Patent tree and construction method thereof

Abstract

The invention relates to the technical field of patent data management, in particular to a patent tree and a construction method thereof, and the method comprises the following steps: analyzing technical features in a patent text based on patent text data, and classifying and integrating the technical features of each patent to obtain a patent technical feature set. According to the method, accurate positioning of patent technical elements is achieved through intelligent information extraction and semantic analysis, deep semantic analysis of verb and noun phrases is combined, technical targets and implementation modes of patents are extracted in detail, functions of the technical targets and the implementation modes are decomposed and integrated, and patent core content is presented in a clear and structured mode. The technical features are further segmented and integrated according to time nodes and application fields, and historical dependency features are identified in combination with fluctuation feature analysis of a time sequence, so that the development state of the patent technology can be effectively predicted. And accurate grouping and cross-domain feature combination analysis of patents can be realized through feature matching and similarity calculation.

Description

Technical Field

[0001] The present invention relates to the technical field of patent data management, and in particular to a patent tree and a construction method thereof. Background Art

[0002] Patent data management technology uses systematic data management tools and methods to efficiently organize, store, retrieve and analyze patent data to support the strategic management of intellectual property rights. Its technology usually covers data mining, patent classification, trend analysis, patent portfolio management, etc., helping enterprises and research institutions to tap the value of patent data, achieve in-depth insights into technical intelligence and optimize decision-making. Patent data management technology plays an important role in intellectual property protection, R&D direction planning, technology early warning, competition analysis, etc.

[0003] Among them, the patent tree construction method is a tool that focuses on systematic classification and display of corporate patent assets. It uses a tree structure to hierarchically classify patents by technical themes, which can help companies quickly identify technology research and development priorities and patent protection gaps. This method is used for intellectual property management, supporting companies in the effective management of patent assets, strategic planning of technical routes, and layout optimization for high-quality patent development. Through the construction of a patent tree, companies can obtain a clear perspective on technological development, and then rationally allocate R&D resources and improve patent layout.

[0004] It is difficult for existing technologies to effectively reflect the technological evolution process in patent management, and at the same time, they show obvious lag in information changes in terms of patent citation growth, technology expansion rate, etc. It is also difficult for existing technologies to quickly identify key technology nodes in the changes of patent life cycle stages, which makes it difficult for enterprises to give priority to patent resources with high development potential in technology planning, and the resource allocation efficiency is low. In addition, it is also difficult to systematically grasp the inherent hierarchical relationship of the patent structure in the display of existing patent models, which increases the difficulty of technology route sorting and strategic planning, and is not conducive to the refined operation of enterprises in patent management and research and development layout. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a patent tree and a construction method thereof.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: a patent tree and a construction method thereof, comprising the following steps: S1: Based on patent text data, analyze the technical features in the patent text, classify and integrate the technical features of each patent, and obtain the patent technical feature set; S2: Based on the patent technology feature set, the patent data is sorted and integrated by time nodes, and the historical dependency characteristics are captured in combination with the data fluctuations at the time nodes to obtain the patent technology status prediction results; S3: Based on the patent technology status prediction results, the patents are classified according to the life cycle stages, and according to the classification results, the path levels of each type of patent are divided to generate hierarchical patent path information; S4: Extracting recurring technical terms and functional descriptions based on the patent technical feature data and hierarchical patent path information in the patent technical feature set, determining potential feature item combinations, and obtaining cross-domain feature combination data; S5: According to the cross-domain feature combination data, the technical features of the patents are converted into vector form, the patents with similarity matching are classified into the same group, the importance and co-occurrence frequency of the technical features are analyzed and the hierarchical nodes are divided to obtain a patent structure layout list; S6: Based on the patent structure layout list, weights are assigned to the patent nodes in the path, and the patent tree layout is optimized and adjusted to generate a patent tree structure with optimized sorting.

[0007] As a further solution of the present invention, the patent technical feature set includes a set of patent technical terms, a functional description structure and an application scenario classification; the patent technical status prediction result includes a change curve of patent citations, an expansion speed index and a change rate of citations; the hierarchical patent path information includes the position of patents in each stage in the path, the stage division of patent growth and application; the cross-domain feature combination data includes repeated technical terms in each field and their co-occurrence frequency and correlation scores; the patent structure layout list includes a hierarchical arrangement of key technical features in a patent group and a superior-subordinate node relationship of patents in the same group; the sorted optimized patent tree structure includes main path sorting, node priority and hierarchical distribution information.

[0008] As a further solution of the present invention, based on patent text data, the technical features in the patent text are analyzed, the technical features of each patent are classified and integrated, and the specific steps of obtaining the patent technical feature set are as follows: S101: Based on the patent text data, extract the patent technical terms, functional descriptions, and key information of application scenarios, retrieve and match the corresponding technical terms, identify the functional description sentences in the text, screen the key technical terms, annotate the matching functional information, and generate the initial patent information set; S102: Based on the initial patent information set, semantic analysis of verbs and noun phrases is performed, where verbs are about the content of technical objectives and nouns are about the content of implementation methods, and technical objective markers and functional implementation contents are integrated to obtain a core content set of the patent; S103: Based on the core content set of the patent, the integrated technical objectives and implementation method data are divided into independent functional parts, and classified according to technical functional fields and application scenarios to obtain a patent technical feature set.

[0009] As a further solution of the present invention, based on the patent technology feature set, the patent data is sorted and integrated by time nodes, combined with the data fluctuations at the time nodes, the historical dependency features are captured, and the specific steps for obtaining the patent technology status prediction results are as follows: S201: Based on the patent technical feature set, extract the patent citation count, patent family size, and citation count data, segment the data by time dimension, and summarize the data by year and quarter to obtain a segmented patent data set; S202: Based on the segmented patent data set, sort and integrate each data item by time node, extract time change features, mark the time node of each data change by constructing a sequence, and generate a patent time series set; S203: Based on the patent time series set, data fluctuation feature analysis is performed to extract patent citation growth pattern, patent family expansion rate, and time-varying features of the number of citations. By identifying fluctuations in time-varying trends, historical dependency features are captured to obtain patent technology status prediction results.

[0010] As a further solution of the present invention, based on the patent technology status prediction results, the patents are classified according to the life cycle stage, and according to the classification results, the path level of each type of patent is divided. The specific steps of generating hierarchical patent path information are as follows: S301: Based on the patent technology status prediction results, the patent life cycle classification standard is generated by mapping the patent growth rate, citation frequency change and patent family expansion trend data to the life cycle stage according to the life cycle stage; S302: Based on the patent life cycle classification standard, the patents are classified into inconsistent categories according to the life cycle stages corresponding to their data, and the patents that meet the life cycle characteristics are grouped to obtain patent life cycle classification results; S303: Based on the patent life cycle classification results, a path hierarchy is constructed according to the grouping results of each category, and the life cycle stages are mapped to hierarchical path nodes to generate hierarchical patent path information.

[0011] As a further solution of the present invention, the specific steps of extracting repeated technical terms and functional descriptions, determining potential feature item combinations, and obtaining cross-domain feature combination data based on the patent technical feature data and hierarchical patent path information in the patent technical feature set are as follows: S401: Based on the patent technical feature set and the hierarchical patent path information, compare patent feature items in multiple technical fields, screen repeated technical terms and functional descriptions, and obtain patent feature comparison data; S402: Based on the patent feature comparison data, statistics are collected on the distribution of the selected feature item combinations in the technical field, the occurrence frequency and application scenarios of the feature combinations are recorded, and feature distribution data is generated; S403: Based on the feature distribution data, calculate the correlation strength of the feature item combination, analyze the co-occurrence frequency and the correlation score, screen the matching cross-domain feature combination, and obtain the cross-domain feature combination data.

[0012] As a further solution of the present invention, according to the cross-domain feature combination data, the technical features of the patents are converted into vector form, the patents with similarity matching are classified into the same group, the importance and co-occurrence frequency of the technical features are analyzed and the hierarchical nodes are divided, and the specific steps of obtaining the patent structure layout list are as follows: S501: Based on the cross-domain feature combination data, separate the technical feature content of the patent, convert each technical feature into a standard vector form using numerical mapping, and integrate all feature vectors for data standardization processing to generate a vectorized feature set; S502: Based on the vectorized feature set, feature matching analysis is performed between patents, the vector values ​​are numerically compared by calculating the similarity between the features, patents with high similarity are combined, and a patent similarity analysis result is generated; S503: Based on the patent similarity analysis results, feature similarity clustering is performed, hierarchical nodes are established for the matched patent features according to importance and co-occurrence frequency, upper nodes and lower nodes are divided according to the degree of technical impact, and a patent structure layout list is generated.

[0013] As a further solution of the present invention, the similarity between features is calculated by vector value, using the formula: ; Get the feature similarity between two patents ; in, The first patent was The value of the feature, The second patent is The value of a feature.

[0014] As a further solution of the present invention, based on the patent structure layout list, weights are assigned to patent nodes in the path, and the patent tree layout is optimized and adjusted to generate a patent tree structure with optimized sorting. The specific steps are as follows: S601: Based on the patent structure layout list and according to preset indicators, weights are assigned to patent nodes in each path according to the life cycle stage to which they belong, to obtain a path node weight set; S602: Based on the path node weight set, accumulate the node weights in the path, obtain the total weight value and average weight value of each path, sort the paths according to the weight values, and obtain a path weight sorting set; S603: Based on the path weight ranking set, adjust the patent tree layout, set the high-weight path as the main path, and assign the lower-weight paths to the secondary branches in order to generate a patent tree structure with optimized ranking.

[0015] A patent tree includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned patent tree and the construction method thereof when executing the computer program.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, the precise positioning of patent technical elements is achieved through intelligent information extraction and semantic analysis. Combined with the deep semantic analysis of verbs and noun phrases, the technical goals and implementation methods of the patent are carefully extracted and their functions are decomposed and integrated, so that the core content of the patent is presented in a clear and structured manner. The technical features are further segmented and integrated according to time nodes and application fields, and combined with the fluctuation characteristics of the time series, the historical dependency characteristics are identified, which can effectively predict the development status of the patent technology. Through feature matching and similarity calculation, the accurate grouping of patents and the analysis of cross-domain feature combinations can be achieved. Through the path weight sorting set, the high-weight path in the patent tree is set as the main path, which optimizes the hierarchical structure of the patent tree, enabling enterprises to clearly identify the relationship between the primary and secondary paths in the patent technology layout, thereby allocating resources more efficiently. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the workflow of the present invention; Figure 2 A detailed flow chart of step S1 of the present invention; Figure 3 A detailed flow chart of step S2 of the present invention; Figure 4 A detailed flow chart of step S3 of the present invention; Figure 5 A detailed flow chart of step S4 of the present invention; Figure 6 A detailed flow chart of step S5 of the present invention; Figure 7 The flowchart of step S6 of the present invention is refined. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0020] See also Figure 1 The present invention provides a technical solution: a patent tree and a construction method thereof, comprising the following steps: S1: Based on patent text data, analyze the technical features in the patent text, classify and integrate the technical features of each patent, and obtain the patent technical feature set; S2: Based on the patent technology feature set, the patent data is sorted and integrated by time node, combined with the data fluctuations at the time node, the historical dependency characteristics are captured, and the patent technology status prediction results are obtained; S3: Based on the prediction results of patent technology status, patents are classified according to the life cycle stage. According to the classification results, each type of patent is divided into path hierarchies to generate hierarchical patent path information; S4: Based on the patent technical feature data and hierarchical patent path information in the patent technical feature set, recurring technical terms and functional descriptions are extracted to determine potential feature item combinations and obtain cross-domain feature combination data; S5: Based on the cross-domain feature combination data, the technical features of the patents are converted into vector form, the patents with similarity matching are grouped into the same group, the importance and co-occurrence frequency of the technical features are analyzed and the hierarchical nodes are divided to obtain a patent structure layout list; S6: Based on the patent structure layout list, assign weights to the patent nodes in the path, optimize and adjust the patent tree layout, and generate a patent tree structure with optimized sorting.

[0021] The patent technical feature set includes the patent's technical terminology set, functional description structure and application scenario classification. The patent technology status prediction results include the change curve of patent citations, the expansion speed index and the change rate of citation times. The hierarchical patent path information includes the position of the patent in each stage in the path, the stage division of patent growth and application, and the cross-domain feature combination data includes the repeated technical terms in each field and their co-occurrence frequency and correlation score. The patent structure layout list includes the hierarchical arrangement of key technical features in the patent group and the upper and lower node relationships of patents in the same group. The sorted and optimized patent tree structure includes the main path sorting, node priority and hierarchical distribution information.

[0022] See also Figure 2 Based on the patent text data, the specific steps of analyzing the technical features in the patent text and classifying and integrating the technical features of each patent to obtain the patent technical feature set are as follows: S101: Based on the patent text data, extract the patent technical terms, functional descriptions, and key information of application scenarios, retrieve and match the corresponding technical terms, identify the functional description sentences in the text, screen the key technical terms, annotate the matching functional information, and generate the initial patent information set; First, natural language processing is used to segment the text content, and the segmentation results are semantically analyzed in combination with the context. Patent terms that meet the characteristics of the field are identified, frequently occurring terms are marked, and their technical relevance is determined by counting the word frequency. Then, the semantic role labeling method is used to further identify and extract functional description phrases, extract words and phrases related to technical applications and functional implementation, and eliminate technology-irrelevant or repeated words and phrases through syntactic analysis. All extracted technical terms, functional descriptions, application scenarios, etc. are classified and recorded, and an index table is established according to the characteristics to generate an initial patent information set.

[0023] S102: Based on the initial information set of the patent, semantic analysis of verbs and noun phrases is performed. Verbs are about the content of the technical goal and nouns are about the content of the implementation method. The technical goal mark and the function implementation content are integrated to obtain the core content set of the patent; First, the verb phrases in the patent are extracted and marked, and the functional verbs are associated with the technical goals. The applicability of the verbs to the application fields is analyzed. Then, the semantic role labeling method is used to extract the content representing the implementation method in the noun phrases. The keywords in these phrases are matched with the specific implementation steps, and a combination relationship between the goals and implementation methods is constructed. Each group of goals and implementation content is compared and analyzed item by item, and the content with similar application characteristics is integrated into the same category. Finally, all the combined content is classified and grouped according to the technical functional field to obtain the core content set of the patent.

[0024] S103: Based on the core content set of the patent, the integrated technical objectives and implementation method data are divided into independent functional parts, and classified according to technical functional fields and application scenarios to obtain the patent technical feature set; First, the hierarchical clustering method is used to divide different technical functions and application scenarios, and the technical goals and corresponding application scenarios are classified respectively. The correlation of the implementation methods contained in each application scenario is verified, and the combination of goals and implementation methods is mapped into the functional field table. Then, the classification is refined within each category according to the applicability of the application field, and the technical features of each application field are hierarchically allocated and organized to form independent functional units, thus obtaining a patent technical feature set.

[0025] See also Figure 3 Based on the patent technology feature set, the patent data is sorted and integrated by time nodes, combined with the data fluctuations at time nodes, and the historical dependency characteristics are captured to obtain the specific steps of patent technology status prediction results as follows: S201: Based on the patent technical feature set, extract the patent citation count, patent family size, and citation count data, segment the data by time dimension, and summarize the data by year and quarter to obtain a segmented patent data set; First, each data is sorted according to the time node of the patent, and each data is grouped by time. The data is summarized year by year to obtain the growth of the annual data. At the same time, the quarterly data is segmented and statistically analyzed to observe the quarterly changes. Then, a segmented statistical table is established using the time accumulation method. The time dimension of the data is subdivided according to the set time interval, and the citation times, patent family size, citation times and other values ​​are superimposed. The data are compared according to the specified time period, the differences between the annual and quarterly data are recorded, and a complete record of the segmented values ​​is constructed to obtain a segmented patent data set.

[0026] S202: Based on the segmented patent data set, sort and integrate each data item by time node, extract time change features, mark the time node of each data change by constructing a sequence, and generate a patent time series set; First, the citation times, patent family size and citation times at each time node are arranged in chronological order, and the changing trend of each time node is recorded. The time changes of each data are marked by constructing a time series analysis framework, and the growth and decline at different time points are recorded in turn. The change values ​​of each time node are accumulated, and the amplitude of the numerical changes is recorded in layers by year and quarter, gradually forming a continuous time series. By classifying the trend of the change rate and amplitude of each time node, significant growth or decline trends are marked to generate a patent time series set.

[0027] S203: Based on the patent time series set, data fluctuation characteristics are analyzed to extract the patent citation growth pattern, patent family expansion rate, and time-varying characteristics of the number of citations. By identifying the fluctuations of the time-varying trend, historical dependency characteristics are captured to obtain the patent technology status prediction results; The fluctuation characteristics between time nodes are captured by differential analysis. The data of citation counts, patent family size and citation counts in the time series are obtained to perform adjacent node differences. The differences are recorded in the fluctuation table to identify the change trend between time nodes. Then the sliding average method is used to obtain the average value of the differential results, and the fluctuation mean in the time series is recorded. The time nodes with significant differences are marked as fluctuation nodes, and the time periods with significant differential fluctuations are divided into high fluctuation intervals. The high fluctuation intervals are compared and analyzed with the stable intervals, and the time node characteristics of continuous high fluctuation periods are recorded. Assuming that the number of citations at the previous time node is used as the benchmark, the difference threshold is set to 10%. For example, if the number of citations in the previous quarter is 500, and the number of citations in the next quarter is 560, the increase is (560-500) / 500=0.12, that is, 12%. This increase is greater than the 10% threshold, so the time node is considered to be a significant difference. Based on the recorded continuous high fluctuation periods, the trend of technological change can be identified. For example, high fluctuations at several consecutive time nodes (rapid increase in citations and patent family size) indicate that the technology is in an expansion period, which may continue to grow in the future. If there are several stable or declining time nodes after the high fluctuation period, it may enter a stable or declining period. Therefore, by analyzing these time node characteristics, the future activity and potential life cycle stages can be inferred, thus obtaining the prediction results of the patent technology status.

[0028] See also Figure 4 Based on the prediction results of patent technology status, the patents are classified according to the life cycle stage. According to the classification results, the path level of each type of patent is divided. The specific steps for generating hierarchical patent path information are as follows: S301: Based on the prediction results of patent technology status, the patent life cycle stages are divided into different stages. By mapping the growth rate of patents, changes in citation frequency and patent family expansion trend data to the life cycle stages, a patent life cycle classification standard is generated; The patent growth rate, citation frequency change and patent family expansion trend data are matched through the life cycle mapping method. First, the annual growth rate of the patent is analyzed, and several indicator ranges such as high growth rate, stable growth and negative growth are set. For example, in the high growth stage, if the annual citation growth rate of a patent exceeds 20%, and the citation frequency continues to rise in the past three years, and the number of patent families rapidly expands from the initial 2 to more than 10, then the patent is classified as the high growth stage. In the stable growth stage, if the annual citation growth rate of a patent fluctuates between 5% and 10%, and the citation frequency remains stable or increases slightly in the past three years, and the number of patent families remains in a stable range of 5-8, then the patent is classified as the stable growth stage. In the decline stage, if the annual citation growth rate of a patent drops to a negative value (for example, -5%), the citation frequency has dropped significantly in recent years, and the expansion of the patent family stagnates and remains at a scale of less than 3, then the patent is classified as the decline stage. Patents within each indicator range are classified, and then the changes in the patent's citation frequency are correlated and matched with the life cycle stage. High-frequency citations, continuous growth and reduction are mapped to different life cycle intervals. The expansion trend is then judged based on the growth curve of patent family expansion. If the patent family expands rapidly in a short period of time, it is marked as a high expansion stage, and if the growth slows down, it is a stable expansion stage. The patent family expansion is comprehensively mapped with the data on citation frequency and growth rate to generate a patent life cycle classification standard.

[0029] S302: Based on the patent life cycle classification standard, the patents are classified into inconsistent categories according to the life cycle stages corresponding to their data, and the patents that meet the life cycle characteristics are grouped to obtain the patent life cycle classification results; Patents are divided into their respective categories according to the corresponding life cycle stages through the classification screening method. First, the characteristic values ​​of each stage in the classification standard are compared, and the growth rate, change in citation frequency and patent family expansion characteristics of the patent are confirmed one by one to see whether they meet a certain category. According to the set value of the standard and the degree of conformity of the actual patent data, patents that meet specific life cycle characteristics are classified into corresponding groups. If the growth rate and change in citation frequency of a patent meet different categories, it is preferentially classified into the most suitable stage according to the main characteristics. The patents in each life cycle are arranged in order, numbered and marked with characteristics in the table to obtain the patent life cycle classification results.

[0030] S303: Based on the patent life cycle classification results, according to the grouping results of each category, a path hierarchy is constructed, the life cycle stages are mapped to hierarchical path nodes, and hierarchical patent path information is generated; The classification results are processed into hierarchical paths through the hierarchical path construction method, and each life cycle category is mapped to a hierarchical path structure. First, the high-growth category in the life cycle classification is set as the main node of the path, and the patents in this category are set as the upper node of the path. Then, the categories with steady growth and slow expansion are set as the secondary nodes of the path, and the association hierarchy is formed one by one. The technical correlation within each category is used to construct a patent path table, and a hierarchical path structure is formed according to the life cycle stage. The path hierarchy is adjusted and arranged to obtain hierarchical patent path information.

[0031] See also Figure 5 , according to the patent technical feature data and hierarchical patent path information in the patent technical feature set, the specific steps of extracting repeated technical terms and functional descriptions, determining potential feature item combinations, and obtaining cross-domain feature combination data are as follows: S401: Based on the patent technical feature set and hierarchical patent path information, compare patent feature items in various technical fields, screen repeated technical terms and functional descriptions, and obtain patent feature comparison data; Patent feature items in various technical fields are compared through comparative analysis methods. First, the technical terms, functional descriptions, application scenarios and other features of each patent are extracted. Then, keyword matching screening tools are used for repeated screening. The technical terms and functional description items that appear repeatedly in various fields are identified and summarized, and common terms that appear multiple times are marked. For terms or descriptions that appear multiple times but have similar meanings, semantic analysis tools are used to compare word meanings to determine whether they are classified as similar terms. Then, the frequency of occurrence of all terms and functional description items is classified, common items and field-specific items are marked, and patent feature comparison data is output.

[0032] S402: Based on the patent feature comparison data, statistics are collected on the distribution of the selected feature item combinations in the technical field, the occurrence frequency and application scenarios of the feature combinations are recorded, and feature distribution data is generated; The distribution of the selected feature item combinations in the technical field is statistically analyzed through the frequency statistics method. First, the occurrence frequency of the feature combination in each field is obtained, and the high-frequency occurrence items of the feature items are marked as common items or cross-field combinations using the frequency classification method. Then, the high-frequency usage scenarios, specific applications and applicable technical fields of each feature combination are registered. For the combinations with higher frequencies, their corresponding application scenarios and technical branches are recorded one by one, and a feature distribution table is constructed. Finally, the distribution of feature items in all fields is sorted out to generate feature distribution data including application scenarios and occurrence frequencies.

[0033] S403: Based on the feature distribution data, calculate the correlation strength of the feature item combination, analyze the co-occurrence frequency and the correlation score, screen the matching cross-domain feature combination, and obtain the cross-domain feature combination data; Based on the characteristic distribution data, according to the formula Calculate the strength of association of cross-domain feature combinations. It represents the correlation strength of feature combinations, which is used to measure the strong correlation of feature combinations across fields. n is the total number of fields covered by each feature combination. It represents the frequency of occurrence of the feature combination in the i-th field, which is calculated by dividing the number of occurrences of the feature combination by the total number of patents in the field. It represents the relevance score of the feature combination in the i-th field. It is quantified based on the applicable scope of the features and the cross-field correlation score, and is scored by the breadth of application scenario coverage. Assuming that the feature combination is "image recognition and edge detection", its relevance score is scored in the following fields: Field A: Medical image analysis, score: 0.9. In medical image analysis, image recognition and edge detection are crucial. For example, they are widely used in tumor boundary recognition, so the score is high. Field B: Autonomous driving technology, score: 0.7. In autonomous driving, image recognition and edge detection are used to identify roads and obstacles, but their applications are less concentrated in specific scenarios, and the relevance is relatively low. Field C: Agricultural detection, score: 0.5. In agriculture, this feature combination is used for plant health monitoring and soil analysis, but it mainly relies on other feature recognitions, so the relevance score is low.

[0034] If the feature combination appears 50 times in a certain field record, and the total number of field records is 500, then the calculated frequency is , if the feature combination is widely used in multiple fields and has significant technical effects, the score is set to 0.8. Substitute it into the formula for calculation, if the frequency and score of this feature combination in n different fields are respectively , , , , the calculation formula is: ; If the average value of the correlation strength of similar feature combinations in the past was 0.15 and the obtained value of 0.22 is higher than the historical average, it means that the correlation strength of the feature combination in the current application has been significantly improved. The results show that the feature combination has gradually strengthened in cross-domain applications, and its potential value has increased significantly, making it suitable for further promotion to other fields to improve the coverage of the technology.

[0035] See also Figure 6 , according to the cross-domain feature combination data, the technical features of the patents are converted into vector form, the patents with similarity matching are classified into the same group, the importance and co-occurrence frequency of the technical features are analyzed and the hierarchical nodes are divided. The specific steps to obtain the patent structure layout list are as follows: S501: Based on the cross-domain feature combination data, separate the technical feature content of the patent, use numerical mapping to convert each technical feature into a standard vector form, and integrate all feature vectors for data standardization to generate a vectorized feature set; This is achieved through feature extraction and numerical mapping processes. First, extract the technical keywords in each patent, apply natural language processing (NLP) tools (such as Python's NLTK or spaCy) for text segmentation and part-of-speech tagging, identify technical vocabulary and terms in feature descriptions, remove stop words and retain high-frequency technical words. Next, use a vocabulary embedding model (such as Word2Vec or GloVe) to convert these keywords into numerical vector representations to obtain the initial vector value of each keyword. For the vectors generated by vocabulary embedding, further encoding adjustments are made to ensure that each vector value reflects the similarity between technical features. Subsequently, the vector data is adjusted to a uniform interval (such as [0,1]) through data normalization methods to keep the proportions between data consistent for subsequent similarity analysis. Finally, the vector values ​​of all technical keywords are integrated to form a complete technical feature vector of the patent, ensuring that each patent has an independent and standardized representation in the vector space, completing the vectorization of the technical features of all patents and summarizing them into a standard vectorized feature set.

[0036] S502: Based on the vectorized feature set, feature matching analysis is performed between patents. The vector values ​​are numerically compared by calculating the similarity between the features, and patents with high similarity are combined to generate patent similarity analysis results; To calculate the similarity between features, we use the following formula: ; Get the feature similarity between two patents ; in, The closer the cosine value of is to 1, the higher the similarity between patent features. The first patent was The numerical value on the item feature is obtained through feature extraction and numerical mapping. After keyword extraction with the help of feature extraction tools (such as Python's NLTK or spaCy), it can be converted into a numerical vector through a vocabulary embedding model (such as Word2Vec, GloVe). The second patent is The value of the feature is obtained in the same way as The same ensures that the corresponding dimensions between numerical vectors are consistent. and : represent the modulus length of eigenvectors A and B respectively.

[0037] Use text extraction tools (such as NLTK) to obtain core technical vocabulary from patent documents, and then use the word embedding model (Word2Vec) to convert these vocabulary into vector form. Assume that the feature vectors of patent A and patent B are and , respectively represent the relative weights between technical features.

[0038] Dot product part: ; Vector modulus: For patent A, ; For Patent B, ; Cosine similarity calculation: ; In order to determine which patent features have significant similarities, a similarity threshold is set. Assume that the threshold is set to 0.75, that is, if The two patents are considered to have similar features and common technical features or application scenarios. The threshold can be adjusted according to the specific research or technical field.

[0039] Assume that the similarity of the following patent features has been calculated, and the results are as follows: the similarity between Patent A and Patent B is 0.82, the similarity between Patent A and Patent C is 0.6, the similarity between Patent B and Patent C is 0.78, and the similarity between Patent A and Patent D is 0.85. According to the set similarity threshold of 0.75, the similarity results are grouped as follows: High Similarity Group 1: Patent A and Patent B, Patent A and Patent D. The similarity values ​​of these two pairs are 0.82 and 0.85, respectively, both higher than the threshold, indicating that the three patents are highly similar in technical features. They can be put into the same group, such as the same node in the tree structure, to highlight the close relationship between their technical contents. High Similarity Group 2: Patent B and Patent C, with a similarity of 0.78, meets the threshold standard, indicating that Patent B and Patent C also have high similarity in features and are suitable for being classified into another similar group. By comparing and grouping patent similarities, technical branches can be constructed in the patent tree. For example, a group containing patents A, B, and D can be marked on a branch of the patent tree to show their close technical connection, and their common features can be noted in the branch name so that their development trends can be more clearly viewed in subsequent technical evolution and patent layout analysis. The grouping of patents B and C is marked on another branch, which can be further expanded to include other patents with similar technical features.

[0040] S503: Based on the results of patent similarity analysis, feature similarity clustering is performed, hierarchical nodes are established for the matched patent features according to importance and co-occurrence frequency, upper nodes and lower nodes are divided according to the degree of technical impact, and a patent structure layout list is generated; First, the similarity values ​​between patents in the similarity matrix are used as initial data, and the patent technical features are clustered by the hierarchical clustering algorithm. According to the principle of high similarity priority grouping, the patents with higher similarity are gradually grouped together to form clusters. In the clustering process, the co-occurrence frequency of the feature vector is used as an important measurement indicator. The co-occurrence frequency of technical features in each cluster is recorded by the frequency statistics tool, and the high-frequency technical features are marked to reflect their core position in each cluster. After the clustering is completed, the hierarchical structure is established using the clustering partitioning method, and the feature nodes are graded according to the technical influence and importance. For technical feature nodes with high co-occurrence frequency and high values ​​in the similarity matrix, they are set as upper nodes to represent the technical core; for nodes with low co-occurrence frequency and relatively low similarity values, they are set as lower nodes to represent secondary or auxiliary technologies in the field. After the hierarchical structure is completed, a patent structure layout list containing the hierarchical results of each patent feature is generated, presenting a patent structure layout list based on the similarity of technical features.

[0041] See also Figure 7 Based on the patent structure layout list, weights are assigned to patent nodes in the path, and the patent tree layout is optimized and adjusted. The specific steps to generate a patent tree structure with optimized sorting are as follows: S601: Based on the patent structure layout list and according to preset indicators, weights are assigned to patent nodes in each path according to the life cycle stage to which they belong, to obtain a path node weight set; Through the preset life cycle stage indicators, each patent node in the path is assigned a weight value according to its life cycle stage. First, according to the life cycle stage, the nodes are divided into high growth stage, mature stage and decline stage, and corresponding weights are assigned to each stage, such as 3 for high growth stage, 2 for mature stage and 1 for decline stage. Then, each node in the path is traversed in turn, and the corresponding weight value is added to the node attribute set according to its life cycle stage to form an initial path node weight set. The weight value of each node is recorded one by one through the document processing tool, and the weight values ​​of all nodes are recorded in the path weight set, so that the weight of each path can be subsequently integrated in the path weight set. S602: Based on the path node weight set, accumulate the node weights in the path, obtain the total weight value and average weight value of each path, sort the paths according to the weight values, and obtain a path weight sorting set; By counting the node weights in the path, we first traverse the nodes in each path, call the weight values ​​stored in the path node weight set, and add the weight values ​​of all nodes to form the total weight value of each path. Then, divide the total weight value by the number of path nodes to obtain the average weight value of the path. For different paths, the total weight values ​​and average weight values ​​of all paths are recorded in sequence, and these values ​​are passed to the patent management tool for sorting. The sorted path weight set is used to construct the final path weight result to ensure that the weight sorting is reasonable and the data is clear.

[0042] S603: Based on the path weight ranking set, adjust the patent tree layout, set the high-weight path as the main path, and assign the lower-weight paths to the secondary branches in order, so as to generate a patent tree structure with optimized ranking; The patent tree layout is adjusted and optimized based on the weight sorting results. By comparing the weight values ​​in the path weight sorting set, the paths with higher weights in the sorting are assigned as main paths, and the main paths are marked at the core branch positions of the patent tree, and are displayed preferentially on the main branches of the tree structure. At the same time, the second highest weighted paths are arranged to the secondary branches in sequence according to the weights. By appropriately allocating the weight sorting sets of the secondary paths, it is ensured that the high-weighted paths are displayed preferentially in the main paths of the patent tree, and finally a patent tree structure layout that has been sorted and optimized is formed.

[0043] A patent tree includes a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of the above-mentioned patent tree and the construction method thereof are implemented.

[0044] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for constructing a patent tree, characterized in that: The following steps are involved: Based on patent text data, analyze the technical features in the patent text, classify and integrate the technical features of each patent, and obtain the patent technical feature set; Based on the patent technology feature set, the patent data is sorted and integrated by time nodes, and the historical dependency features are captured in combination with the data fluctuations at the time nodes to obtain the patent technology status prediction results; Based on the patent technology status prediction results, the patents are classified according to the life cycle stages, and according to the classification results, the path levels of each type of patent are divided to generate hierarchical patent path information; Based on the patent technical feature data and hierarchical patent path information in the patent technical feature set, recurring technical terms and functional descriptions are extracted, potential feature item combinations are determined, and cross-domain feature combination data is obtained; According to the cross-domain feature combination data, the technical features of the patents are converted into vector form, patents with similarity matching are grouped into the same group, the importance and co-occurrence frequency of the technical features are analyzed and the hierarchical nodes are divided to obtain a patent structure layout list; Based on the patent structure layout list, weights are assigned to the patent nodes in the path, and the patent tree layout is optimized and adjusted to generate a patent tree structure with optimized sorting.

2. The method for constructing a patent tree according to claim 1, characterized in that: The patent technical feature set includes a set of patent technical terms, a functional description structure and a classification of application scenarios; the patent technical status prediction results include a change curve of patent citations, an expansion speed index and a change rate of citations; the hierarchical patent path information includes the position of patents in each stage in the path, the stage division of patent growth and application; the cross-domain feature combination data includes repeated technical terms in each field and their co-occurrence frequency and correlation scores; the patent structure layout list includes the hierarchical arrangement of key technical features in the patent group and the superior and subordinate node relationships of patents in the same group; the sorted and optimized patent tree structure includes the main path sorting, node priority and hierarchical structure distribution information.

3. The method for constructing a patent tree according to claim 1, characterized in that: Based on the patent text data, the specific steps of analyzing the technical features in the patent text, classifying and integrating the technical features of each patent, and obtaining the patent technical feature set are as follows: Based on patent text data, extract patent technical terms, functional descriptions, and key information of application scenarios, retrieve and match corresponding technical terms, identify functional description sentences in the text, screen key technical terms, annotate matching functional information, and generate an initial patent information set; Based on the initial patent information set, semantic analysis of verbs and noun phrases is performed. Verbs are about the content of technical goals and nouns are about the content of implementation methods. The technical goal mark and the function implementation content are integrated to obtain the core content set of the patent. Based on the core content set of the patent, the integrated technical objectives and implementation method data are divided into independent functional parts, and classified according to technical functional fields and application scenarios to obtain the patent technical feature set.

4. The method for constructing a patent tree according to claim 1, characterized in that: Based on the patent technology feature set, the patent data is sorted and integrated by time nodes, combined with the data fluctuations at time nodes, and the historical dependency features are captured to obtain the specific steps of patent technology status prediction results as follows: Based on the patent technical feature set, extract the patent citation times, patent family size, and citation times data, segment the data by time dimension, and summarize the data by year and quarter to obtain a segmented patent data set; Based on the segmented patent data set, each data item is sorted and integrated by time node, time change characteristics are extracted, and the time node of each data change is marked by constructing a sequence to generate a patent time series set; Based on the patent time series set, data fluctuation characteristics are analyzed to extract the patent citation growth pattern, patent family expansion rate, and time-varying characteristics of the number of citations. By identifying the fluctuations of time-varying trends and capturing historical dependency characteristics, the patent technology status prediction results are obtained.

5. The method for constructing a patent tree according to claim 1, characterized in that: Based on the patent technology status prediction results, the patents are classified according to the life cycle stage. According to the classification results, the path level of each type of patent is divided. The specific steps for generating hierarchical patent path information are as follows: Based on the patent technology status prediction results, the patent life cycle classification standard is generated by mapping the patent growth rate, citation frequency change and patent family expansion trend data to the life cycle stage according to the life cycle stage; Based on the patent life cycle classification standard, patents are classified into inconsistent categories according to the life cycle stages corresponding to their data, and patents that meet the life cycle characteristics are grouped to obtain patent life cycle classification results; Based on the patent life cycle classification results, a path hierarchy is constructed according to the grouping results of each category, the life cycle stages are mapped to hierarchical path nodes, and hierarchical patent path information is generated.

6. The method for constructing a patent tree according to claim 1, characterized in that: According to the patent technical feature data and hierarchical patent path information in the patent technical feature set, the specific steps of extracting repeated technical terms and functional descriptions, determining potential feature item combinations, and obtaining cross-domain feature combination data are as follows: Based on the patent technical feature set and the hierarchical patent path information, patent feature items in various technical fields are compared, and repeated technical terms and functional descriptions are screened to obtain patent feature comparison data; Based on the patent feature comparison data, statistics are collected on the distribution of the selected feature item combinations in the technical field, the occurrence frequency and application scenarios of the feature combinations are recorded, and feature distribution data is generated; Based on the feature distribution data, the correlation strength of the feature item combination is calculated, the co-occurrence frequency and the correlation score are analyzed, and the matching cross-domain feature combination is screened to obtain the cross-domain feature combination data.

7. The method for constructing a patent tree according to claim 1, characterized in that: According to the cross-domain feature combination data, the technical features of the patents are converted into vector form, the patents with similarity matching are grouped into the same group, the importance and co-occurrence frequency of the technical features are analyzed and the hierarchical nodes are divided, and the specific steps for obtaining the patent structure layout list are as follows: Based on the cross-domain feature combination data, separate the technical feature content of the patent, use numerical mapping to convert each technical feature into a standard vector form, and integrate all feature vectors for data standardization to generate a vectorized feature set; Based on the vectorized feature set, feature matching analysis is performed between patents, the vector values ​​are numerically compared by calculating the similarity between the features, patents with high similarity are combined, and patent similarity analysis results are generated; Based on the patent similarity analysis results, feature similarity clustering is performed, hierarchical nodes are established for the matching patent features according to importance and co-occurrence frequency, upper and lower nodes are divided according to the degree of technical impact, and a patent structure layout list is generated.

8. The method for constructing a patent tree according to claim 7, characterized in that: To calculate the similarity between features, we use the following formula: ； Get the feature similarity between two patents ; in, The first patent was The value of the feature, The second patent is The value of a feature.

9. The method for constructing a patent tree according to claim 1, characterized in that: Based on the patent structure layout list, weights are assigned to the patent nodes in the path, and the patent tree layout is optimized and adjusted. The specific steps for generating a patent tree structure with optimized sorting are as follows: Based on the patent structure layout list, according to preset indicators, the patent nodes in each path are weighted according to the life cycle stage to which they belong, and a path node weight set is obtained; Based on the path node weight set, the node weights in the path are accumulated to obtain the total weight value and the average weight value of each path, and the paths are sorted by weight value to obtain a path weight sorting set; Based on the path weight ranking set, the patent tree layout is adjusted, high-weight paths are set as main paths, and lower-weight paths are assigned to secondary branches in order to generate a ranking-optimized patent tree structure.

10. A patent tree, comprising a memory and a processor, characterized in that: The memory stores a computer program, and when the processor executes the computer program, the steps of the method for constructing a patent tree according to any one of claims 1 to 9 are implemented.

Images