Intellectual property management method based on big data technology

Abstract

The invention discloses an intellectual property management method based on a big data technology, and belongs to the technical field of patent early warning, and the method specifically comprises the steps: obtaining all patent data of an associated enterprise having the same industry chain node with a target enterprise in the technical field; calculating a difference value between the target enterprise and any associated enterprise according to the patent data, and determining a competitive enterprise according to the difference value; extracting reference patent data sets of the competitive enterprise in past time T, grouping the reference patent data sets to obtain a plurality of sub-data sets, and obtaining a theme corresponding to any sub-data set and a risk score of the theme; acquiring all patent data of the target enterprise in the sub-data set corresponding to the subject, calculating by combining the patent data to obtain a correction coefficient, and correcting the risk score of the subject according to the correction coefficient; and if the corrected risk score is greater than or equal to a preset risk score threshold, carrying out subject patent early warning on the target enterprise, and realizing effective analysis and early warning on the enterprise technology development direction.

Description

Technical Field

[0001] The present invention relates to the field of patent early warning technology, and in particular to an intellectual property management method based on big data technology. Background Art

[0002] Intellectual property management covers patent analysis, intellectual property protection, technological competitive intelligence and other contents, aiming to help enterprises effectively manage their intellectual property, protect innovative achievements and improve market competitiveness.

[0003] The analysis method for technology protection strategy recommendations based on designated patent data is a patent data-based analytical strategy designed to help companies develop and optimize technology protection strategies. By conducting in-depth analysis of designated patent data, it identifies the direction and trends of technological innovation and facilitates the development of more effective intellectual property protection plans. The primary goal is to provide effective technology protection strategy recommendations to ensure companies can protect their unique technological achievements, prevent infringement, and maintain competitive advantage in a highly competitive market. This method allows companies to better understand the current patent landscape in a technology field and identify potential technological risks and opportunities. This helps companies optimize R&D directions, reduce technological risks, and strengthen the protection of core technologies, thereby improving the sustainability of innovation and protecting market share.

[0004] Existing technologies usually use patent risk assessment based on the number of patents, technical field coverage or citation relationships, or identify potential infringement risks through text similarity analysis. They do not conduct correlation analysis between the time series characteristics of patent data and the evolution of technical topics, resulting in deviations between warning results and actual needs. Summary of the Invention

[0005] The purpose of this invention is to provide an intellectual property management method based on big data technology to solve the following technical problems:

[0006] Existing technologies usually use patent risk assessment based on the number of patents, technical field coverage or citation relationships, or identify potential infringement risks through text similarity analysis. They do not conduct correlation analysis between the time series characteristics of patent data and the evolution of technical topics, resulting in deviations between warning results and actual needs.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] A method for intellectual property management based on big data technology, comprising the following steps:

[0009] S1, all patent data corresponding to any technical field of the target enterprise is acquired, and all patent data of associated enterprises of the target enterprise at the nodes of the industrial chain in the technical field is acquired, a difference value of the target enterprise and any associated enterprise is calculated according to the patent data, if the difference value is less than or equal to a preset threshold value, the associated enterprise is marked as a competitive enterprise, and the patent data includes patent text, IPC classification number and application time;

[0010] S2, any competitive enterprise is selected, all patent data of the competitive enterprise in the technical field in the past time T is extracted and marked as a reference patent data set, a feature vector of each reference patent data in the reference patent data set is acquired, the reference patent data set is divided into a plurality of sub-data sets according to the feature vector, and theme identification is performed on any sub-data set to obtain themes corresponding to all sub-data sets;

[0011] S3, any theme is selected, all patent data in the sub-data set corresponding to the theme is acquired and sorted in ascending order of application time, the time interval between any two adjacent patent data is acquired in sequence, and a time interval change curve is generated with the serial number of the patent data pair as the horizontal coordinate and the time interval as the vertical coordinate, and the risk score of the theme is calculated according to the time interval change curve;

[0012] S4, S2 is repeated to obtain all patent data of the target enterprise in the sub-data set corresponding to the theme, a correction coefficient is calculated combined with the patent data, the risk score of the theme is corrected according to the correction coefficient, and if the corrected risk score is greater than or equal to a preset risk score threshold value, the target enterprise is warned of the theme patent.

[0013] As a further scheme of the application, in S1, the specific calculation process of the difference value is:

[0014]

[0015] Wherein, n is the total number of IPC categories involved in all patent data corresponding to the technical field of the target enterprise, ci is the number of patent data corresponding to the i th IPC category involved in the associated enterprise, and ai is the number of patent data corresponding to the i th IPC category involved in the target enterprise.

[0016] As a further scheme of the application, in S2, the specific process of dividing a plurality of sub-data sets is:

[0017] S11, the patent text is extracted from any patent data, the patent text is subjected to word segmentation processing, and each word is converted into an embedding vector using a pre-trained word embedding model, the embedding vectors of all words are weighted and averaged to obtain the feature vector of the patent data;

[0018] S12, mapping all reference patent data into the LSH table using a hash algorithm to obtain several hash buckets, selecting any hash bucket and calculating the cosine distance values ​​of the eigenvectors between all reference patent data in the hash bucket, setting a standard cosine distance value R, and calculating the density P of the reference patent data within the standard cosine distance value R with any reference patent data as the center, obtaining the average density AvgP of the reference patent data within the standard cosine distance value R of all reference patent data; if there is any reference patent data whose corresponding reference patent data density P is greater than or equal to AvgP, generating a category cluster with the reference patent data as the core point, and classifying all reference patent data within the radius R into the category cluster;

[0019] Detect the non-core point reference patent data in the category cluster. If there is a non-core point reference patent data whose density P within the standard cosine distance value R is also greater than MinP, then merge the category cluster generated by the non-core point with the original category cluster to generate several category clusters.

[0020] S13, repeat the above process to obtain several category clusters corresponding to all hash buckets, and mark each category cluster as a sub-dataset.

[0021] As a further solution of the present invention, the patent text is also preprocessed. The specific process of the preprocessing is as follows:

[0022] Special characters in the patent text are filtered using regular expressions and deleted. The special characters include punctuation marks and HTML tags.

[0023] As a further solution of the present invention: In S3, the specific calculation process of the risk score is:

[0024] Get the starting point, end point, peak point and valley point of the time interval change curve and use them as feature points, calculate the slope between adjacent feature points, and get the average slope k. According to the calculation formula The risk score C is calculated, where T is the length of the past time, kq is the slope between the qth adjacent feature points, tq ​​represents the time length between the qth adjacent feature points, and m is the number of all patent data in the sub-dataset corresponding to the topic.

[0025] As a further solution of the present invention: In S4, the specific process of correcting the risk score of the subject is:

[0026]

[0027] Where J is the total amount of patent data on the subject by all enterprises in the past time T, L is the total amount of patent data on the subject by the target enterprise in the past time T, and β is the preset unit coefficient.

[0028] As a further solution of the present invention: if there are multiple competing enterprises, then according to the calculation formula The second risk score Cc is calculated and used as the risk score of the topic; where Y is the number of competing companies, Gw is the total amount of patent data on the topic of the w-th competing industry in the past time T, G is the total amount of patent data on the topic of all competing companies in the past time T, and Cw is the risk score of the w-th competing industry.

[0029] As a further solution of the present invention: the specific process of topic identification is:

[0030] Extract patent text from any patent data, extract keywords from the title and abstract of the patent text based on the graph model, and generate a keyword co-occurrence matrix. Input the keyword co-occurrence matrix into the preset LDA model to obtain the topic corresponding to the sub-dataset.

[0031] Beneficial effects of the present invention:

[0032] By comparing and calculating the difference between the patent data of the target enterprise and its affiliated enterprises, affiliated enterprises with a difference value less than or equal to a preset threshold are labeled as competitors. It can be understood that if the difference value is small, it means that the technical fields are similar, and therefore there may be technological competition. By selecting the patent data of competing enterprises and performing feature extraction and theme identification, it can be understood that the patent feature vectors are clustered, and each subset represents a technical theme. The distribution of competitors' patents on various technical themes can be understood. By sorting the patents under each theme by time, the time interval between adjacent patents is calculated to generate a curve, and then the risk score is calculated. It can be understood that the larger the risk score, the greater the change in the competitor's investment in this field, and the smaller the risk score, the more stable the change in the competitor's investment in this field. At the same time, combined with the development of the target enterprise in this field, it can be understood that the higher the proportion of the target enterprise's patents in this field, the less external dependence the target enterprise has in this field. It can be understood that the lower the proportion of the target enterprise's patents in this field, the greater the external dependence of the target enterprise in this field. By comprehensively evaluating the target enterprise's external dependence in this field and the greater changes in competitor investment, it can effectively warn of technical risks that may affect the development of the enterprise, and achieve effective analysis and warning of the direction of the enterprise's technological development. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The present invention will be further described below with reference to the accompanying drawings.

[0034] Figure 1 This is a flow chart of an intellectual property management method based on big data technology of the present invention. DETAILED DESCRIPTION

[0035] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0036] See also Figure 1 As shown, the present invention is an intellectual property management method based on big data technology, comprising the following steps:

[0037] S1: Obtain all patent data corresponding to any technical field of the target enterprise, and obtain all patent data of affiliated enterprises in the same industrial chain node as the target enterprise in the same technical field, calculate the difference value between the target enterprise and any affiliated enterprise based on the patent data, and if the difference value is less than or equal to a preset threshold, mark the affiliated enterprise as a competitor; the patent data includes patent text, IPC classification number and application time;

[0038] S2: Select any competing enterprise, extract all patent data of the competing enterprise in the technical field within the past time T and mark them as a reference patent dataset, obtain the feature vector of each reference patent data in the reference patent dataset, divide the reference patent dataset into several sub-datasets based on the feature vector, and perform topic identification on any sub-dataset to obtain the topics corresponding to all sub-datasets;

[0039] S3: Select any topic, obtain all patent data in the sub-dataset corresponding to the topic, and sort them in ascending order of application time. Obtain the time interval between any two adjacent patent data in sequence, and generate a time interval change curve with the sequence number of the patent data pair as the horizontal axis and the time interval as the vertical axis. Calculate the risk score of the topic based on the time interval change curve.

[0040] S4, repeat S2, obtain all patent data of the target enterprise in the sub-dataset corresponding to the subject, calculate the correction coefficient based on the patent data, and correct the risk score of the subject according to the correction coefficient. If the corrected risk score is greater than or equal to the preset risk score threshold, then issue a patent warning for the subject to the target enterprise.

[0041] By comparing and calculating the difference between the patent data of the target enterprise and its affiliated enterprises, affiliated enterprises with a difference value less than or equal to a preset threshold are labeled as competitors. It can be understood that if the difference value is small, it means that the technical fields are similar, and therefore there may be technological competition. By selecting the patent data of competing enterprises and performing feature extraction and theme identification, it can be understood that the patent feature vectors are clustered, and each subset represents a technical theme. The distribution of competitors' patents on various technical themes can be understood. By sorting the patents under each theme by time, the time interval between adjacent patents is calculated to generate a curve, and then the risk score is calculated. It can be understood that the larger the risk score, the greater the change in the competitor's investment in this field, and the smaller the risk score, the more stable the change in the competitor's investment in this field. At the same time, combined with the development of the target enterprise in this field, it can be understood that the higher the proportion of the target enterprise's patents in this field, the less external dependence the target enterprise has in this field. It can be understood that the lower the proportion of the target enterprise's patents in this field, the greater the external dependence of the target enterprise in this field. By comprehensively evaluating the target enterprise's external dependence in this field and the greater changes in competitor investment, it can effectively warn of technical risks that may affect the development of the enterprise, and achieve effective analysis and warning of the direction of the enterprise's technological development.

[0042] In a preferred embodiment of the present invention, in S1, the specific calculation process of the difference value is:

[0043]

[0044] Where n is the total number of IPC categories involved in all patent data corresponding to the target enterprise in the technical field, ci is the number of patent data corresponding to the i-th IPC category involved by the associated enterprise, and ai is the number of patent data corresponding to the i-th IPC category involved by the target enterprise.

[0045] In a preferred embodiment of the present invention, the specific process of dividing the plurality of sub-data sets in S2 is as follows:

[0046] S11, extracting patent text from any patent data, performing word segmentation on the patent text, and converting each word into an embedding vector using a pre-trained word embedding model. The embedding vectors of all words are weighted averaged to obtain a feature vector for the patent data;

[0047] S12, mapping all reference patent data into the LSH table using a hash algorithm to obtain several hash buckets, selecting any hash bucket and calculating the cosine distance values ​​of the eigenvectors between all reference patent data in the hash bucket, setting a standard cosine distance value R, and calculating the density P of the reference patent data within the standard cosine distance value R with any reference patent data as the center, obtaining the average density AvgP of the reference patent data within the standard cosine distance value R of all reference patent data; if there is any reference patent data whose corresponding reference patent data density P is greater than or equal to AvgP, generating a category cluster with the reference patent data as the core point, and classifying all reference patent data within the radius R into the category cluster;

[0048] Detect the non-core point reference patent data in the category cluster. If there is a non-core point reference patent data whose density P within the standard cosine distance value R is also greater than MinP, then merge the category cluster generated by the non-core point with the original category cluster to generate several category clusters.

[0049] S13, repeat the above process to obtain several category clusters corresponding to all hash buckets, and mark each category cluster as a sub-dataset.

[0050] By using the Locality Sensitive Hashing (LSH) algorithm, patents with similar feature vectors can be efficiently mapped to the same hash bucket, significantly reducing the number of patent pairs requiring cosine similarity calculations and avoiding the need to directly calculate similarity across all patent data. This significantly improves computational efficiency for large datasets. Using a density-based clustering method, patent data can be divided into clusters based on similarity and density. This method does not require a predefined number of clusters, but automatically determines the clusters based on the actual data distribution. This better adapts to the natural distribution of patent data and avoids the manual setting of the number of clusters required in traditional methods. During the clustering process, the density and cosine distance of core points help identify clusters with high density, while points with low density are considered noise or outliers. This effectively eliminates noise data and improves the accuracy of clustering results. With this feature vector-based clustering method, each cluster corresponds to a group of similar patent data, making it easier for users to understand and analyze patent data within a particular category. By identifying core points and density distribution, users can more clearly identify the characteristics of each cluster. The hashing method and density-based clustering can process larger datasets while maintaining computational efficiency.

[0051] In a preferred embodiment of the present invention, the patent text is further preprocessed. The specific process of the preprocessing is as follows:

[0052] Special characters in the patent text are filtered using regular expressions and deleted. The special characters include punctuation marks and HTML tags.

[0053] Patent texts may contain numerous punctuation marks and HTML tags, which often have no practical significance for text analysis and processing. Removing these special characters can make the text more concise and consistent, facilitating subsequent analysis and processing. For example, after removing HTML tags, the remaining text is purer, making it easier to extract content and refine information. Removing irrelevant characters (such as HTML tags and punctuation marks) can reduce text complexity and simplify subsequent natural language processing (NLP) tasks such as word segmentation, entity recognition, and text classification. By removing these interfering characters, the algorithm can focus more on the core content of the patent, improving processing efficiency and accuracy.

[0054] In a preferred embodiment of the present invention, the specific calculation process of the risk score in S3 is:

[0055] Get the starting point, end point, peak point and valley point of the time interval change curve and use them as feature points, calculate the slope between adjacent feature points, and get the average slope k. According to the calculation formula The risk score C is calculated, where T is the length of the past time, kq is the slope between the qth adjacent feature points, tq ​​represents the time length between the qth adjacent feature points, and m is the number of all patent data in the sub-dataset corresponding to the topic.

[0056] In a preferred embodiment of the present invention, the specific process of modifying the risk score of the subject in S4 is as follows:

[0057]

[0058] Where J is the total amount of patent data on the subject by all enterprises in the past time T, L is the total amount of patent data on the subject by the target enterprise in the past time T, and β is the preset unit coefficient.

[0059] It is understandable that the higher the proportion of patents held by the target enterprise in this field, the less external dependence the target enterprise has in this field. It is understandable that the lower the proportion of patents held by the target enterprise in this field, the greater the external dependence the target enterprise has in this field.

[0060] In a preferred embodiment of the present invention, if there are multiple competing enterprises, the calculation formula The second risk score Cc is calculated and used as the risk score of the topic; where Y is the number of competing companies, Gw is the total amount of patent data on the topic of the w-th competing industry in the past time T, G is the total amount of patent data on the topic of all competing companies in the past time T, and Cw is the risk score of the w-th competing industry.

[0061] By considering multiple competing companies and their patent data volumes, a more comprehensive assessment of the market competition risk facing a specific topic can be achieved. The patent activity and existing patent technologies of different competitors in a given field directly influence the risk score, helping to identify potential challenges under specific market conditions. Using the patent data volume and risk score for each competing industry allows for a more precise quantification of the risk contribution of different competitors. Calculations based on the patent data volume over a period of time T dynamically reflect changes in the competitive landscape. Over time, some companies may become more active in patent applications, while others may slow down their technological innovation. This time-based data processing approach helps identify potential market shifts and competitive risks. In the risk assessment process, comprehensively considering the patent data of multiple competitors and existing risk scores facilitates a more accurate assessment of actual risk, thereby enabling companies or decision-makers to make more precise strategic plans and decisions. For example, these risk scores can be used to adjust decisions regarding R&D investment, technological collaboration, and market entry. The resulting second risk score, Cc, helps companies identify competing industries with greater risk, allowing them to proactively prepare for technological innovation and patent development to avoid being overtaken by competitors. At the same time, if certain competing industries have lower risk scores, they may also provide companies with market entry or partnership opportunities. In terms of resource allocation, based on the calculated risk scores, companies can rationally optimize the allocation of R&D and marketing resources, prioritizing investment in higher-risk areas and reducing unnecessary waste. For example, if a competing industry has an extremely high risk score, companies may choose to adopt more defensive strategies.

[0062] Calculating the volume and risk scores of competitors' patent data on specific topics can motivate companies to strengthen their intellectual property management and protection. By understanding competitors' patent portfolios and technological capabilities, companies can proactively identify potential intellectual property risks and take appropriate protective measures.

[0063] In a preferred embodiment of the present invention, the specific process of topic identification is as follows:

[0064] Extract patent text from any patent data, extract keywords from the title and abstract of the patent text based on the graph model, and generate a keyword co-occurrence matrix. Input the keyword co-occurrence matrix into the preset LDA model to obtain the topic corresponding to the sub-dataset.

[0065] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for intellectual property management based on big data technology, characterized in that: The following steps are involved: S1: Obtain all patent data corresponding to any technical field of the target enterprise, and obtain all patent data of affiliated enterprises in the same industrial chain node as the target enterprise in the same technical field, calculate the difference value between the target enterprise and any affiliated enterprise based on the patent data, and if the difference value is less than or equal to a preset threshold, mark the affiliated enterprise as a competitor; the patent data includes patent text, IPC classification number and application time; S2: Select any competing enterprise, extract all patent data of the competing enterprise in the technical field within the past time T and mark them as a reference patent dataset, obtain the feature vector of each reference patent data in the reference patent dataset, divide the reference patent dataset into several sub-datasets based on the feature vector, and perform topic identification on any sub-dataset to obtain the topics corresponding to all sub-datasets; S3: Select any topic, obtain all patent data in the sub-dataset corresponding to the topic, and sort them in ascending order of application time. Obtain the time interval between any two adjacent patent data in sequence, and generate a time interval change curve with the sequence number of the patent data pair as the horizontal axis and the time interval as the vertical axis. Calculate the risk score of the topic based on the time interval change curve. S4, repeat S2, obtain all patent data of the target enterprise in the sub-dataset corresponding to the topic and calculate the correction coefficient, correct the risk score of the topic according to the correction coefficient, and if the corrected risk score is greater than or equal to the preset risk score threshold, issue a patent warning for the topic to the target enterprise.

2. The intellectual property management method based on big data technology according to claim 1, characterized in that: In S1, the specific calculation process of the difference value is: Where n is the total number of IPC categories involved in all patent data corresponding to the target enterprise in the technical field, ci is the number of patent data corresponding to the i-th IPC category involved by the associated enterprise, and ai is the number of patent data corresponding to the i-th IPC category involved by the target enterprise.

3. The intellectual property management method based on big data technology according to claim 1, characterized in that: In S2, the specific process of dividing the data into several sub-data sets is as follows: S11, extracting patent text from any patent data, performing word segmentation on the patent text, and converting each word into an embedding vector using a pre-trained word embedding model. The embedding vectors of all words are weighted averaged to obtain a feature vector for the patent data; S12, mapping all reference patent data into the LSH table using a hash algorithm to obtain several hash buckets, selecting any hash bucket and calculating the cosine distance values ​​of the eigenvectors between all reference patent data in the hash bucket, setting a standard cosine distance value R, and calculating the density P of the reference patent data within the standard cosine distance value R with any reference patent data as the center, obtaining the average density AvgP of the reference patent data within the standard cosine distance value R of all reference patent data; if there is any reference patent data whose corresponding reference patent data density P is greater than or equal to AvgP, generating a category cluster with the reference patent data as the core point, and classifying all reference patent data within the radius R into the category cluster; Detect the non-core point reference patent data in the category cluster. If there is a non-core point reference patent data whose density P within the standard cosine distance value R is also greater than MinP, then merge the category cluster generated by the non-core point with the original category cluster to generate several category clusters. S13, repeat the above process to obtain several category clusters corresponding to all hash buckets, and mark each category cluster as a sub-dataset.

4. The intellectual property management method based on big data technology according to claim 3 is characterized in that: It also includes preprocessing of patent texts. The specific process of preprocessing is as follows: Special characters in the patent text are filtered using regular expressions and deleted. The special characters include punctuation marks and HTML tags.

5. The intellectual property management method based on big data technology according to claim 1, characterized in that: In S3, the specific calculation process of the risk score is: Get the starting point, end point, peak point and valley point of the time interval change curve and use them as feature points, calculate the slope between adjacent feature points, and get the average slope k. According to the calculation formula The risk score C is calculated, where T is the past time, kq is the slope between the qth adjacent feature points, tq ​​represents the time length between the qth adjacent feature points, and m is the number of all patent data in the sub-dataset corresponding to the topic.

6. The intellectual property management method based on big data technology according to claim 1, characterized in that: In S4, the specific process of correcting the risk score of the subject is as follows: Where J is the total amount of patent data on the subject by all enterprises in the past time T, L is the total amount of patent data on the subject by the target enterprise in the past time T, and β is the preset unit coefficient.

7. The intellectual property management method based on big data technology according to claim 5, characterized in that: It also includes if there are multiple competing companies, according to the calculation formula The second risk score Cc is calculated and used as the risk score of the topic; where Y is the number of competing companies, Gw is the total amount of patent data on the topic of the w-th competing industry in the past time T, G is the total amount of patent data on the topic of all competing companies in the past time T, and Cw is the risk score of the w-th competing industry.

8. The intellectual property management method based on big data technology according to claim 1, characterized in that: The specific process of topic identification is: Extract patent text from any patent data, extract keywords from the title and abstract of the patent text based on the graph model, and generate a keyword co-occurrence matrix. Input the keyword co-occurrence matrix into the preset LDA model to obtain the topic corresponding to the sub-dataset.

Images