KR20110018169A - Apparatus and method for forecasting patent trouble - Google Patents

Abstract

PURPOSE: A patent dispute apparatus for acquiring prediction information through an estimating unit is provided to predict the possibility of trouble through various information about the patent. CONSTITUTION: An input data storage unit(110) stores input data by an analysis unit. An output data storage unit(150) stores output data including analysis information about dispute possibility. A map learning unit(120) generates a mining model in order that the input data is classified by the output data. A prediction unit(130) inputs input data about prediction target patent to learned neural network. The prediction unit acquires dispute prediction information.

Description

Apparatus and method for forecasting patent trouble}

The present invention relates to the use of patent information, and more particularly, to a method and apparatus for predicting the possibility of future dispute using various information about a patent.

Patent documents refer to documents that are published or registered through the Patent Office. Such general patent documents are very useful for identifying trends such as a specific technical flow or a specific applicant's research and development direction, unlike general information that is scattered through the Internet.

However, at present, there is a problem in that the patent analysis is merely provided by analyzing and providing the patent trend, and the information analyzed and analyzed is not processed to predict and provide a possibility of patent dispute.

The present invention is to provide a patent dispute prediction apparatus and method that can be provided by predicting the possibility of dispute using a variety of information on the patent.

According to one aspect of the present invention, an apparatus capable of predicting the possibility of dispute over a patent is provided.

According to an embodiment of the present invention, an input data storage unit for storing input data including patent holder information, technical classification information, or a combination thereof for each analysis unit; An output data storage unit for dividing and storing output data including classification information on a possibility of dispute for each analysis unit; And a guidance learning unit configured to generate a mining model so that the input data is classified according to the output data.

The classification information is a historical dispute, the mining model is a neural network model, the technical classification information may include a technical classification, patent classification and literature vector.

The document vector may be a size of a term frequency (TF) -inverse document frequency (IDF) value of each keyword set extracted from a claim for the patent for each analysis unit.

The keyword set may be a dimension of a representative similar word of a keyword set extracted from a claim for a patent for each analysis unit.

The analysis unit is one patent or a plurality of patents, and when the analysis unit is a plurality of patents, the analysis unit is a patent owned by the same patent holder, and the similarity between family patents, patents owned by the same patent owner in a citation relationship, and document vectors. May be any one or a combination of patents owned by the same patent holder with a reference value or more.

When the analysis unit is a plurality of patents, the patents owned by the same patent holder may be clustered, and a representative value of input data for each of the clustered patents may be calculated to configure input data corresponding to the analysis unit.

The representative value may be calculated using any one of an average value, a maximum value, a minimum value, and a sum.

The input data may include at least two technical classifications.

According to another aspect of the present invention, a method for predicting a dispute over a patent by a patent dispute prediction apparatus is provided.

According to an embodiment of the present invention, in a method in which a patent dispute prediction apparatus predicts a dispute over a patent, (a) an analysis unit for dispute prediction using one or more of patent information and bibliographic information on a patent for each analysis unit Generating input data for each patent and registering the same in a database; (b) inputting the input data corresponding to the registered patent among the input data into the neural network for predicting the dispute of the patent, and using the dispute prediction information outputted and the dispute information preset according to the registered patent, Learning by changing weights; And (c) inputting the input data corresponding to the predicted patent to the learned neural network to obtain dispute prediction information about the predicted patent.

Step (a) may include extracting a keyword set from patent information of the patent using an association rule algorithm; Calculating a terminal frequency (TF) and an inverse document frequency (IDF) using the extracted keyword set; And constructing a literature vector corresponding to the patent using the calculated TF and IDF.

If the analysis unit is one patent or multi-patent, and the analysis unit is a multi-patent, step (a),

If the analysis unit is a multi-patent, step (a), the step of clustering similar patents owned by the same patent holder to the patent; And calculating representative values for the clustered similar patents.

The extracting of the similar patent may include extracting, as a similar patent, at least one of a family patent, a patent that is directly or indirectly cited or cited in correspondence with a set patent, or a patent having a similarity of a reference vector or higher.

The input data includes a plurality of technical classifications, including a literature vector corresponding to the patent, wherein steps (b) and (c) input at least two or more technical classifications of the plurality of technical classifications into the neural network. can do.

By providing a patent dispute prediction apparatus and method according to the present invention, it is possible to predict and provide a possibility of dispute using a variety of information about the patent.

As the invention allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail in the written description. However, this is not intended to limit the present invention to specific embodiments, it should be understood to include all transformations, equivalents, and substitutes included in the spirit and scope of the present invention. In the following description of the present invention, if it is determined that the detailed description of the related known technology may obscure the gist of the present invention, the detailed description thereof will be omitted.

The terms first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another.

The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting of the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, the terms "comprise" or "have" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, and one or more other features. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, components, or a combination thereof.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram schematically illustrating an internal configuration of a patent dispute prediction apparatus according to an embodiment of the present invention, and FIG. 2 is a diagram schematically showing an internal configuration of a data registration unit according to an embodiment of the present invention. 3 is a flowchart illustrating a method of constructing a document vector according to an embodiment of the present invention, and FIG. 4 is a flowchart illustrating a method of clustering similar patents according to an embodiment of the present invention.

Referring to FIG. 1, the patent dispute prediction apparatus 100 according to the present embodiment includes an input data storage unit 110, a map learning unit 120, a prediction unit 130, a correction unit 140, and an output data storage unit. And 150.

The input data storage unit 110 processes input data for dispute prediction by using one or more of patent information and bibliographic information on patents for each analysis unit, and registers the processed input data in a database (not shown). To perform.

For example, when the mining model generated by the patent dispute prediction apparatus 100 according to the present embodiment is a neural network model, the input data may be a filing date, a registration date, a claim number, an independent claim number, a technical classification, a patent classification such as an IPC, and a document. It may include a vector, a patent index, patent expiry status, rights transfer information, patent holder information, and the like.

In this embodiment, technical classification, patent classification, and document vector will be referred to as technical classification information.

In addition, in this embodiment, the patent information is referred to as information that can be obtained from the patent document of the patent, and all the information issued by the respective patent offices or acquired by each company corresponding to the corresponding patent other than collectively referred to as bibliographic information Shall be.

In addition, the input data may include patent holder information, and the patent holder information may include the name of the patent holder, the number of patent holders' and / or registrations, the number of technology transfers, and the number of past disputes.

In another example, if the mining model is a linear model, such as a decision tree model, the input data may include the application stay period (i.e., the period between the filing date and the registration date), the IPC-technical classification corresponding probability value (the specific IPC of the entire IPC is specified). It may further include a separate variable representing the relationship between the variables, such as the ratio belonging to the technical classification).

In addition, the input data may be processed using patent information and bibliographic information on a single patent, and may be processed using patent information and bibliographic information on a plurality of patents according to an analysis unit to be set. That is, the analysis unit may be set to one patent or a plurality of patents.

The input data storage unit 110 will be described in more detail with reference to FIG. 2.

As shown in FIG. 2, the input data storage unit 110 includes a setting module 210, a vector calculating module 220, and a registration module 230.

The setting module 210 performs a function of setting a patent analysis unit. That is, the setting module 210 sets the analysis unit in accordance with the information input by the user. For example, the analysis unit is set to either one patent or multiple patents.

One patent refers to processing and registering input data corresponding to one patent, and multiple patents refers to processing and registering input data using a plurality of patents.

The registration module 230 may register input data corresponding to one patent or a plurality of patents according to the set analysis unit.

In the present embodiment, it is assumed that the input data is processed and registered using at least one of patent information and bibliographic information. However, the input data may be input by the user.

The vector calculation module 220 performs a function of constructing a document vector corresponding to each patent. Referring to FIG. 3, a method of configuring the document vector by the vector calculation module 220 will be described.

The vector calculation module 220 extracts a keyword set according to a predetermined method for patent information (eg, a claim) for each patent in order to construct a literature vector (step 310). Of course, the keyword set may be extracted using various patent information in addition to the claims, but in the present specification, it is assumed that the keyword set is extracted using the claims of the patent for convenience of understanding and explanation. For example, a keyword set is extracted using an association rule algorithm, and the association rule algorithm may be any one of an apriori algorithm, aprioriTID algorithm, aprioriHybrid algorithm, and DHP algorithm. Since the association rule algorithm is obvious to those skilled in the art, a separate description thereof will be omitted.

Also, the keyword set may be extracted differently according to the setting condition. Here, the setting condition may be any one of a sentence unit and a clause unit.

If set in sentence units, a keyword set is extracted based on each claim of each patent as a base unit. However, if the set condition is a clause unit, the keyword set is extracted separately from each claim according to a symbol (eg, comma, semicolon, colon, etc.) defined in the claim.

Also, the keyword set may be extracted in a word unit or in a phrase unit.

Since a method of extracting a keyword set using an association rule algorithm is obvious to those skilled in the art, a detailed description of a detailed method of extracting a keyword set will be omitted.

When the keyword set is extracted, the vector calculation module 220 calculates the TF-IDF using the extracted keyword set, and constructs a document vector corresponding to the patent using the extracted keyword set (step 320).

That is, the vector calculation module 220 calculates the frequency of occurrence of the keyword set in the patent information of the patent. This will be referred to as TF (term frequency).

Subsequently, the vector calculation module 220 calculates the frequency of each patent for which the keyword set exists, which will be referred to as a document frequency (DF). Here, the vector calculation module 220 calculates an inverse document frequency (IDF) by adjusting to generate a high value for a keyword set with a low frequency and adjusting a low value for a keyword set with a high frequency.

As such, when the IDF is calculated, the vector calculation module 220 calculates the weight of each keyword set by multiplying the TF by the IDF.

Subsequently, the vector calculation module 220 may configure the weight set as a document vector.

Here, the vector calculation module 220 reduces the overall processing speed of the system when the keyword set becomes too large, and since the document vector is distorted, the keyword set is grouped using the similarly registered similar information to prevent this. Clustering (step 330).

For example, the vector calculation module 220 may group keyword sets using similar word information and construct a literature vector using keyword sets having an IDF greater than or equal to a threshold.

In addition, the vector calculation module 220 may cluster the configured document vector using a predetermined algorithm. For example, the vector calculation module 220 may cluster the document vector constructed using any one algorithm of K-mean, K-medoids, cliqeue, Proclus, and pcluster. Since these algorithms are obvious to those skilled in the art, a separate description thereof will be omitted.

The registration module 230 processes the input data using the patent information and the bibliographic information of the patent corresponding to the set analysis unit, and registers the input data in the database.

The registration module 230 may process all the input data except for the literature vector and register it in the database.

For example, when the analysis unit is one patent, the registration module 230 may generate input data by using patent information and bibliographic information on the patent and register it in a database.

However, when the analysis unit is a plurality of patents, the registration module 230 may extract similar patents corresponding to the set patents, and register them in a clustered database. Hereinafter, this will be described in more detail with reference to FIG. 4.

In operation 410, the registration module 230 extracts a similar patent corresponding to the set patent from the database. For example, the registration module 230 may provide a reference value (for example, similarity between a family patent of a set patent, a patent of the same patent owner (or inventor), a patent or a literature vector that is directly or indirectly cited or cited according to the set patent). , 0.9) or more patents can be extracted as similar patents.

In step 420, registration module 230 clusters similar patents.

For example, the registration module 230 extracts input data corresponding to the extracted similar patent, respectively. And, the registration module 230 represents each input data extracted corresponding to the similar patent.

For example, the registration module 230 may represent each extracted input data using one or more of a minimum value, an average value, a maximum value, and a sum. Of course, in addition to the median it can be represented using a variety of methods.

In operation 430, the registration module 230 compares the clustered patents and removes duplicates. That is, the registration module 230 determines that the clustered patents are duplicated if they are completely identical and removes them.

Again, referring to FIG. 1, the input data storage unit 110 may process input data by using one or more of patent information and bibliographic information of patents for each analysis condition and register the input data in a database.

In addition, the input data storage unit 110 according to the present embodiment may normalize the generated input data according to a predetermined method and register it in a database. For example, the input data can be normalized to a value between 0 and 1 and registered in the database.

The supervised learning unit 120 performs a function of learning the neural network by inputting input data corresponding to a registered patent among input data registered by the input data storage unit 110 to a formed neural network. Since the operation of the neural network is obvious to those skilled in the art, a separate description thereof will be omitted. In the present embodiment, it is assumed that the map learning unit 120 uses a neural network as a mining model. However, the map learning unit 120 may also use a predecessor model such as a decision tree model.

That is, the map learning unit 120 may include a mining model (eg, a neural network) such that input data registered in the input data storage unit 110 is classified according to output data registered in the output data storage unit 150 described below. ) And create a function to train it.

The supervising learning unit 120 inputs input data corresponding to the registered patent to the formed neural network to change the weight of each connection information of the neural network by using the dispute prediction information outputted and the dispute information preset according to the registered patent. Train the neural network.

In addition, the map learning unit 120 may train the neural network using input data registered in the storage unit 110 in the input data so as to correspond to output data including classification information regarding the possibility of dispute. Here, the classification information may be one or more of past disputes, warning letter issued / sent information, standard patents and promising patents.

In addition, the map learning unit 120 may train the neural network using at least two technical classification information from the input data. As described above, the technical classification information is a technical classification, a patent classification, and a literature vector, and the technical classification information is input to the neural network in at least two combinations. That is, technology classification and patent classification, patent classification and document vector, document vector, and technology classification are simultaneously input to the neural network to be learned.

After the neural network learning is completed by the map learning unit 120, the prediction unit 130 obtains dispute prediction information by inputting input data corresponding to the predicted patent to the learned neural network.

Here, the prediction unit 130 obtains dispute prediction information by inputting a plurality of classification data into a neural network learned together among the classification data among the input data corresponding to the patent to be predicted.

The correction unit 140 corrects the dispute prediction information obtained by the prediction unit 130 according to a predetermined method. For example, the correction unit 140 calculates the accuracy of the neural network using the misclassification matrix, and corrects the calculated accuracy by multiplying the calculated accuracy by the dispute prediction information obtained by the prediction unit 130. In this embodiment, since the misclassification matrix is obvious to those skilled in the art, a separate description of a method for calculating the accuracy of the neural network using the same will be omitted.

The output data storage unit 150 stores the output data for the patent for each classification unit. The output data includes classification information regarding the likelihood of dispute. The classification information may include whether there is a past dispute, a past litigation, a judgment, whether a party re-examination, a warning or the like.

5 is a flowchart illustrating a method of predicting a patent dispute by a patent dispute prediction apparatus according to an embodiment of the present invention. Each step described below is performed by each internal component of the patent dispute prediction apparatus, but will be collectively described as a patent dispute prediction apparatus for the convenience of understanding and explanation.

In operation 510, the patent dispute prediction apparatus 100 processes input data for dispute prediction using one or more of patent information and bibliographic information on a patent for each analysis condition and registers it in a database. Since this is the same as described above, a separate description thereof will be omitted.

In operation 515, the patent dispute prediction apparatus 100 trains a neural network using input data of a registered patent.

As described above, the neural network can be trained by changing the weights of the connection information of the neural network using the dispute prediction information and the predetermined dispute information outputted by inputting input data for the registered patent into the neural network.

In operation 520, the patent dispute prediction apparatus 100 obtains dispute prediction information of the prediction target patent by inputting input data of the prediction target patent to the learned neural network.

In operation 525, the patent dispute prediction apparatus 100 corrects and outputs the dispute prediction information obtained by using the accuracy of the neural network.

As described above, the accuracy of the neural network can be calculated using the misclassification matrix, and the calculated accuracy can be set as the accuracy of the neural network. Accordingly, the patent dispute prediction apparatus 100 may correct the dispute prediction information by multiplying the set accuracy by the dispute prediction information obtained for the patent to be predicted.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the invention as defined in the appended claims. It will be understood that the invention may be varied and varied without departing from the scope of the invention.

1 is a block diagram schematically illustrating an internal configuration of a patent dispute prediction apparatus according to an embodiment of the present invention.

2 is a diagram schematically showing an internal configuration of a data registration unit according to an embodiment of the present invention.

3 is a flow chart illustrating a method of constructing a document vector according to an embodiment of the present invention.

4 is a flow chart illustrating a method for clustering similar patents in accordance with an embodiment of the present invention.

5 is a flowchart illustrating a method for predicting a patent dispute by a patent dispute prediction apparatus according to an embodiment of the present invention.

Claims (17)

An input data storage unit for storing input data including patent holder information, technology classification information, or a combination thereof for each analysis unit; An output data storage unit for dividing and storing output data including classification information on a possibility of dispute for each analysis unit; And Patent dispute prediction device including a map learning unit for generating a mining model so that the input data is classified according to the output data. According to claim 1, The classification information is a patent dispute prediction device, characterized in that the presence of a past dispute. According to claim 1, The mining model is a patent dispute prediction device, characterized in that the neural network model. According to claim 1, The technical classification information is a patent dispute prediction device, characterized in that it comprises a technical classification, patent classification and literature vector. 5. The method of claim 4, The document vector is a patent dispute prediction device, characterized in that the size of the TF (term frequency) -IDF (inverse document frequency) value of each keyword set extracted from the claim for the patent for each analysis unit. 6. The method of claim 5, The keyword set is a patent dispute prediction device characterized in that the dimension of the representative similar word of the keyword set extracted from the claim for the patent for each analysis unit. According to claim 1, The analysis unit is a patent dispute prediction device, characterized in that one patent or a plurality of patents. 8. The method of claim 7, When the analysis unit is a plurality of patents, the analysis unit is a patent owned by the same patent owner, any one of a family patent, a patent owned by the same patent owner in a citation relationship, a patent owned by the same patent owner whose similarity between document vectors is equal to or greater than a reference value, or It is a combination of these, The patent dispute prediction apparatus characterized by the above-mentioned. The method of claim 8, When the analysis unit has a plurality of patents, the patents owned by the same patent holder are clustered, and a representative value of input data for each of the clustered patents is calculated to form input data corresponding to the analysis unit. Patent dispute prediction device. The method of claim 9, And the representative value is calculated using any one of an average value, a maximum value, a minimum value, and a sum. According to claim 1, And the input data includes at least two technical classifications. In the method of predicting a dispute over a patent, a patent dispute prediction device, (a) generating input data of patents of analysis units for dispute prediction using at least one of patent information and bibliographic information of patents of analysis units and registering them in a database; (b) inputting the input data corresponding to the registered patent among the input data into the neural network for predicting the dispute of the patent, and using the dispute prediction information outputted and the dispute information preset according to the registered patent, Learning by changing weights; And (c) obtaining dispute prediction information about the predicted patent by inputting input data corresponding to the predicted patent to the learned neural network. 13. The method of claim 12, In step (a), Extracting a keyword set from patent information of the patent using an association rule algorithm; Calculating a terminal frequency (TF) and an inverse document frequency (IDF) using the extracted keyword set; And And constructing a literature vector corresponding to the patent using the calculated TF and IDF. 13. The method of claim 12, The analysis unit is a patent dispute prediction method, characterized in that one patent or multi-patent. The method of claim 14, When the analysis unit is a multi-patent, step (a), Clustering similar patents owned by the same patent holder to the patent; And Calculating a representative value for the clustered similar patents. The method of claim 15, Clustering the similar patents, A patent dispute prediction method comprising extracting, as a similar patent, at least one of a family patent, a patent that is directly or indirectly cited or cited according to a set patent, or a similarity of a reference vector or more. The method of claim 9, The input data includes a plurality of technical classifications, including literature vectors corresponding to the patent, Step (b) and step (c) is a patent dispute prediction method, characterized in that for inputting at least two or more technical classifications of the plurality of technical classifications to the neural network.

Images