Patent analysis method, system, device and storage medium based on artificial intelligence

Abstract

The present application relates to a patent analysis method, system, device and storage medium based on artificial intelligence. The method inputs patent text into a patent feature extraction model to obtain technical fields, technical features and technical problems; obtains the number of patent applications corresponding to the technical field based on a first scoring model, builds and determines the life cycle based on a quantity curve, and outputs a first scoring value; obtains the number of related patents of associated binary groups generated by technical features and technical problems based on a second scoring model, and outputs a second scoring value; the second scoring model obtains patent comparative documents based on associated triples generated by the technical field, technical features and technical problems; the third scoring model performs stability analysis on the patent text based on the patent comparative document text, and outputs a third scoring value; performs weighted fusion processing on the obtained scoring values, and outputs the analysis results of the patent to be analyzed; compared with the prior art, the technical solution of the present application can improve the accuracy and efficiency of patent analysis.

Description

Technical Field

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a patent analysis method, system, device and storage medium based on artificial intelligence. Background Art

[0002] With the increasing awareness of intellectual property protection, patent analysis has become an important tool for enterprises, scientific research institutions, etc. to conduct technological research and market layout; however, traditional patent analysis methods rely on manual reading and understanding of patent documents, which is inefficient and easily affected by subjective factors, resulting in low accuracy of analysis results. Summary of the invention

[0003] This application provides a patent analysis method, system, device and storage medium based on artificial intelligence to improve the accuracy and efficiency of patent analysis.

[0004] In a first aspect, the present application provides a patent analysis method based on artificial intelligence, including: obtaining a patent text of a patent to be analyzed, inputting the patent text into a pre-trained patent feature extraction model, so that the patent feature extraction model outputs patent feature data; wherein the patent feature data includes a technical field, technical features and technical problems; inputting the technical field into a pre-trained first scoring model, so that the first scoring model obtains the number of patent applications corresponding to the technical field from a patent database, and based on the number of patent applications, constructs a quantity curve, and based on the quantity curve, determines the life cycle of the patent to be analyzed, and outputs a first scoring value based on the life cycle; associates the technical features and the technical problems to obtain an associated binary group, and inputs the associated binary group into a pre-trained second scoring model, so that the second The scoring model obtains the number of related patents corresponding to the associated binary group from the patent database, and outputs a second scoring value based on the number of related patents; performs association processing on the technical field, the technical features and the technical problems to obtain associated triples, and inputs the associated triples into the pre-trained patent retrieval model so that the patent retrieval model obtains patent comparative documents from the patent database; inputs the patent text and the patent comparative document text corresponding to the patent comparative document into the third scoring model, so that the third scoring model performs stability analysis on the patent text based on the patent comparative document text, and outputs a third scoring value based on the stability analysis result; inputs the first scoring value, the second scoring value and the third scoring value into the pre-trained weighted fusion model for weighted fusion processing, and outputs the analysis result of the patent to be analyzed.

[0005] In a possible implementation, before obtaining the patent text of the patent to be analyzed, it also includes: constructing a fusion model, wherein the fusion model includes the patent feature extraction model, the first scoring model, the second scoring model, the third scoring model, the patent retrieval model and the weighted fusion model; wherein the output end of the patent feature extraction model is respectively connected to the input end of the first scoring model, the second scoring model and the patent retrieval model, and the output end of the patent retrieval model is connected to the input end of the third scoring model; the output ends of the first scoring model, the second scoring model and the third scoring model are connected to the input end of the weighted fusion model.

[0006] In a possible implementation, the patent text is input into a pre-trained patent feature extraction model so that the patent feature extraction model outputs patent feature data; wherein the patent feature data includes technical fields, technical features and technical problems, specifically including: the patent feature extraction model includes a text segmentation sub-model, a first feature extraction sub-model, a second feature extraction sub-model and a third feature extraction sub-model, the output end of the text segmentation sub-model is respectively connected to the input end of the first feature extraction sub-model, the second feature extraction sub-model and the third feature extraction sub-model; the patent text is input into the text segmentation sub-model so that the text segmentation sub-model performs segmentation processing on the patent text to obtain claim text data, technical field text data and technical problem text data; the claim text data Input into the first feature extraction sub-model, so that the first feature extraction sub-model performs sentence processing on the claim text data to obtain multiple text data, and respectively performs keyword extraction on the multiple text data to obtain multiple keywords, and uses the multiple keywords as output technical features; input the technical field text data into the second feature extraction sub-model, so that the second feature extraction sub-model performs target phrase positioning processing on the technical field text data, obtains and determines the technical field position based on the first target phrase position and the second target phrase position, and extracts and outputs the technical field based on the technical field position; input the technical problem text data into the third feature extraction sub-model, so that the third feature extraction model performs intent analysis processing on the technical problem text data, obtains and outputs the technical problem.

[0007] In a possible implementation, the obtaining of the number of patent applications corresponding to the technical field from the patent database and constructing a quantity curve based on the number of patent applications specifically includes: based on the technical field, traversing the fields corresponding to all patents in the patent database to obtain all first target patents corresponding to the technical field, and classifying all first target patents based on the patent application times corresponding to all first target patents to obtain the number of first patent applications corresponding to different years, and constructing a first quantity curve with the first number of patent applications as the ordinate and the year as the abscissa; obtaining similar technical fields corresponding to the technical field, traversing the fields corresponding to all patents in the patent database based on the similar technical fields to obtain all second target patents corresponding to the similar technical fields, and classifying all second target patents based on the patent application times corresponding to all second target patents to obtain the number of second patent applications corresponding to different years, and constructing a similar quantity curve with the second number of patent applications as the ordinate and the year as the abscissa; and correcting the first quantity curve based on the similar quantity curve to obtain a quantity curve.

[0008] In a possible implementation, obtaining the number of related patents corresponding to the associated binary group from the patent database specifically includes: obtaining the first technical feature and the first technical problem corresponding to each patent in the patent database, and performing association processing on the first technical feature and the first technical problem to obtain the first associated binary group corresponding to each patent; performing vectorization processing on the associated binary group to obtain a first associated vector, and performing vectorization processing on each first associated binary group respectively to obtain a corresponding second associated vector; performing dimensionality reduction processing on the first associated vector and each second associated vector respectively to obtain the first two-dimensional coordinates of the first associated vector in a preset vector space, and the second two-dimensional coordinates of each second associated vector in the preset vector space; calculating the coordinate distance between the first two-dimensional coordinates and the second two-dimensional coordinates, and based on the coordinate distance, screening all second associated vectors to obtain a target second associated vector set, counting the number of vectors in the second associated vector set, and using the number of vectors as the number of related patents corresponding to the associated binary group.

[0009] In a possible implementation, the inputting of the associated triples into a pre-trained patent retrieval model so that the patent retrieval model obtains patent comparative documents from the patent database specifically includes: obtaining the first technical field, the first technical feature and the first technical problem corresponding to each patent in the patent database, and performing association processing on the first technical field, the first technical feature and the first technical problem to obtain the first associated triples corresponding to each patent; converting the associated triples into first binary data, and converting each first associated triplet into its corresponding second binary data, respectively calculating the similarity between the first binary data and each second binary data, and obtaining the patent corresponding to the highest similarity value as the patent comparative document.

[0010] In a possible implementation, the training process of the third scoring model specifically includes: obtaining a plurality of historical patent invalidation case data, wherein each historical invalidation case data includes an invalidation case and an invalidation comparative document; performing feature annotation on the invalidation case to obtain a first invalidation feature, and performing feature annotation on the invalidation comparative document to obtain a second invalidation feature, respectively performing association processing on the first invalidation feature and the second invalidation feature to obtain an associated invalidation feature, and setting an invalidation reason, an invalidation legal provision, and a stability score for the associated invalidation feature; taking the invalidation case, the invalidation comparative document, the associated invalidation feature, and the invalidation reason, the invalidation legal provision, and the stability score corresponding to each historical invalidation case data as a training sample to generate a training sample set; taking the invalidation case sample and the invalidation comparative document sample corresponding to each training sample in the training sample set as inputs of the initial third scoring model, and taking the associated invalidation feature sample corresponding to each training sample in the training sample set, and the invalidation reason sample, the invalidation legal provision sample, and the stability score sample corresponding to the associated invalidation feature sample as outputs of the initial third scoring model, and performing model training on the initial third scoring model to obtain a trained third scoring model.

[0011] In the second aspect, the present application provides a patent analysis system based on artificial intelligence, including: a patent feature extraction module, a first scoring module, a second scoring module, a patent search module, a third scoring module and a patent analysis result generation module; wherein the patent feature extraction module is used to obtain the patent text of the patent to be analyzed, and input the patent text into a pre-trained patent feature extraction model so that the patent feature extraction model outputs patent feature data; wherein the patent feature data includes technical fields, technical features and technical problems; the first scoring module is used to input the technical field into a pre-trained first scoring model so that the first scoring model obtains the number of patent applications corresponding to the technical field from the patent database, and based on the number of patent applications, constructs a quantity curve, and based on the quantity curve, determines the life cycle of the patent to be analyzed, and outputs a first scoring value based on the life cycle; the second scoring module is used to associate the technical features and the technical problems to obtain associated tuples, and associate the associated tuples The binary group is input into the pre-trained second scoring model, so that the second scoring model obtains the number of related patents corresponding to the associated binary group from the patent database, and outputs a second scoring value based on the number of related patents; the patent search module is used to associate the technical field, the technical features and the technical problem to obtain an associated triplet, and the associated triplet is input into the pre-trained patent search model, so that the patent search model obtains patent reference documents from the patent database; the third scoring module is used to input the patent text and the patent reference document text corresponding to the patent reference document into the third scoring model, so that the third scoring model performs stability analysis on the patent text based on the patent reference document text, and outputs a third scoring value based on the stability analysis result; the patent analysis result generation module is used to input the first scoring value, the second scoring value and the third scoring value into the pre-trained weighted fusion model for weighted fusion processing, and output the analysis result of the patent to be analyzed.

[0012] In the second aspect, the present application provides a patent analysis system based on artificial intelligence, which also includes: a fusion model construction module; wherein the fusion model construction module is used to construct a fusion model, wherein the fusion model includes the patent feature extraction model, the first scoring model, the second scoring model, the third scoring model, the patent retrieval model and a weighted fusion model; wherein the output end of the patent feature extraction model is respectively connected to the input ends of the first scoring model, the second scoring model and the patent retrieval model, and the output end of the patent retrieval model is connected to the input end of the third scoring model; the output ends of the first scoring model, the second scoring model and the third scoring model are connected to the input end of the weighted fusion model.

[0013] In a possible implementation, the patent feature extraction module is used to input the patent text into a pre-trained patent feature extraction model so that the patent feature extraction model outputs patent feature data; wherein the patent feature data includes technical fields, technical features and technical problems, specifically including: the patent feature extraction model includes a text segmentation sub-model, a first feature extraction sub-model, a second feature extraction sub-model and a third feature extraction sub-model, the output end of the text segmentation sub-model is respectively connected to the input end of the first feature extraction sub-model, the second feature extraction sub-model and the third feature extraction sub-model; the patent text is input into the text segmentation sub-model so that the text segmentation sub-model performs segmentation processing on the patent text to obtain claim text data, technical field text data and technical problem text data ... The requirement text data is input into the first feature extraction sub-model so that the first feature extraction sub-model performs sentence processing on the claim text data to obtain multiple text data, and respectively performs keyword extraction on the multiple text data to obtain multiple keywords, and uses the multiple keywords as the output technical features; the technical field text data is input into the second feature extraction sub-model so that the second feature extraction sub-model performs target phrase positioning processing on the technical field text data to obtain and determine the technical field position based on the first target phrase position and the second target phrase position, and extracts and outputs the technical field based on the technical field position; the technical problem text data is input into the third feature extraction sub-model so that the third feature extraction model performs intent analysis processing on the technical problem text data to obtain and output the technical problem.

[0014] In a possible implementation, the first scoring module is used to obtain the number of patent applications corresponding to the technical field from the patent database, and construct a quantity curve based on the number of patent applications, specifically including: based on the technical field, traversing the fields corresponding to all patents in the patent database to obtain all first target patents corresponding to the technical field, and classifying all first target patents based on the patent application time corresponding to each of the first target patents to obtain the number of first patent applications corresponding to different years, and constructing a first quantity curve with the first number of patent applications as the ordinate and the year as the abscissa; obtaining similar technical fields corresponding to the technical field, traversing the fields corresponding to all patents in the patent database based on the similar technical fields to obtain all second target patents corresponding to the similar technical fields, and classifying all second target patents based on the patent application time corresponding to each of the second target patents to obtain the number of second patent applications corresponding to different years, and constructing a similar quantity curve with the second number of patent applications as the ordinate and the year as the abscissa; correcting the first quantity curve based on the similar quantity curve to obtain a quantity curve.

[0015] In a possible implementation, the second scoring module is used to obtain the number of related patents corresponding to the associated binary group from the patent database, specifically including: obtaining the first technical feature and the first technical problem corresponding to each patent in the patent database, and associating the first technical feature and the first technical problem to obtain the first associated binary group corresponding to each patent; vectorizing the associated binary group to obtain a first associated vector, and vectorizing each first associated binary group respectively to obtain a corresponding second associated vector; performing dimensionality reduction processing on the first associated vector and each second associated vector respectively to obtain the first two-dimensional coordinates of the first associated vector in the preset vector space, and the second two-dimensional coordinates of each second associated vector in the preset vector space; calculating the coordinate distance between the first two-dimensional coordinates and the second two-dimensional coordinates, based on the coordinate distance, screening all second associated vectors to obtain a target second associated vector set, counting the number of vectors in the second associated vector set, and using the number of vectors as the number of related patents corresponding to the associated binary group.

[0016] In a possible implementation, the patent search module is used to input the associated triples into a pre-trained patent search model so that the patent search model obtains patent comparative documents from the patent database, specifically including: obtaining the first technical field, the first technical feature and the first technical problem corresponding to each patent in the patent database, and performing association processing on the first technical field, the first technical feature and the first technical problem to obtain the first associated triples corresponding to each patent; converting the associated triples into first binary data, and converting each first associated triplet into its corresponding second binary data, respectively calculating the similarity between the first binary data and each second binary data, and obtaining the patent corresponding to the highest similarity value as the patent comparative document.

[0017] In a possible implementation, the training process of the third scoring model specifically includes: obtaining a plurality of historical patent invalidation case data, wherein each historical invalidation case data includes an invalidation case and an invalidation comparative document; performing feature annotation on the invalidation case to obtain a first invalidation feature, and performing feature annotation on the invalidation comparative document to obtain a second invalidation feature, respectively performing association processing on the first invalidation feature and the second invalidation feature to obtain an associated invalidation feature, and setting an invalidation reason, an invalidation legal provision, and a stability score for the associated invalidation feature; taking the invalidation case, the invalidation comparative document, the associated invalidation feature, and the invalidation reason, the invalidation legal provision, and the stability score corresponding to each historical invalidation case data as a training sample to generate a training sample set; taking the invalidation case sample and the invalidation comparative document sample corresponding to each training sample in the training sample set as inputs of the initial third scoring model, and taking the associated invalidation feature sample corresponding to each training sample in the training sample set, and the invalidation reason sample, the invalidation legal provision sample, and the stability score sample corresponding to the associated invalidation feature sample as outputs of the initial third scoring model, and performing model training on the initial third scoring model to obtain a trained third scoring model.

[0018] In the third aspect, the present application provides an electronic device comprising: at least one communication interface; at least one bus connected to the at least one communication interface; at least one processor connected to the at least one bus; and at least one memory connected to the at least one bus, wherein the processor is used to implement the artificial intelligence-based patent analysis steps described in any one of the above items of the present application when executing a program stored in the memory.

[0019] In a fourth aspect, the present application also provides a computer storage medium storing computer executable instructions, wherein the computer executable instructions are used to execute the artificial intelligence-based patent analysis method described in any one of the above items of the present application.

[0020] The above technical solution provided by the embodiment of the present application has the following advantages over the prior art: through the pre-trained patent feature extraction model, key information such as technical fields, technical features and technical issues can be extracted from the patent text, which helps to conduct a more comprehensive and in-depth analysis of the patent and ensure the comprehensiveness of the analysis results; secondly, the patent to be analyzed is scored through the first scoring model, the second scoring model and the third scoring model, the first scoring value, the second scoring value and the third scoring value, and finally, the first scoring value, the second scoring value and the third scoring value are input into the weighted fusion model for weighted fusion processing, and the analysis result of the patent to be analyzed is output; this comprehensive scoring mechanism can provide a multi-dimensional patent evaluation result to help decision makers make more comprehensive and reasonable decisions; this application realizes the automatic analysis and processing of patents through the collaborative work of multiple models and the use of artificial intelligence technology, thereby improving the efficiency and accuracy of patent analysis. Compared with traditional manual analysis, this method can reduce human errors and improve the accuracy and efficiency of patent analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0023] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplified descriptions do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and unless otherwise stated, the figures in the drawings do not constitute proportional limitations.

[0024] Figure 1 It is a flowchart of an embodiment of a patent analysis method based on artificial intelligence provided by the present application;

[0025] Figure 2 It is a structural schematic diagram of an embodiment of a patent analysis system based on artificial intelligence provided by the present application;

[0026] Figure 3It is a structural schematic diagram of another embodiment of a patent analysis system based on artificial intelligence provided by the present application;

[0027] Figure 4 It is a structural schematic diagram of an electronic device provided by this application. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0029] The disclosure below provides many different embodiments or examples to implement different structures of the present invention. In order to simplify the disclosure of the present invention, the parts and settings of specific examples are described below. Of course, they are only examples, and the purpose is not to limit the present invention. In addition, the present invention can repeat reference numbers and / or letters in different examples. This repetition is for the purpose of simplification and clarity, and does not itself indicate the relationship between the various embodiments and / or settings discussed.

[0030] Example 1, see Figure 1 , Figure 1 is a flowchart of an embodiment of a patent analysis method based on artificial intelligence provided by this application, such as Figure 1 As shown, the method includes steps 101 to 106, which are specifically as follows:

[0031] Step 101: Obtain the patent text of the patent to be analyzed, and input the patent text into a pre-trained patent feature extraction model so that the patent feature extraction model outputs patent feature data; wherein the patent feature data includes technical fields, technical features and technical issues.

[0032] In one embodiment, a fusion model is constructed, wherein the fusion model includes the patent feature extraction model, the first scoring model, the second scoring model, the third scoring model, the patent retrieval model and a weighted fusion model.

[0033] Specifically, the output end of the patent feature extraction model is connected to the input end of the first scoring model, the second scoring model and the patent retrieval model respectively, and the output end of the patent retrieval model is connected to the input end of the third scoring model; the output ends of the first scoring model, the second scoring model and the third scoring model are connected to the input end of the weighted fusion model.

[0034] Specifically, the input end of the patent feature extraction model and the input end of the third scoring model are used to receive the patent text of the patent to be analyzed.

[0035] Specifically, by building a fusion model, each model involved is connected in series so that data can automatically flow from one model to another based on the connection relationship between the models, and no human intervention is required in the process, thus ensuring the continuity and smoothness of the process. In addition, since data is directly transmitted between models, waiting time is reduced and the response speed of the system is improved. At the same time, in the fusion model, each model has its own specific tasks. This modular design also makes the system easy to maintain and upgrade.

[0036] In one embodiment, the patent feature extraction model includes a text segmentation sub-model, a first feature extraction sub-model, a second feature extraction sub-model and a third feature extraction sub-model, and the output end of the text segmentation sub-model is respectively connected to the input ends of the first feature extraction sub-model, the second feature extraction sub-model and the third feature extraction sub-model.

[0037] Specifically, the patent feature extraction model is a composite model, in which each sub-model focuses on processing a specific part of the patent text, and can extract information more professionally and accurately. Similarly, sub-models can be added or modified as needed to meet different analysis needs.

[0038] In one embodiment, the patent text is input into the text segmentation sub-model so that the text segmentation sub-model performs segmentation processing on the patent text to obtain claim text data, technical field text data and technical problem text data.

[0039] Specifically, the training process of the text segmentation sub-model includes: acquiring multiple historical patent texts to obtain a historical patent text set, segmenting each historical patent text in the historical patent text set respectively to obtain historical claim text data, historical technical field text data and historical technical problem text data corresponding to each historical patent text; integrating all historical patent texts and their corresponding historical claim text data, historical technical field text data and historical technical problem text data to generate a text segmentation sample set, and using the historical patent texts in the text segmentation sample set as the input of the initial text segmentation sub-model, and using the historical claim text data, historical technical field text data and historical technical problem text data corresponding to the historical patent text as the output of the initial text segmentation sample set, training the initial text segmentation sub-model until the model converges or the number of model training times meets the preset requirements, stopping the model training, and obtaining the text segmentation sub-model.

[0040] Preferably, in addition to setting the text segmentation sub-model to perform segmentation processing on the patent text, the text segmentation sub-model can also be set to perform segmentation processing on the patent text image.

[0041] In one embodiment, the claim text data is input into the first feature extraction sub-model so that the first feature extraction sub-model performs sentence processing on the claim text data to obtain multiple text data, and performs keyword extraction on the multiple text data respectively to obtain multiple keywords, and uses the multiple keywords as output technical features.

[0042] Specifically, the first feature extraction sub-model performs sentence processing on the claim text data, and when multiple text data are obtained, the claim text data is obtained, the end characters in the claim text data are traversed, and based on the end characters, the claim text data is segmented to obtain multiple text data, and each text data is segmented based on the jieba library in python to obtain a word set corresponding to each text data, wherein the word set corresponding to each text data is composed of one or more phrases.

[0043] Specifically, keyword extraction is performed on the multiple text data respectively, and when multiple keywords are obtained, a theme set is set, and target themes are randomly assigned to the multiple text data based on the theme set, and based on a preset prominence value calculation algorithm, the prominence value corresponding to each word in the word set corresponding to each text data is calculated respectively, and the prominence value corresponding to each word is normalized to obtain a normalized prominence value, and each word in each word set is screened based on the normalized prominence value to obtain multiple target words, and the target words are used as multiple keywords.

[0044] Specifically, the salient value calculation algorithm is as follows:

[0045]

[0046] Where s is a single phrase in each text data, t is the total number of phrases in each text data, r is the topic, P(s) is the prominence value of word s, f(r|s) is the probability distribution function of word s appearing under topic r, f(r) is the probability distribution function of topic r appearing in the topic set, R is the topic set, and a, b, and c are constants.

[0047] Preferably, the first feature extraction model can adopt a topic model, such as an LDA model, which can identify potential topics and keywords in the patent text and help understand the technical features of the patent.

[0048] In one embodiment, the technical field text data is input into the second feature extraction sub-model so that the second feature extraction sub-model performs target phrase positioning processing on the technical field text data, obtains and determines the technical field position based on the first target phrase position and the second target phrase position, and extracts and outputs the technical field based on the technical field position.

[0049] Specifically, in patent application documents, the technical field is usually expressed as: This application relates to XXX technical field. Therefore, the first target phrase is set to "involving" and the second target phrase is set to "technical field".

[0050] Specifically, the second feature extraction sub-model traverses the first target phrase and the second target phrase in the technical field text data, and after traversing to the first target phrase, performs positioning processing on the right side of the first target phrase to determine the position of the first target phrase, and after traversing to the second target phrase, performs positioning processing on the left side of the second target phrase to determine the position of the second target phrase; taking the first target phrase position as the starting position of the technical field position and the second target phrase position as the ending position of the technical field position, performs text extraction on the technical field text data to obtain target extracted text, and uses the target extracted text as the technical field.

[0051] In one embodiment, the technical problem text data is input into the third feature extraction sub-model, so that the third feature extraction model performs intent analysis on the technical problem text data to obtain and output the technical problem.

[0052] Specifically, the third feature extraction model uses natural language processing technology NLP to perform intent recognition on technical question text data to obtain the user's target intent.

[0053] Natural language processing (NLP) technology can be used to understand and process human language, including language generation, understanding, and application; in intent recognition, NLP technology is used to analyze user input (such as text or voice) to determine its intent or purpose.

[0054] Step 102: Input the technical field into a pre-trained first scoring model, so that the first scoring model obtains the number of patent applications corresponding to the technical field from a patent database, and constructs a quantity curve based on the number of patent applications, and determines the life cycle of the patent to be analyzed based on the quantity curve, and outputs a first scoring value based on the life cycle.

[0055] In one embodiment, after receiving the technical field, the first scoring model traverses the fields corresponding to all patents in the patent database based on the technical field to obtain all first target patents corresponding to the technical field, and classifies all first target patents based on the patent application time corresponding to each of the first target patents to obtain the number of first patent applications corresponding to different years, and constructs a first quantity curve with the number of first patent applications as the vertical coordinate and the year as the horizontal coordinate.

[0056] Specifically, based on the technical field, when traversing the fields corresponding to all the patents in the patent database, the fields corresponding to all the patents in the patent database are obtained, and it is determined whether the technical field is the same as the field corresponding to each patent in the patent database. If they are the same, the patent corresponding to the field that is the same as the technical field is used as the first target patent corresponding to the technical field.

[0057] Specifically, based on the patent application time corresponding to all the first target patents, the patent application year corresponding to each first target patent is determined, and based on the patent application year, all the first target patents are classified and processed to obtain the number of first patent applications corresponding to each patent application year.

[0058] In one embodiment, similar technical fields corresponding to the technical field are obtained, and based on the similar technical fields, the fields corresponding to all patents in the patent database are traversed to obtain all second target patents corresponding to the similar technical fields, and based on the patent application time corresponding to each of the second target patents, all the second target patents are classified to obtain the number of second patent applications corresponding to different years, and a similarity quantity curve is constructed with the number of second patent applications as the vertical coordinate and the year as the horizontal coordinate.

[0059] Specifically, since the terms used in some technical fields are similar or close, if patent traversal is only based on a specific technical field, the statistics or traversal results may be incomplete; taking the field of artificial intelligence as an example, its similar or close technical fields also include deep learning, neural networks, etc. Therefore, in this embodiment, similar technical fields corresponding to the technical field are obtained, and the first quantity curve constructed based on the specific technical field is corrected based on all second target patents corresponding to the similar technical field, which can improve the comprehensiveness of the traversal results.

[0060] In one embodiment, based on the similar quantity curve, the first quantity curve is corrected to obtain a quantity curve.

[0061] Specifically, the similar quantity curve is normalized to obtain a first normalized quantity curve, and the first quantity curve is normalized to obtain a second normalized quantity curve; the first normalized quantity curve and the second normalized quantity curve are weighted averaged to obtain a fitting quantity curve, and the fitting quantity curve is used as the quantity curve.

[0062] In one embodiment, when determining the life cycle of the patent to be analyzed based on the quantity curve, the standard life cycle curves corresponding to different life cycles are obtained, and the similarity calculation process is performed on the quantity curve and the standard life cycle curves corresponding to different life cycles to obtain the similarity corresponding to each standard life cycle curve, and based on the similarity, the target life cycle is determined, and the target life cycle is set as the life cycle of the patent to be analyzed.

[0063] In one embodiment, corresponding scoring values ​​are set for different life cycles. After the life cycle of the patent to be analyzed is determined, the target scoring value corresponding to the life cycle is obtained, and the target scoring value is used as the first scoring value of the output.

[0064] Step 103: Associating the technical features and the technical problems to obtain associated binary tuples, and inputting the associated binary tuples into a pre-trained second scoring model so that the second scoring model obtains the number of related patents corresponding to the associated binary tuples from the patent database, and outputs a second scoring value based on the number of related patents.

[0065] In one embodiment, when the technical feature and the technical problem are associated, the technical feature is used as the first key name, the technical problem is used as the first key value, a first key-value pair is constructed in the form of key name-key value, and the first key-value pair is used as an associated tuple.

[0066] In one embodiment, the first technical feature and the first technical problem corresponding to each patent in the patent database are obtained, and the first technical feature and the first technical problem are associated to obtain the first associated tuple corresponding to each patent.

[0067] In one embodiment, vectorization processing is performed on the associated two-tuple to obtain a first associated vector, and vectorization processing is performed on each first associated two-tuple to obtain a corresponding second associated vector.

[0068] Specifically, word2vec technology is used to perform word vector processing on the associated bigram and each first associated bigram, so as to obtain a first associated word vector corresponding to the associated bigram and a second associated vector corresponding to each first associated bigram.

[0069] In one embodiment, dimensionality reduction processing is performed on the first association vector and each second association vector to obtain first two-dimensional coordinates of the first association vector in a preset vector space and second two-dimensional coordinates of each second association vector in the preset vector space.

[0070] Specifically, the first association vector and each second association vector are reduced to two-dimensional word vectors respectively through a dimensionality reduction algorithm, and they are marked in a preset space vector to obtain the first two-dimensional coordinates of the first association vector in the preset vector space and the second two-dimensional coordinates of each second association vector in the preset vector space, wherein the dimensionality reduction algorithm includes but is not limited to a singular value decomposition algorithm, a principal component analysis algorithm, a factor analysis algorithm and an independent component analysis algorithm.

[0071] In one embodiment, the coordinate distance between the first two-dimensional coordinate and the second two-dimensional coordinate is calculated, and based on the coordinate distance, all second association vectors are screened to obtain a target second association vector set, and the number of vectors in the second association vector set is counted, and the number of vectors is used as the number of related patents corresponding to the associated binary group.

[0072] Specifically, based on the coordinate distance, when all second associated vectors are screened, by setting a coordinate distance threshold, when it is detected that the coordinate distance is greater than the coordinate distance threshold, the second associated vector corresponding to the second two-dimensional coordinate is deleted; if it is detected that the coordinate distance is not greater than the coordinate distance threshold, the second associated vector corresponding to the second two-dimensional coordinate is retained, and all retained second associated vectors are integrated to obtain a target second associated vector set.

[0073] In one embodiment, when outputting the second scoring value based on the number of relevant patents, multiple patent quantity ranges are set, and a corresponding scoring value is set for each patent quantity range. After obtaining the number of relevant patents, the target patent quantity range is determined by judging the attribution relationship between the number of relevant patents and the multiple patent quantity ranges, and the scoring value corresponding to the target patent quantity range is used as the second scoring value.

[0074] Step 104: Associating the technical field, the technical features and the technical problems to obtain associated triples, and inputting the associated triples into a pre-trained patent search model so that the patent search model can obtain patent comparative documents from the patent database.

[0075] In one embodiment, the technical feature is used as the first key name, the technical problem is used as the first key value, and a first key-value pair is constructed in the form of key name-key value. The first key-value pair is used as the second key value, and the technical field is used as the second key name. A second key-value pair is constructed in the form of key name-key value, and the second key-value pair is used as an associated triplet.

[0076] In one embodiment, the first technical field, the first technical feature and the first technical problem corresponding to each patent in the patent database are obtained, and the first technical field, the first technical feature and the first technical problem are associated to obtain the first associated triplet corresponding to each patent.

[0077] In one embodiment, the associated triples are converted into first binary data, and each first associated triplet is converted into corresponding second binary data, and the similarity between the first binary data and each second binary data is calculated, and the patent corresponding to the highest similarity value is obtained as a patent comparison document.

[0078] Specifically, the binary number conversion process is performed using the struct module in Python, where the struct module can define a format string that specifies the data type and size of each element in the triple. Then, the pack method is used to pack the elements in the triple into binary data; for example, if the triple is (technical field, technical features, technical issues), and each element is a string type, a suitable format string can be used to encode these strings.

[0079] Specifically, when respectively calculating the similarity between the first binary data and each second binary data, the similarity between the first binary data and each second binary data can be compared by calculating cosine similarity or Jaccard similarity, etc. This can be achieved by converting text into vectors through a vector space model and then calculating the similarity between the vectors.

[0080] Step 105: Input the patent text and the patent reference document text corresponding to the patent reference document into a third scoring model, so that the third scoring model performs a stability analysis on the patent text based on the patent reference document text, and outputs a third scoring value based on the stability analysis result.

[0081] In one embodiment, the training process of the third scoring model obtains a plurality of historical patent invalidation case data, wherein each historical invalidation case data includes an invalidation case and an invalidation comparative document; performs feature annotation on the invalidation case to obtain a first invalidation feature, performs feature annotation on the invalidation comparative document to obtain a second invalidation feature, performs association processing on the first invalidation feature and the second invalidation feature to obtain an associated invalidation feature, and sets an invalidation reason, an invalidation legal provision, and a stability score for the associated invalidation feature; uses the invalidation case, the invalidation comparative document, the associated invalidation feature, and the invalidation reason, the invalidation legal provision, and the stability score corresponding to each historical invalidation case data as a training sample to generate a training sample set; uses the invalidation case sample and the invalidation comparative document sample corresponding to each training sample in the training sample set as inputs of the initial third scoring model, and uses the associated invalidation feature sample corresponding to each training sample in the training sample set, and the invalidation reason sample, the invalidation legal provision sample, and the stability score sample corresponding to the associated invalidation feature sample as outputs of the initial third scoring model, performs model training on the initial third scoring model, and obtains a trained third scoring model.

[0082] In one embodiment, when the initial third scoring model is trained based on the training sample set, the parameters of the initial learning rate, learning decay method and number of training rounds of the initial third scoring model are set; specifically, the initial learning rate is 0.02, the learning rate decay method is exponential decay, and the number of training rounds is 170 epochs.

[0083] Step 106: Input the first scoring value, the second scoring value and the third scoring value into a pre-trained weighted fusion model for weighted fusion processing, and output the analysis result of the patent to be analyzed.

[0084] In one embodiment, the weighted fusion model sets corresponding weight values ​​according to the importance of the first score value, the second score value, and the third score value.

[0085] Preferably, the weight values ​​can be determined by a variety of methods, including empirical weights based on expert knowledge, optimized weights through cross-validation, or using more complex algorithms such as gradient boosted decision trees (GBDT) to dynamically adjust weights.

[0086] In one embodiment, by calculating a first product value of the first score value and a first weight value corresponding to the first score value, calculating a second product value of the second score value and a second weight value corresponding to the second score value, and calculating a third product value of the third score value and a third weight value corresponding to the third score value, the first product value, the second product value, and the third product value are added to obtain a weighted fusion score value, and based on the weighted fusion score value, the analysis result of the patent to be analyzed is determined.

[0087] In one embodiment, the overall value of the patent to be analyzed is determined based on the score of the weighted fusion score value; specifically, the patent to be analyzed can be graded according to its overall value, and a corresponding grade score range can be set for each grade. Based on the weighted fusion score value, its corresponding target grade score range can be determined, and based on the target grade score range, the overall value grade of the patent to be analyzed can be determined, so as to achieve an intuitive understanding of the patent.

[0088] In the prior art, when a patent is subjected to rapid preliminary examination or priority examination evaluation, it is generally analyzed and processed manually by an agent. It is necessary not only to determine whether the requirements for rapid preliminary examination or priority examination evaluation are met, but also to determine whether the patent is highly innovative. This process is highly subjective and can easily lead to inaccurate evaluation and analysis results. Based on this, in this embodiment, after analyzing the overall value level of the patent, it is also possible to determine whether the patent classification number meets the requirements for patent examination, thereby quickly helping the agent to determine whether the patent meets the requirements for patent examination and improving the accuracy of the analysis.

[0089] In one embodiment, it is determined whether the patent to be analyzed is an unsubmitted patent. If so, the patent comparison file corresponding to the patent to be analyzed output by the patent search module is obtained, and the patent classification number corresponding to the patent comparison file is obtained, and it is determined whether the patent classification number belongs to the target classification number set. If it does, it is determined that the patent to be analyzed meets the patent examination requirements.

[0090] Specifically, the target classification code set includes but is not limited to the fast pre-examination classification code set or the priority review classification code set corresponding to each region.

[0091] Example 2, see Figure 2 , Figure 2 is a schematic diagram of a structure of an embodiment of a patent analysis system based on artificial intelligence provided by the present application, such as Figure 2 As shown, the system includes a patent feature extraction module 201, a first scoring module 202, a second scoring module 203, a patent search module 204, a third scoring module 205 and a patent analysis result generation module 206.

[0092] The patent feature extraction module 201 is used to obtain the patent text of the patent to be analyzed, and input the patent text into a pre-trained patent feature extraction model so that the patent feature extraction model outputs patent feature data; wherein the patent feature data includes technical fields, technical features and technical issues.

[0093] The first scoring module 202 is used to input the technical field into the pre-trained first scoring model, so that the first scoring model obtains the number of patent applications corresponding to the technical field from the patent database, and constructs a quantity curve based on the number of patent applications, and determines the life cycle of the patent to be analyzed based on the quantity curve, and outputs a first scoring value based on the life cycle.

[0094] The second scoring module 203 is used to associate the technical features and the technical problems to obtain associated binary groups, and input the associated binary groups into a pre-trained second scoring model so that the second scoring model obtains the number of related patents corresponding to the associated binary groups from the patent database, and outputs a second scoring value based on the number of related patents.

[0095] The patent search module 204 is used to associate the technical field, the technical features and the technical problems to obtain associated triples, and input the associated triples into a pre-trained patent search model so that the patent search model can obtain patent reference documents from the patent database.

[0096] The third scoring module 205 is used to input the patent text and the patent reference document text corresponding to the patent reference document into a third scoring model, so that the third scoring model performs a stability analysis on the patent text based on the patent reference document text, and outputs a third scoring value based on the stability analysis result.

[0097] The patent analysis result generation module 206 is used to input the first score value, the second score value and the third score value into a pre-trained weighted fusion model for weighted fusion processing, and output the analysis result of the patent to be analyzed.

[0098] The patent analysis system based on artificial intelligence provided in this embodiment also includes: a fusion model building module 200, such as Figure 3 As shown, Figure 3 It is a structural diagram of another embodiment of a patent analysis system based on artificial intelligence provided by the present application.

[0099] The fusion model construction module 200 is used to construct a fusion model, wherein the fusion model includes the patent feature extraction model, the first scoring model, the second scoring model, the third scoring model, the patent retrieval model and a weighted fusion model; wherein the output end of the patent feature extraction model is respectively connected to the input ends of the first scoring model, the second scoring model and the patent retrieval model, and the output end of the patent retrieval model is connected to the input end of the third scoring model; the output ends of the first scoring model, the second scoring model and the third scoring model are connected to the input end of the weighted fusion model.

[0100] In one embodiment, the patent feature extraction module 201 is used to input the patent text into a pre-trained patent feature extraction model so that the patent feature extraction model outputs patent feature data; wherein the patent feature data includes technical fields, technical features and technical problems, specifically including: the patent feature extraction model includes a text segmentation sub-model, a first feature extraction sub-model, a second feature extraction sub-model and a third feature extraction sub-model, the output end of the text segmentation sub-model is respectively connected to the input end of the first feature extraction sub-model, the second feature extraction sub-model and the third feature extraction sub-model; the patent text is input into the text segmentation sub-model so that the text segmentation sub-model performs segmentation processing on the patent text to obtain claim text data, technical field text data and technical problem text data; the claim The claim text data is input into the first feature extraction sub-model, so that the first feature extraction sub-model performs sentence processing on the claim text data to obtain multiple text data, and respectively performs keyword extraction on the multiple text data to obtain multiple keywords, and uses the multiple keywords as the output technical features; the technical field text data is input into the second feature extraction sub-model, so that the second feature extraction sub-model performs target phrase positioning processing on the technical field text data, obtains and determines the technical field position based on the first target phrase position and the second target phrase position, and extracts and outputs the technical field based on the technical field position; the technical problem text data is input into the third feature extraction sub-model, so that the third feature extraction model performs intent analysis processing on the technical problem text data to obtain and output the technical problem.

[0101] In one embodiment, the first scoring module 202 is used to obtain the number of patent applications corresponding to the technical field from the patent database, and construct a quantity curve based on the number of patent applications, specifically including: based on the technical field, traversing the fields corresponding to all patents in the patent database to obtain all first target patents corresponding to the technical field, and classifying all first target patents based on the patent application time corresponding to each of the first target patents to obtain the number of first patent applications corresponding to different years, and constructing a first quantity curve with the first number of patent applications as the ordinate and the year as the abscissa; obtaining similar technical fields corresponding to the technical field, traversing the fields corresponding to all patents in the patent database based on the similar technical fields to obtain all second target patents corresponding to the similar technical fields, and classifying all second target patents based on the patent application time corresponding to each of the second target patents to obtain the number of second patent applications corresponding to different years, and constructing a similar quantity curve with the second number of patent applications as the ordinate and the year as the abscissa; correcting the first quantity curve based on the similar quantity curve to obtain a quantity curve.

[0102] In one embodiment, the second scoring module 203 is used to obtain the number of related patents corresponding to the associated binary group from the patent database, specifically including: obtaining the first technical feature and the first technical problem corresponding to each patent in the patent database, and performing association processing on the first technical feature and the first technical problem to obtain the first associated binary group corresponding to each patent; performing vectorization processing on the associated binary group to obtain a first associated vector, and performing vectorization processing on each first associated binary group respectively to obtain a corresponding second associated vector; performing dimensionality reduction processing on the first associated vector and each second associated vector respectively to obtain the first two-dimensional coordinates of the first associated vector in the preset vector space, and the second two-dimensional coordinates of each second associated vector in the preset vector space; calculating the coordinate distance between the first two-dimensional coordinates and the second two-dimensional coordinates, based on the coordinate distance, screening all second associated vectors to obtain a target second associated vector set, counting the number of vectors in the second associated vector set, and using the number of vectors as the number of related patents corresponding to the associated binary group.

[0103] In one embodiment, the patent search module 204 is used to input the associated triples into a pre-trained patent search model so that the patent search model obtains patent comparative documents from the patent database, specifically including: obtaining the first technical field, the first technical feature and the first technical problem corresponding to each patent in the patent database, and performing association processing on the first technical field, the first technical feature and the first technical problem to obtain the first associated triples corresponding to each patent; converting the associated triples into first binary data, and converting each first associated triplet into its corresponding second binary data, respectively calculating the similarity between the first binary data and each second binary data, and obtaining the patent corresponding to the highest similarity value as the patent comparative document.

[0104] In one embodiment, the training process of the third scoring model specifically includes: obtaining a plurality of historical patent invalidation case data, wherein each historical invalidation case data includes an invalidation case and an invalidation comparative document; performing feature annotation on the invalidation case to obtain a first invalidation feature, and performing feature annotation on the invalidation comparative document to obtain a second invalidation feature, respectively performing association processing on the first invalidation feature and the second invalidation feature to obtain an associated invalidation feature, and setting an invalidation reason, an invalidation legal provision, and a stability score for the associated invalidation feature; taking the invalidation case, the invalidation comparative document, the associated invalidation feature, and the invalidation reason, the invalidation legal provision, and the stability score corresponding to each historical invalidation case data as a training sample to generate a training sample set; respectively taking the invalidation case sample and the invalidation comparative document sample corresponding to each training sample in the training sample set as the input of the initial third scoring model, and taking the associated invalidation feature sample corresponding to each training sample in the training sample set, and the invalidation reason sample, the invalidation legal provision sample, and the stability score sample corresponding to the associated invalidation feature sample as the output of the initial third scoring model, and performing model training on the initial third scoring model to obtain a trained third scoring model.

[0105] The patent analysis system based on artificial intelligence can implement the patent analysis method based on artificial intelligence in the above method embodiment. The options in the above method embodiment are also applicable to this embodiment and will not be described in detail here.

[0106] like Figure 4 As shown, Figure 4 It is a structural diagram of an electronic device provided by the present application; it includes a processor 111, a communication interface 112, a memory 113 and a communication bus 114, wherein the processor 111, the communication interface 112, and the memory 113 communicate with each other through the communication bus 114, and the memory 113 is used to store computer programs.

[0107] In one embodiment of the present application, the processor 111 is used to implement the artificial intelligence-based patent analysis method provided by any of the aforementioned method embodiments when executing the program stored in the memory 113.

[0108] An embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the artificial intelligence-based patent analysis method provided in any of the aforementioned method embodiments are implemented.

[0109] The system embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0110] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0111] It should be understood that the terms used herein are only for the purpose of describing specific example embodiments and are not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms "one", "an" and "said" as used herein may also be meant to include plural forms. The terms "include", "comprise", "contain", and "have" are inclusive, and therefore specify the existence of stated features, steps, operations, elements and / or parts, but do not exclude the existence or addition of one or more other features, steps, operations, elements, parts, and / or combinations thereof. The method steps, processes, and operations described herein are not interpreted as necessarily requiring them to be performed in the specific order described or illustrated, unless the execution order is clearly indicated. It should also be understood that additional or alternative steps may be used.

[0112] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A patent analysis method based on artificial intelligence, characterized in that: include: Obtaining a patent text of a patent to be analyzed, and inputting the patent text into a pre-trained patent feature extraction model, so that the patent feature extraction model outputs patent feature data; wherein the patent feature data includes technical fields, technical features, and technical issues; Input the technical field into a pre-trained first scoring model, so that the first scoring model obtains the number of patent applications corresponding to the technical field from a patent database, and constructs a quantity curve based on the number of patent applications, and determines the life cycle of the patent to be analyzed based on the quantity curve, and outputs a first scoring value based on the life cycle; Associating the technical features with the technical problems to obtain associated binary groups, and inputting the associated binary groups into a pre-trained second scoring model, so that the second scoring model obtains the number of related patents corresponding to the associated binary groups from the patent database, and outputs a second scoring value based on the number of related patents; Associating the technical field, the technical features and the technical problem to obtain associated triples, and inputting the associated triples into a pre-trained patent search model so that the patent search model obtains patent reference documents from the patent database; Inputting the patent text and the patent reference document text corresponding to the patent reference document into a third scoring model, so that the third scoring model performs stability analysis on the patent text based on the patent reference document text, and outputs a third scoring value based on the stability analysis result; The first scoring value, the second scoring value and the third scoring value are input into a pre-trained weighted fusion model for weighted fusion processing, and the analysis result of the patent to be analyzed is output.

2. The method according to claim 1, characterized in that Before obtaining the patent text of the patent to be analyzed, it also includes: Constructing a fusion model, wherein the fusion model includes the patent feature extraction model, the first scoring model, the second scoring model, the third scoring model, the patent retrieval model and a weighted fusion model; The output end of the patent feature extraction model is connected to the input ends of the first scoring model, the second scoring model and the patent search model respectively, and the output end of the patent search model is connected to the input end of the third scoring model; The output ends of the first scoring model, the second scoring model and the third scoring model are connected to the input end of the weighted fusion model.

3. The method according to claim 1, characterized in that: The patent text is input into a pre-trained patent feature extraction model so that the patent feature extraction model outputs patent feature data; wherein the patent feature data includes technical fields, technical features and technical issues, specifically including: The patent feature extraction model includes a text segmentation sub-model, a first feature extraction sub-model, a second feature extraction sub-model and a third feature extraction sub-model, and the output end of the text segmentation sub-model is connected to the input ends of the first feature extraction sub-model, the second feature extraction sub-model and the third feature extraction sub-model respectively; Inputting the patent text into the text segmentation sub-model so that the text segmentation sub-model performs segmentation processing on the patent text to obtain claim text data, technical field text data and technical problem text data; Inputting the claim text data into the first feature extraction sub-model, so that the first feature extraction sub-model performs sentence processing on the claim text data to obtain multiple text data, and respectively extracts keywords from the multiple text data to obtain multiple keywords, and uses the multiple keywords as output technical features; Inputting the technical field text data into the second feature extraction sub-model so that the second feature extraction sub-model performs target phrase location processing on the technical field text data, obtains and determines the technical field position based on the first target phrase position and the second target phrase position, and extracts and outputs the technical field based on the technical field position; The technical problem text data is input into the third feature extraction sub-model, so that the third feature extraction model performs intent analysis on the technical problem text data to obtain and output the technical problem.

4. The method according to claim 1, characterized in that: The obtaining of the number of patent applications corresponding to the technical field from the patent database and constructing a quantity curve based on the number of patent applications specifically includes: Based on the technical field, the fields corresponding to all patents in the patent database are traversed to obtain all first target patents corresponding to the technical field, and based on the patent application time corresponding to each of the first target patents, all the first target patents are classified to obtain the number of first patent applications corresponding to different years, and a first quantity curve is constructed with the number of the first patent applications as the ordinate and the year as the abscissa; Obtain similar technical fields corresponding to the technical field, and based on the similar technical fields, traverse the fields corresponding to all patents in the patent database to obtain all second target patents corresponding to the similar technical fields, and classify all second target patents based on the patent application time corresponding to each of the second target patents to obtain the number of second patent applications corresponding to different years, and construct a similarity quantity curve with the number of second patent applications as the ordinate and the year as the abscissa; Based on the similar quantity curve, the first quantity curve is corrected to obtain a quantity curve.

5. The method according to claim 1, characterized in that: The obtaining the number of related patents corresponding to the associated binary group from the patent database specifically includes: Obtaining a first technical feature and a first technical problem corresponding to each patent in the patent database, and performing association processing on the first technical feature and the first technical problem to obtain a first associated tuple corresponding to each patent; Performing vectorization processing on the associated two-tuple to obtain a first associated vector, and performing vectorization processing on each first associated two-tuple to obtain a corresponding second associated vector; Performing dimensionality reduction processing on the first association vector and each second association vector respectively to obtain first two-dimensional coordinates of the first association vector in a preset vector space and second two-dimensional coordinates of each second association vector in the preset vector space; Calculate the coordinate distance between the first two-dimensional coordinate and the second two-dimensional coordinate, screen all second associated vectors based on the coordinate distance to obtain a target second associated vector set, count the number of vectors in the second associated vector set, and use the number of vectors as the number of related patents corresponding to the associated binary group.

6. The method according to claim 1, characterized in that: The step of inputting the associated triples into a pre-trained patent search model so that the patent search model acquires patent reference documents from the patent database specifically includes: Obtaining the first technical field, the first technical feature, and the first technical problem corresponding to each patent in the patent database, and performing association processing on the first technical field, the first technical feature, and the first technical problem to obtain a first association triplet corresponding to each patent; Convert the associated triples into first binary data, and convert each first associated triplet into corresponding second binary data, respectively calculate the similarity between the first binary data and each second binary data, and obtain the patent corresponding to the highest similarity value as a patent comparison document.

7. The method according to claim 1, characterized in that: The training process of the third scoring model specifically includes: Obtain multiple historical patent invalidation case data, wherein each historical invalidation case data includes an invalidation case and an invalidation comparative document; Performing feature marking on the invalid case to obtain a first invalid feature, and performing feature marking on the invalid comparative document to obtain a second invalid feature, respectively performing association processing on the first invalid feature and the second invalid feature to obtain an associated invalid feature, and setting an invalid reason, an invalid law provision, and a stability score for the associated invalid feature; The invalid case, the invalid comparison document, the associated invalid feature, and the invalid reason, the invalid law and the stability score corresponding to each historical invalid case data are used as a training sample to generate a training sample set; The invalid case samples and the invalid comparison document samples corresponding to each training sample in the training sample set are respectively used as inputs of the initial third scoring model, and the associated invalid feature samples corresponding to each training sample in the training sample set, as well as the invalid reason samples, the invalid law article samples and the stability score samples corresponding to the associated invalid feature samples are used as outputs of the initial third scoring model, and the initial third scoring model is trained to obtain a trained third scoring model.

8. A patent analysis system based on artificial intelligence, characterized in that: include: Patent feature extraction module, first scoring module, second scoring module, patent search module, third scoring module and patent analysis result generation module; The patent feature extraction module is used to obtain the patent text of the patent to be analyzed, and input the patent text into the pre-trained patent feature extraction model so that the patent feature extraction model outputs patent feature data; wherein the patent feature data includes technical fields, technical features and technical issues; The first scoring module is used to input the technical field into a pre-trained first scoring model, so that the first scoring model obtains the number of patent applications corresponding to the technical field from a patent database, and constructs a quantity curve based on the number of patent applications, and determines the life cycle of the patent to be analyzed based on the quantity curve, and outputs a first scoring value based on the life cycle; The second scoring module is used to associate the technical features with the technical problems to obtain an associated binary group, and input the associated binary group into a pre-trained second scoring model so that the second scoring model obtains the number of related patents corresponding to the associated binary group from the patent database, and outputs a second scoring value based on the number of related patents; The patent search module is used to associate the technical field, the technical features and the technical problem to obtain an associated triplet, and input the associated triplet into a pre-trained patent search model so that the patent search model obtains patent reference documents from the patent database; The third scoring module is used to input the patent text and the patent reference document text corresponding to the patent reference document into a third scoring model, so that the third scoring model performs stability analysis on the patent text based on the patent reference document text, and outputs a third scoring value based on the stability analysis result; The patent analysis result generation module is used to input the first score value, the second score value and the third score value into a pre-trained weighted fusion model for weighted fusion processing, and output the analysis result of the patent to be analyzed.

9. An electronic device, comprising: at least one communication interface; at least one bus connected to the at least one communication interface; at least one processor coupled to the at least one bus; At least one memory connected to the at least one bus, wherein the processor is used to implement the artificial intelligence-based patent analysis method described in any one of claims 1-7 when executing the program stored in the memory.

10. A computer storage medium storing computer executable instructions, wherein the computer executable instructions are used to execute the artificial intelligence-based patent analysis method described in any one of claims 1 to 7 of the present application.

Images