, the two groups must have the same number of data. In the present invention, 1) the number of patents belonging to set 2 and 3 belonging to the comparison target is different, and therefore, the corresponding sample t test cannot be performed. It is more appropriate to use the t test.

In addition, the technique according to the prior art finds a higher similarity between the patents of Set 4 and Set 2 which are randomly extracted than the Patents of Set 2 and Set 3, which incorrectly selects the patents to be manually read. It represents the inferiority of vector-based techniques.

Finally, when analyzing the hit ratio, that is, the case classified correctly, when calculating the number of core patents and directly related patents among the top n similar patents, FIGS. 11, 12, and 13 For reference, in all cases, the performance of the proposed semantic based patent claim analysis method is better than that of the keyword vector based method. Taken together, the method according to the embodiment of the present invention is verified to have good performance in terms of accuracy and robustness, and appears to provide more accurate indicators for determining the possibility of patent infringement.

As described above, because the number of related patents is very large, manual reading to detect the risk of patent infringement is time consuming and labor intensive, making it impractical. The method according to an embodiment of the present invention proposes a more efficient and automated method for screening infringing patents requiring further investigation.

The method according to an embodiment of the present invention is based on a hierarchical SAO vector, which analyzes textual information as well as semantic (dependent) relationships between patent claims. This not only develops similarity indicators from comparisons between claims, but also presents a tree matching algorithm to account for potential patent infringement.

Comparative analysis of the DNA chip technology shows that the method according to the embodiment of the present invention has a significant advantage in terms of accuracy and robustness over the technique according to the prior art.

The present invention is noted in terms of contribution and potential utility. First, the method according to an embodiment of the present invention may contribute to the technology management theory as a first attempt to analyze the risk of patent infringement through patent comparison between claims. The present invention overcomes the limitations of the prior art relying on patent keyword comparisons in view of the semantic-based relationship of the claims.

In addition, the present invention has shown excellent performance in extracting, expressing and analyzing patent infringement information, and its functional efficiency is enhanced by self-developed software which helps to quickly investigate the possibility of patent infringement. In addition, the method according to an embodiment of the present invention can be used in various research fields, such as intellectual property management and R & D screening, and may be a starting point for developing a more general model.

Secondly, according to an embodiment of the present invention, hierarchical SAO vectors are developed for transforming semi-structured documents into structural data, and algorithms designed to measure similarity between them are presented. Although the present invention has been developed focusing on patent infringement, it is not limited thereto and may be applied to other types of documents for other purposes.

A large number of business and technical documents, represented in XML form on the Web, have not yet been analyzed, which may represent a gap in business or technology, important as a potential source for creating new businesses and developing new technologies, and in the future This is the area to be studied. Hierarchical SAO vectors and tree matching algorithms in accordance with the present invention can help with this effort.

Despite the various possibilities presented by the present invention, this research is still in its infancy. In order to improve the performance of the present invention, the following points can be considered. That is, although the opinions of experts can significantly improve the accuracy and reliability of the present invention, the present invention has not yet reflected administrative and technical judgments. Thus, other types of methods, such as SAO analysis, may be incorporated to improve the accuracy of the present invention.

In addition, in view of limitations, the following areas need to be studied in the future. First, in order to improve the performance of the semantic-based patent claim analysis of the present invention, a systematic technique needs to be developed, and a method of integrating other methods such as SAO analysis and expert input method can be defined in detail.

Second, in order to apply the present invention to product infringement against patents, there is a specific need for a method of associating a product's components with patent information, and an analysis tool such as a product / technical tree and a qualitative function arrangement. Features and associations can be usefully represented.

Third, developing other indicators related to different types of patent infringement will contribute to expanding and diversifying the scope of usefulness of the present invention. Finally, more verification of patents of various scopes may help to increase the external validity of the present invention. However, it is clear that the accuracy and robustness proposed by the present invention substantially contribute to the research and actual situation.

In addition, a detailed system configuration diagram of a patent infringement determination apparatus based on semantic-based patent claim analysis according to an embodiment of the present invention, a common platform technology such as an embedded system, an O / S, and interface standardization technology such as a communication protocol and an I / O interface Detailed description of the present invention will be omitted since it is obvious to those skilled in the art to which the present invention pertains.

The patent infringement determination method based on the semantic-based patent claim analysis according to the present invention may be implemented in the form of program instructions that can be executed by various computer means and recorded in a computer-readable medium. That is, the recording medium may be a computer-readable recording medium having recorded thereon a program for causing a computer to execute the steps described above.

The computer readable medium may include a program command, a data file, a data structure, etc. alone or in combination. The program instructions recorded on the medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as a hard disk, a floppy disk and a magnetic tape, optical recording media such as CD-ROM and DVD, magnetic recording media such as a floppy disk Optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like.

In addition, each of the above-described components may be implemented in one physically adjacent component or may be implemented in different components. In the latter case, each component may be controlled by being located adjacent or in different zones. In this case, the present invention is provided with a separate control means or control room for controlling each component to control each component by wire or wirelessly. You may.

In the above description, the respective components and / or functions described in the embodiments may be combined and combined, and those skilled in the art will understand that the present invention It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention.

100: patent infringement determination device 110: data collection unit
120: vector construction unit 121: keyword extraction unit
123: SAO set generator 125: SAO vector generator
127: tree structure generation unit 129: node setting unit
130: infringement determination unit 131: node matching unit
133: similarity measurement unit 135: patent infringement determination unit
140: control unit

Claims (22)

A data collection unit for extracting information on the claims of the patent;
A vector constructing unit converting the information on the claims to a hierarchical subject-action-object vector; And
Including a violation determination unit for applying a tree matching algorithm to the transformed hierarchical SAO vector and measuring the similarity with the target technology to determine the possibility of patent infringement,
The vector construction unit,
A tree structure generation unit for constructing a citation relationship of a plurality of claims included in the claims in a tree structure; And
Further comprising a node setting unit for setting the level of the node corresponding to the citation level of the plurality of claims included in the claims,
The infringement determination unit determines a possibility of patent infringement by applying a tree matching algorithm to the transformed hierarchical SAO vector and measuring the similarity between the claims between the first patent and the second patent,
The infringement determination unit may further include a node matching unit configured to measure a first similarity with respect to a claim having the same level of the node and match each other in the order of the highest similarity,
The first similarity is measured by the following cosine similarity index,

X is the SAO vector of the claims of the first patent, Y is the SAO vector of the claims of the second patent,
The infringement determination unit,
And a similarity measuring unit for determining whether or not similarity is obtained by measuring a second similarity defined as the following formula using the matched claim as an element.

Where L is the number of levels contained in the claim element, Nj is the number of nodes at the j-th level, w (j) is the similarity weight for the j-th level, w (Nj) is the similarity weight of the node at level j The similarity between corresponding nodes Lj is the cosine similarity index value between matched nodes.
The method of claim 1,
The vector construction unit,
A keyword extraction unit for extracting SAO related keywords from the claims;
A SAO set generating unit generating a SAO _j (j = 1 to n) set by combining the extracted keywords; And
If it contains SAO _j the claims included in the claims, the generated first, the generated if they do not contain the SAO _j means for allocating zero including SAO vector generation unit for generating a SAO vector based on patent claim Analysis Based on patent infringement judgment device.
delete delete delete delete delete delete The method of claim 1,
And w (N _j ) is an inverse of the number of nodes located at level j (1 / N _j ).
The method of claim 1,
The w (j) is a patent infringement determination apparatus based on the semantic-based patent claim analysis, characterized in that the following formula.

The method of claim 1,
The infringement determination unit,
Based on the meaning-based patent claim analysis further comprising a patent infringement determination unit for determining the likelihood of patent infringement between the first patent and the second patent by deriving a third similarity using the second similarity as follows. Patent infringement judgment device.

Where n is the number of elements included in the claims.
The method of claim 1,
The infringement determination unit,
Apparatus for determining patent infringement based on analysis of semantic-based patent claims, wherein a dummy node is added to the second patent when there is no claim corresponding to the level of the node of the first patent among the claims of the second patent.
13. The method of claim 12,
Apparatus for determining patent infringement based on semantic based patent claim analysis, wherein the SAO vector of the dummy node is a 0 vector.
delete In the method of determining patent infringement by a patent infringement determination apparatus based on semantic-based patent claim analysis,
Extracting information about the claims of the patent;
Converting information about the claims into a hierarchical subject-action-object (SAO) vector; And
Applying a tree matching algorithm to the transformed hierarchical SAO vector and measuring similarity with a target technology to determine a possibility of patent infringement,
The vector conversion step,
Constructing a citation relationship of a plurality of claims included in the claims in a tree structure; And
Setting a level of the node corresponding to the level of citation of the plurality of claims included in the claims;
Determining the possibility of patent infringement,
Applying a tree matching algorithm to the transformed hierarchical SAO vector and determining the likelihood of patent infringement by measuring the similarity between the claims between the first and second patents,
A first similarity is measured with respect to the claim of the same level of the node, and the first similarity is matched with each other in order of high first,
The first similarity is measured by the following cosine similarity index,

X is the SAO vector of the claims of the first patent, Y is the SAO vector of the claims of the second patent,
The method for determining patent infringement based on a semantic-based patent claim analysis further comprising a similarity measuring unit for determining whether the similarity is determined by measuring a second similarity defined as the following formula using the matched claim as an element.

Where L is the number of levels contained in the claim element, Nj is the number of nodes at the j-th level, w (j) is the similarity weight for the j-th level, w (Nj) is the similarity weight of the node at level j The similarity between corresponding nodes Lj is the cosine similarity index value between matched nodes.
16. The method of claim 15,
The information extraction step,
Extracting SAO-related keywords from the claims;
Combining the extracted keywords to generate a SAO _j (j = 1 to n) set; And
1, the means based on patent claim analyzed further comprising generating a SAO vector by assigning a 0 if it does not contain the generated SAO _j if it contains SAO _j the claims included in the claims, the generated Based on patent infringement determination method.
delete delete delete delete 16. The method of claim 15,
In determining the possibility of patent infringement,
Determination of patent infringement based on semantic-based patent claim analysis characterized by determining a possibility of patent infringement between the first patent and the second patent by deriving a third similarity defined using the second similarity as follows. Way.

Where n is the number of elements included in the claims.
A program of instructions that can be executed by a digital processing device is tangibly implemented and digitally processed to perform a patent infringement determination method based on the semantic-based patent claim analysis according to any one of claims 15, 16 and 21. Recording medium recording a program that can be read by the device.

Images