A key technology identification method and related apparatus

Abstract

The application discloses a key technology identification method and related device, which is applied to the field of big data. In the application, first, to-be-identified patent data is acquired, and key technology patents meeting key estimation conditions are filtered based on the to-be-identified patent data, wherein the key estimation conditions are used to measure and determine the key degree of the to-be-identified patent data. Then, the BERT model is used to determine the target theme words of the key technology patents, the number of the target theme words in the key technology patents is counted, and the abstract and the patent theme corresponding to the key technology patent whose number meets a preset threshold are spliced to obtain a target document. Finally, the target document is identified through a text generation model, and an identification result of the key technology is generated. The key technology can be identified in an automatic manner without relying on human experience, and the efficiency and accuracy of the key technology identification are improved.

Description

Technical Field

[0001] This application relates to the field of big data technology, and in particular to a key technology identification method and related apparatus. Background Technology

[0002] With social development, the current patent data is growing rapidly. This large amount of patent data contains a wealth of technical information. Identifying the key technologies within it can help us grasp the trends in technological development and carry out targeted technological innovation.

[0003] In related technologies, the identification of key technologies in patent data is often done manually, relying on human experience and judgment, which cannot accurately and efficiently identify key technologies.

[0004] Given the shortcomings of the aforementioned technologies, there is an urgent need for a method that can efficiently and accurately identify key technologies. Summary of the Invention

[0005] In view of this, embodiments of this application provide a key technology identification method and related apparatus, aiming to improve the efficiency and accuracy of key technology identification.

[0006] In a first aspect, embodiments of this application provide a method for identifying key technologies, the method comprising:

[0007] Obtain the patent data to be identified;

[0008] Based on the patent data to be identified, key technology patents that meet the key prediction conditions are screened. The key prediction conditions are used to determine the keyness of the patent data to be identified.

[0009] The target keywords for the key technology patents were determined based on the BERT model.

[0010] The number of target keywords in the key technology patents is counted, and the abstracts and patent topics corresponding to the key technology patents whose numbers meet a preset threshold are concatenated to obtain the target document;

[0011] The target document is identified using a text generation model to generate identification results for key technologies.

[0012] Optionally, the key prediction criteria include a score of the patent data to be identified. After screening key technology patents that meet the key prediction criteria based on the patent data to be identified, the method further includes:

[0013] Determine the scores and rank the key technology patents;

[0014] The first k key technology patents in the sorted list are taken as target patents, where k>1;

[0015] Extract the abstract from the target patent;

[0016] Based on the extracted summary, a description of the key technology is generated using a large language model (LLM).

[0017] Optionally, the determination of target keywords for the key technology patents based on the BERT model includes:

[0018] Obtain the subject and keywords of the key technology patent, wherein the subject is used to indicate the technical direction of the key technology patent and the keywords are used to indicate the technical application of the key technology patent;

[0019] The topic vector is obtained by encoding the topic based on the BERT model, and the topic word is obtained by encoding the topic word.

[0020] Calculate the cosine similarity between the topic vector and the topic word vector, and determine the topic words whose cosine similarity meets the preset conditions as target topic words.

[0021] Optionally, the method further includes:

[0022] Construct a stop word list, which includes words that are not related to the key technology;

[0023] The key technology patent includes an abstract text, and obtaining the subject and keywords of the key technology patent includes:

[0024] The abstract text of the key technology patent is used as the text to be processed, which includes Chinese text and / or English text.

[0025] The Chinese text and / or the English text are segmented and tagged with parts of speech.

[0026] Based on the results of the word segmentation and part-of-speech tagging, the Chinese text and / or the English text are filtered using the stop word list to obtain keyword data;

[0027] The subject and keywords of the key technology patent are determined based on the keyword data.

[0028] Optionally, determining the subject and keywords of the key technology patent based on the keyword data includes:

[0029] Based on the keyword data, a first distribution and a second distribution are initialized by the LDA model. The first distribution indicates the distribution between the keyword data and the topic, and the second distribution indicates the distribution between the topic and the term.

[0030] The topic allocation is adjusted based on Gibbs sampling, and the first distribution and the second distribution are updated according to the allocated topics;

[0031] The subject of the key technology patent is determined based on the updated first distribution, and the subject terms of the key technology patent are determined based on the updated second distribution.

[0032] Optionally, determining the subject and keywords of the key technology patent based on the keyword data includes:

[0033] Obtain the keyword data and the summary text;

[0034] The first feature vector is obtained by encoding the summary text based on the BERT model;

[0035] Based on the keyword data, the first distribution obtained by using the LDA model is used to determine the probability distribution of the topic corresponding to the document, and the probability distribution is used as the second feature vector. The first distribution indicates the distribution between the keyword data and the topic.

[0036] The subject and keywords of the key technology patent are determined based on the first feature vector and the second feature vector.

[0037] Optionally, the method further includes:

[0038] Establish a lexicon of attributive terms;

[0039] The step of recognizing the target document using a text generation model to generate recognition results for key technologies includes:

[0040] The target document is input into the text generation model to obtain the title of the target document;

[0041] The title is processed by the adjective lexicon, and the title that has undergone the adjective processing is determined as the identification result of the key technology.

[0042] Secondly, embodiments of this application provide a key technology identification device, the device comprising: an acquisition module, a filtering module, a determination module, a statistics module, and an identification module;

[0043] The acquisition module is used to acquire the patent data to be identified;

[0044] The filtering module is used to filter key technology patents that meet the key prediction conditions based on the patent data to be identified. The key prediction conditions are used to determine the keyness of the patent data to be identified.

[0045] The determining module is used to determine the target keywords for the key technology patent based on the BERT model;

[0046] The statistics module is used to count the number of target keywords in the key technology patents, and to combine the abstracts and patent topics of key technology patents whose counts meet a preset threshold to obtain the target document.

[0047] The recognition module is used to recognize the target document through a text generation model and generate recognition results for key technologies.

[0048] Thirdly, this application provides an electronic device, the device comprising: a processor, a memory, and a system bus;

[0049] The processor and the memory are connected via the system bus;

[0050] The memory is used to store one or more programs, the one or more programs including instructions that, when executed by the processor, cause the processor to perform a key technology identification method in any of the implementations mentioned in the first aspect.

[0051] Fourthly, embodiments of this application provide a computer storage medium storing code, wherein when the code is executed, a device running the code implements the key technology identification method in any of the implementations mentioned in the first aspect.

[0052] This application provides a method and related apparatus for identifying key technologies. When executing the method, firstly, patent data to be identified is acquired. Based on the patent data, key technology patents that meet key prediction conditions are screened, whereby the key prediction conditions are used to measure and determine the keyness of the patent data to be identified. Then, target keywords for the key technology patents are determined based on the BERT model, and the number of target keywords in the key technology patents is counted. The abstracts and patent topics corresponding to key technology patents with a number meeting a preset threshold are concatenated to obtain a target document. Finally, the target document is identified using a text generation model to generate the key technology identification result. Thus, key technology patents are screened based on key prediction conditions, target keywords in the key technology patents are determined using the BERT model, the number of target keywords in the key technology patents is counted, and key technology patents with a number meeting a preset threshold are identified. The key technology patent obtained at this time is a patent with a certain number of target subject words. This patent is highly likely to contain key technologies. The target document is obtained by splicing the abstract and subject of the key technology patent. The key technology can be obtained by using a text generation model to identify the target document. The whole process does not rely on human experience, but adopts an automated method to identify key technologies. The final key technology identification result can meet the current identification needs, ensure the accuracy of key technology identification, and improve the efficiency of key technology identification. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings used in the description of the embodiment or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 A flowchart illustrating a key technology identification method provided in this application embodiment;

[0055] Figure 2 A flowchart illustrating a method for determining target keywords provided in this application embodiment;

[0056] Figure 3 This is a schematic diagram of the structure of a key technology identification device provided in an embodiment of this application;

[0057] Figure 4 This is a schematic diagram of another key technology identification device provided in the embodiments of this application. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] In this application, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0060] Research on related technologies has revealed that in order to extract key technologies from massive amounts of patent data, and to grasp technological development trends based on these key technologies, it is crucial to provide directional guidance for technicians and researchers, enabling them to conduct technological innovation in a targeted manner and improve innovation efficiency. The goal is to accurately identify key technologies within patent data. However, in related technologies, this process is often done manually, which is not only extremely resource-intensive but also inefficient.

[0061] Based on this, this application proposes a key technology identification method and related apparatus. It enables the identification of key technologies automatically, without relying on human experience. The final identification results meet current identification requirements, ensuring the accuracy of key technology identification while improving its efficiency.

[0062] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0063] Figure 1 A flowchart of a key technology identification method provided in an embodiment of this application is shown below. Figure 1 As shown, a key technology identification method provided in this application includes:

[0064] S11: Obtain the patent data to be identified.

[0065] Patent data to be identified refers to patent data used to screen key technology patents. This patent data typically comprises one or more patents. In addition to the main text of the patent, this patent data includes, but is not limited to, the following features: patent application number, patent publication number, patent title, patent abstract, patent publication date, and number of citations. A specific method for obtaining the patent data to be identified can be as follows: first, obtain the original patent data; then, process the original patent data to obtain the patent data to be identified.

[0066] Original patent data refers to unprocessed patent data, which typically originates from databases, patent-related websites, etc. Because original patent data may contain blank or redundant features, it is necessary to process it to obtain the patent data to be identified. This process avoids invalid or redundant data that could negatively impact the effectiveness of subsequent key technology identification.

[0067] The aforementioned data processing may include at least one of the following: data cleaning, data filtering, data imputation, and data partitioning. The following sections provide a detailed description of each data processing method:

[0068] (1) Data filtering: Filter the original patent data in the feature data where the content of the patent title, patent abstract, patent publication number, patent application number and patent publication date is empty; filter the original patent data that is neither Chinese nor English according to the patent title and patent abstract (the specific language filtering method can be determined according to the actual situation and application scenario, the above is only an example).

[0069] (2) Data Filling: For other feature data in the original patent data besides the patent title, patent abstract, patent publication number, patent application number, and patent publication date, data filling is performed when the feature is empty. Numerical features are filled with 0, such as the number of citations, the number of citations, etc.; array-type features are filled with empty arrays, such as the IPC classification number, the award level, etc.

[0070] (3) Data division: The original patent data is divided into Chinese and English according to the patent title and patent abstract (other languages ​​are also allowed, which are not limited here).

[0071] (4) Data cleaning: The patent title and patent abstract are cleaned by regular expressions, including removing HTML code characters, replacing line breaks, removing extra spaces, removing leading and trailing whitespace characters, and filtering out original patent data with a patent title length of less than 2 or a patent abstract length of less than 5.

[0072] When processing the original patent data, one or more of the above data processing methods can be selected in combination with the specific situation of the characteristic data for data processing. After the original patent data is processed, the patent data to be recognized can be obtained for subsequent identification of key technologies.

[0073] S12: Screen key technology patents that meet the key prediction conditions based on the patent data to be recognized.

[0074] The key prediction conditions are the criteria for measuring key technology patents. Based on these key prediction conditions, key technology patents can be screened from the patent data to be recognized. That is, the key prediction conditions are used to determine the key degree of the patent data to be recognized, and this key degree can be determined in a quantifiable manner.

[0075] For example, in the embodiments of the present application, the key prediction conditions can be set by setting evaluation conditions. For example, a scoring standard regarding the characteristic data is set for the patent data to be recognized, and based on this scoring standard, specific scores are determined for each patent data to be recognized in a quantifiable manner, and the scoring situation is used as the condition for measuring key technology patents. At this time, the key prediction condition can be that the patent data to be recognized with a score of 90 is a key technology patent, or the patent data to be recognized whose scoring situation is in the top 10% of all patent data to be recognized is a key technology patent, etc.

[0076] The following exemplarily shows a scoring standard for different characteristic data. The specific scoring standard can be set by those skilled in the art according to the actual situation and application scenario, and is not limited herein.

[0077] Whether it is classified as a strategic emerging industry, "yes" = 100 points, "no" = 0 points.

[0078] Average annual citation number X, X = 0, 0 points; 0 < X <= 0.2, 30 points; 0.2 < X <= 0.6, 60 points; 0.6 < X <= 1.25, 80 points; 1.25 < X, 100 points.

[0079] Citation number X, X = 0, zero points; one <= X <= 2, 10 points; three <= X <= 5, 20 points; six <= X <= 7, 40 points; eight <= X <= 9, 70 points; X >= ten, 100 points.

[0080] Technical coverage X, X = 0, zero points; X = 1, 30 points; X = 2, 50 points; X = 3, 70 points; X >= 4, 100 points.

[0081] Number of inventors, X = 0, zero points; X = 1, 10 points; X = 2, 30 points; three <= X <= 5, 60 points; five < X <= nine, 80 points; X >= ten, 100 points.

[0082] The number of applicants X, where X = 0, 0 points; X = 1, 90 points; 1 < X, 100 points.

[0083] The number of pages of the document X, where X = 0, 0 points; X <= 5, 20 points; 5 < X <= 10, 50 points; 10 < X <= 15, 70 points; 15 < X <= 20, 90 points; X > 20, 100 points.

[0084] PCT application, "Yes" = 100 points, "No" = 0 points.

[0085] Patent operation X, where X = 0, 0 points; X = 1, 80 points; X >= 2, 100 points.

[0086] The number of同族专利数X, where X = 1, 0 points; X = 2, 20 points; 2 < X <= 5, 60 points; < X <= 8, 80 points; 8 < X, 100 points.

[0087] The number of citations, X = 0, 0 points; X <= 2, 80 points; 3 <= X, 100 points.

[0088] Award-winning situation, gold award, 100 points; silver award, 80 points; excellent, 60 points; other awards, 40 points; no award, 0 points.

[0089] In addition to setting the above scoring criteria, different weights can also be set for different characteristic data. Based on the above scoring criteria and the corresponding weights, the score of the patent data to be identified can be obtained, and the key technology patent can be determined according to the score. The specific way of assigning weights to each characteristic data can be set by those skilled in the art according to the actual situation and application scenarios, and is not limited herein.

[0090] Combined with the foregoing description, in one possible implementation, the key prediction condition includes the score of the patent data to be identified. After screening the key technology patents that meet the key prediction conditions from the patent data to be identified based on the above scoring situation, the description of the corresponding key technology can be generated based on the abstract of the key technology patent. The method can be as follows: First, determine the scores of the key technology patents and sort them, and take the top k key technology patents in the sorting as the target patents, where k > 1. Then, extract the abstracts of the target patents, and use the large language model (LLM) to generate the description of the key technology based on the extracted abstracts.

[0091] Specifically, based on the key technology patents identified through the scoring of the patent data to be identified, the scores of these key technology patents can be ranked, and the top k key technology patents in the ranking can be selected as target patents. For example, the scores of the key technology patents can be arranged in descending order, and the top four key technology patents can be selected as target patents. The process of determining target patents generally aims to select key technology patents with higher scores.

[0092] After identifying the target patents, their corresponding abstracts can be obtained. Generally, the abstracts clarify the key information of the technical topics involved in the target patent, thus allowing the identification of the corresponding technical topics. Once the abstracts are obtained, they can be input into a Large Language Model (LLM) as representative documents to obtain the technical points predicted by the LLM, and a comprehensive technical description of the topic can be constructed based on these predictions. The construction process involves piecing together the technical points from these representative documents to form the technical description. The technical points refer to the specific technical means employed in the target patent, such as cell culture and separation technology, culture medium generation technology, etc.

[0093] In this embodiment, the Large Language Model (LLM) is pre-trained on a large amount of text data, possessing a deep understanding and processing capability of natural language. This enables it to accurately identify and parse the key technical points in representative documents, thus providing a foundation for constructing comprehensive subject-specific technical descriptions. For example, the LLM can use the Qwen1.5-14B-Chat open-source model. Below is an example of generating key technology descriptions for patent abstracts based on the Qwen1.5-14B-Chat open-source model:

[0094] Patent Abstract: This document provides a method for guiding the differentiation of human pluripotent stem cells into a three-dimensional multilayer skin composition comprising an epidermis, a dermis, and various cells capable of forming functional hair follicles. This document also provides the three-dimensional multilayer engineered skin composition and methods for using it for drug screening, screening compounds for their effects on hair growth, and for other applications.

[0095] Description of key technologies: `ˋ`ˋˋ`json{"result":"Involves stem cell differentiation and tissue engineering, including methods that induce human pluripotent stem cells to form three-dimensional skin structures with epidermal, dermal, and hair follicle cells for drug screening and compound testing."

[0096] "}ˋ`ˋ`ˋ`

[0097] As described above, in this embodiment, key technology patents are identified by scoring the patent data to be identified. The top k key technology patents by score are then selected as target patents. A summary is extracted from each target patent, and a description of the key technology is generated using a Large Language Model (LLM) based on the extracted summary. This description includes the key technical points of the target patent, which facilitates the identification of key technologies. Furthermore, the use of a LLM enables automated generation of key technology descriptions. Extracting summaries from the selected target patents ensures a high degree of relevance between the extracted summary content and the key technologies.

[0098] S13: Determine the target keywords for the key technology patents based on the BERT model.

[0099] BERT (Bidirectional Encoder Representations from Transformers) is a pre-trained language representation model based on the Transformer architecture. It learns text representations through large-scale unsupervised training. In this embodiment, the target keywords for key technology patents need to be determined based on the BERT model. Figure 2 A flowchart illustrating a method for determining target keywords provided in this application embodiment. In one possible implementation, the method for determining target keywords is as follows:

[0100] S131: Obtain the subject matter and keywords of the key technology patent.

[0101] The subject line indicates the technical direction of the key technology patent, while the keywords indicate its technical application. A key technology patent includes both its subject line and its keywords. For example, the subject line could be: human neural stem cell culture, pancreatic islet stem cell therapy for diabetes, etc., and the keywords could be: somatic cells, neural stem cells, pancreatic islets, etc.

[0102] S132: Encode the topic using the BERT model to obtain a topic vector, and encode the topic words to obtain a topic word vector.

[0103] The BERT model can extract vector representations from text, capturing semantic information from the text within these vectors. Specifically, it can encode topics to obtain topic vectors and topic words to obtain topic word vectors, both of which contain semantic information about the corresponding topics and topic words.

[0104] S133: Calculate the cosine similarity between the topic vector and the topic word vector, and determine the topic words whose cosine similarity meets the preset conditions as target topic words.

[0105] Cosine similarity is an index used to measure the similarity between two vectors. In this embodiment, cosine similarity is used to measure the similarity between topic vectors and topic word vectors. The preset condition can be that the cosine similarity between topic word vectors and topic vectors reaches a certain value, or that the cosine similarity value is among the highest in all topic words. For example, assuming the range of cosine similarity is [-1, +1], where a value closer to 1 indicates that the two vectors are more similar, a value closer to -1 indicates that the two vectors are more opposite, and a value of 0 indicates that the two vectors are orthogonal or unrelated. In this case, the preset condition can be set to a cosine similarity value less than -0.6, or a cosine similarity value ranking among the top 15 in all topic words. When the preset condition is a cosine similarity value less than -0.6, all topic words with a cosine similarity value less than -0.6 are identified as target topic words. When the preset condition is a cosine similarity value ranking among the top 15 in all topic words, all topic words with a cosine similarity value ranking among the top 15 in all topic words are identified as target topic words.

[0106] The method for determining target keywords described above can obtain topic vectors and keyword vectors corresponding to each topic and keyword based on the BERT model. By calculating the cosine similarity between the topic vectors and keyword vectors, the keyword vectors that meet preset conditions are selected as target keywords. Using the BERT model allows for efficient and rapid acquisition of topic vectors and keyword vectors. The calculation of cosine similarity enables the determination of target keywords in a quantifiable manner, ensuring a close correlation between the determined target keywords and the corresponding technical direction, thereby improving the accuracy of subsequent key technology identification.

[0107] S14: Count the number of target keywords in the key technology patents, and combine the abstracts and patent topics corresponding to the key technology patents whose number meets the preset threshold to obtain the target document.

[0108] In S13, target keywords were identified. These target keywords are those closely related to the technical direction of the research topic. To ensure the generated key technology patents have a more comprehensive and general theme, it is generally believed that key technology patents containing more target keywords are more likely to cover the core technologies and key concepts of the topic. Therefore, the number of target keywords contained in each key technology patent can be statistically analyzed. The key technology patent containing the most target keywords is selected as the representative document P. top Specifically, it can be determined using the following formula:

[0109] Ptop =arg max(count(P i∈PL (TopW)))

[0110] Where P refers to the individual patent involved, PL refers to the key technology patent, and TopW refers to the target keywords. The aforementioned preset threshold can be freely set by those skilled in the art based on actual circumstances and application scenarios, and is not limited here. For example, the preset threshold can be set to 20. Alternatively, the preset threshold can be determined as the highest achievable number of keywords included.

[0111] The above method allows us to quantify the relevance of each patent to the research topic. The patent title and abstract summarize the main technical content and application areas of the patent, which is beneficial for generating technical topic titles. To construct the target document T, we concatenate the abstract and title of patent P together, removing some meaningless modifiers during the concatenation process, and generate the target document T.

[0112] For example, suppose the abstract is as follows: This invention relates to the use of pancreatic islet stem cells in the treatment of diabetes. This invention provides isolated pancreatic islet stem cells with the function of secreting insulin. Simultaneously, a monoclonal antibody specifically targeting CD3 was prepared, which has the function of activating T cells. Furthermore, when this antibody is used in combination with pancreatic islet stem cells, it can effectively inhibit the immune mechanism in mice and lower blood glucose levels. This can be used to prepare pharmaceutical compositions for the treatment of diabetes and has excellent application prospects. The patent subject is: Use of pancreatic islet stem cells in the treatment of diabetes. A specific splicing result could be: Use of pancreatic islet stem cells in the treatment of diabetes. This invention relates to the use of pancreatic islet stem cells in the treatment of diabetes. This invention provides isolated pancreatic islet stem cells with the function of secreting insulin. Simultaneously, a monoclonal antibody specifically targeting CD3 was prepared, which has the function of activating T cells. Furthermore, when this antibody is used in combination with pancreatic islet stem cells, it can effectively inhibit the immune mechanism in mice and lower blood glucose levels. This can be used to prepare pharmaceutical compositions for the treatment of diabetes and has excellent application prospects.

[0113] When determining the final target document, the concatenated result of the summary and title can be processed by the following modifiers. This processing can be based on a modifier lexicon or some filtering models, which are not limited here.

[0114] S15: The target document is identified using a text generation model to generate identification results for key technologies.

[0115] In this embodiment, a lexicon of attributive terms needs to be constructed. This lexicon refers to a database or corpus containing various attributive terms, which can be used to extract and analyze attributive terms. For example, in this embodiment, the attributive term database may include terms such as "a kind of" and "application". The text generation model can employ various topic generation models, including the alpaca-title-generator-mt0-large model (a neural network model for generating titles), the PromptCULUE model (a pre-trained model for Chinese language understanding and generation tasks), the nlp_mt5_zero-shot-augment_chine-base model (a multi-task learning model based on zero-shot augmentation for Chinese), and the Bert-title-generation-en model (a title generation model based on English). In this embodiment, through experiments on the generality, fluency, and compatibility of the generated titles with both Chinese and English, the alpaca-title-generator-mt0-large model was determined as the text generation model for generating titles for key technologies.

[0116] The following is the comparison process of the experiment:

[0117] The relevant Chinese target document is: Method for increasing osteoblast function. This disclosure provides a method for increasing osteoblast function in a subject, the method comprising systematically administering a population of stem cells and / or their progeny and / or soluble factors derived therefrom to the subject.

[0118] The involved English target documents are: Universal donor stem cells and related methods.Disclosed herein are universal donor stem cells and related methods of their use and production.The universal donor stem cells disclosed herein are useful for overcoming the immune rejection in cell-based transplantationtherapies. cells disclosed herein do not express one or more human leukocyte antigens(eg,HLA-A,HLA-B and / or HLA-C)corresponding to MHC-I and MHC-II human leukocyteantigens,thereby rendering such cells hypoimmunogenic.

[0119] Table 1. Comparison Results of Titles Generated by the Model

[0120]

[0121] The experimental comparison results above show that the titles of the target documents generated by the alpaca-title-generator-mt0-large model have good title summarization, sentence fluency, and compatibility with both Chinese and English.

[0122] The aforementioned S15 mentions "identifying the target document using a text generation model to generate identification results for key technologies." In this embodiment, key technologies can be identified using the following method: First, the target document is input into a text generation model to obtain the title of the target document. Then, the title is processed using a lexicon of adjectives, and the title with the completed adjective processing is determined as the identification result for the key technologies.

[0123] Specifically, the target document needs to be input into a text generation model, which then generates the document's title. The resulting title may contain generic descriptive terms that are not strongly related to the key technology itself and can be considered redundant. Therefore, when identifying the key technology, these descriptive terms can be removed from the title. For example, if the generated title "Application of Pancreatic Stem Cells in the Treatment of Diabetes" contains the term "application" from the descriptive terminology library, then the word "application" in "Application of Pancreatic Stem Cells in the Treatment of Diabetes" will be directly deleted. The final title then becomes "Pancreatic Stem Cells in the Treatment of Diabetes".

[0124] The method for identifying key technologies described above first establishes a lexicon of adjectives. The target document is then input into a text generation model to obtain the title of the target document. The adjectives in the title of the target document are then processed based on the lexicon of adjectives to remove them. This method avoids redundancy in the identified key technologies, making the identification results more concise and accurate.

[0125] This embodiment proposes a key technology identification method. The method first acquires patent data to be identified, and then filters key technology patents that meet key prediction criteria based on this data. These criteria measure and determine the criticality of the patent data. Next, target keywords are determined for the key technology patents using a BERT model. The number of target keywords in the key technology patents is counted, and the abstracts and patent topics corresponding to key technology patents with a number meeting a preset threshold are concatenated to obtain the target document. Finally, key technologies are identified based on the target document using a text generation model. In this way, key technology patents are filtered out using key prediction criteria, target keywords are determined using a BERT model, the number of target keywords in the key technology patents is counted, and key technology patents with a number meeting a preset threshold are identified. The key technology patent obtained at this point is a patent with a certain number of target keywords. This patent has a high probability of containing key technologies. The target document, obtained by concatenating the abstract and subject of the key technology patent, is then identified using a text generation model to determine the key technologies. The entire process is automated, eliminating reliance on human experience. The final identification result meets current requirements, ensuring accuracy and efficiency. In one possible implementation, this embodiment also requires the construction of a stop word list. This list includes words unrelated to the key technologies. "Unrelated words" can be understood as meaningless words or symbols unrelated to the key technologies, such as "this invention," "a," or "comprising."

[0126] When the key technology patent includes an abstract text, the method for "obtaining the subject and keywords of the key technology patent" mentioned in S131 above can be as follows: First, the abstract text of the key technology patent is used as the text to be processed, which includes Chinese text and / or English text. Then, word segmentation and part-of-speech tagging are performed on the Chinese text and / or the English text. Based on the results of word segmentation and part-of-speech tagging, a stop word list is used to filter the Chinese text and / or English text to obtain keyword data. Finally, the subject and keywords of the key technology patent are determined based on the keyword data.

[0127] Specifically, when the obtained key technology patent includes an abstract text, this abstract text can be Chinese text, English text, or a bilingual Chinese-English text. In this embodiment, the abstract text is used as the text to be processed, and word segmentation and part-of-speech tagging are performed on the text to be processed. Word segmentation refers to dividing the abstract text into word units, while part-of-speech tagging is to label each word with its grammatical role in the sentence, such as noun, verb, adjective, etc.

[0128] Specifically, the text to be processed needs to be segmented into words, that is, the abstract text needs to be segmented into words, and the text in the abstract text needs to be divided into words. For example, individual sentences in the abstract text need to be segmented into words. In addition, the text to be processed needs to be annotated with parts of speech, and words with different parts of speech in the abstract text need to be labeled. By annotating the parts of speech, it is easier to remove words such as modifiers that are not related to the key technologies when identifying key technologies later. In the embodiments of this application, natural language processing technology can be used to perform word segmentation and part-of-speech tagging on the text to be processed. For example, pkuseg (a Chinese word segmentation tool) and jieba (a Chinese text processing tool) can be used to perform word segmentation and part-of-speech tagging on Chinese text, and NLTK (Natural Language Toolkit, a toolkit for processing natural language text) can be used to perform word segmentation and part-of-speech tagging on English text.

[0129] After word segmentation and part-of-speech tagging of the abstract text are completed, the results are obtained. Based on these results, a stop word list is used to filter the abstract text, yielding keyword data. The stop word list can be set by those skilled in the art according to the actual situation and application scenario, and is not limited here. Generally, the keyword data obtained by filtering words using the stop word list consists of words not listed in the stop word list and whose part of speech is noun. Based on the keyword data, the subject matter and topic terms of the key technology patent can be determined.

[0130] The method for obtaining topics and keywords described above involves constructing a stop word list, performing word segmentation and part-of-speech tagging on the abstract text of key technology patents, and then using the stop word list to filter the abstract text based on the results of word segmentation and part-of-speech tagging to obtain keyword data. Based on this keyword data, the topic and keywords of the key technology patents are determined. The stop word list provides a clear basis for keyword data filtering, improving the efficiency of keyword data filtering. Furthermore, the pre-processing of the abstract text with word segmentation and part-of-speech tagging facilitates rapid keyword data filtering using the stop word list based on the segmented words and their assigned parts of speech.

[0131] The aforementioned "determining the subject and keywords of key technology patents based on keyword data" can, in one possible implementation, be achieved by using Latent Dirichlet Allocation (LDA) to capture the semantic topics of text. LDA is a topic clustering algorithm that can discover implicit topics from text. Specifically, a method for determining the subject and keywords of key technology patents based on the LDA model can be as follows:

[0132] A1: Based on the keyword data, the first and second distributions are initialized using the LDA model.

[0133] The first distribution indicates the distribution between keyword data and topics, while the second distribution indicates the distribution between topics and terms. Keyword data can be understood as the set of keywords from the acquired key technology patents. Specifically, the first distribution can be understood as the distribution of each keyword data point to each topic, i.e., the probability distribution of each keyword data point to each topic; in other words, the first distribution describes the probability of each keyword data point corresponding to each topic. The second distribution can be understood as the distribution of each topic to each term. In the LDA model, each topic can be represented as a probability distribution with respect to each term, and the topic-term distribution describes the weight or probability of each term within each topic. When keyword data is acquired, the LDA model can initialize the aforementioned first and second distributions.

[0134] A2: Adjust the topic allocation based on Gibbs sampling, and update the first distribution and the second distribution according to the allocated topics.

[0135] The aforementioned keyword data can be understood as a set of keywords from key technology patents. When using the LDA model, the topic distribution of the keyword data can be inferred by iteratively updating the topic of each word in the keyword data through Gibbs sampling. Specifically, Gibbs sampling first randomly initializes the topic assignment of keywords in each keyword data set, and then gradually adjusts the topic assignment of keywords in each keyword data set by calculating conditional probabilities, eventually reaching a stable state. In each iteration, Gibbs sampling considers the topic assignment of the current keyword data set and the topic assignments of keywords in surrounding keyword data sets, updating the topic assignment of keywords in the current keyword data set according to conditional probabilities. This process is iterated repeatedly until a certain number of iterations is reached or the convergence condition is met. By adjusting the topic assignment through Gibbs sampling, the probability distribution of keywords in each keyword data set under each topic can be obtained, thereby inferring the topic distribution of the keyword data set. When the topic assignment is adjusted, the first and second distributions, which are related to the topic, are also updated accordingly.

[0136] A3: The subject of the key technology patent is determined based on the updated first distribution, and the subject terms of the key technology patent are determined based on the updated second distribution.

[0137] When Gibbs sampling reaches a certain number of iterations or meets the convergence condition, the topic of the document corresponding to the keyword data can be extracted by analyzing the first distribution of each keyword data, thus obtaining the key technology patents contained in the topic. Simultaneously, representative words for each topic are identified based on the second distribution, and C preset representative words are selected. These words can serve as the key description of the topic, representing the key technology topic words within the key technology topic.

[0138] The method for determining themes and keywords provided above can combine the LAD model and Gibbs sampling to adjust the first and second distributions in the keyword data. By combining the two distributions, the themes and keywords of key technology patents can be determined. No manual intervention is required in the process. The adjustment of the theme allocation is achieved through Gibbs sampling, which can efficiently determine the themes and keywords of key technology patents.

[0139] The aforementioned "determining the topic and keywords of key technology patents based on keyword data" can, in one possible implementation, be based on BERT-LDA, a topic clustering model that combines the pre-trained models BERT and LDA. When the key technology patent includes abstract text, the method for determining the topic and keywords is as follows: First, obtain keyword data and abstract text. Encode the abstract text using the BERT model to obtain a first feature vector. Then, based on the keyword data, use the first distribution obtained from the LDA model to determine the probability distribution of the topic corresponding to the document. Use this probability distribution as the second feature vector. The first distribution indicates the distribution between the keyword data and the topic. Finally, determine the topic and keywords of the key technology patent based on the first and second feature vectors.

[0140] Specifically, to obtain keyword data and abstract text from key technology patents, a pre-trained BERT model is used to encode the abstract text to obtain a first feature vector. Then, using the keyword data, an LDA model is employed to obtain a probability distribution between the keyword data and a set of topics from the distribution between the keyword data and topics (i.e., the first distribution). This probability distribution can be used as a second feature vector. Finally, the topics and topic names of the key technology patents are determined based on the first and second feature vectors.

[0141] The process of "determining the subject and subject name of key technology patents based on the first and second feature vectors" can be described as follows: The first and second feature vectors are concatenated to obtain a comprehensive representation vector. Since the comprehensive representation vector obtained by direct concatenation may have excessive dimensionality and sparse information, it is not conducive to subject clustering. Therefore, to reduce the clustering difficulty, an autoencoder can be used to learn the low-dimensional latent space representation of this comprehensive representation vector, obtaining a low-dimensional representation vector with condensed information. Then, patents that are similar to each other in terms of subject matter can be assigned to a single cluster. The specific assignment process can be performed using the K-Means clustering algorithm to obtain K subject classes, which in turn yield the key technology patents contained within each subject.

[0142] After obtaining K topic classes, the keyword terms for each topic need to be extracted from the keyword data. Here, Class-TFIDF is used to obtain the keyword terms for each topic class. The keyword data is a set of keywords for key technology patents within each topic class. Class-TFIDF (Class-Term Frequency-Inverse Document Frequency) is a technique used for information retrieval and text mining. It measures the importance of a word in a specific category by calculating term frequency (TF) and inverse document frequency (IDF). The formula for calculating the importance of word t in topic i is:

[0143]

[0144] Among them, t i The number of times word t appears in category i, w i t is the total number of words in category i, m is the total number of keyword data (number of categories), and n is the number of keyword data containing word t. Finally, the top C words with preset importance for each topic category are obtained according to Class-TFIDF and used as topic keywords. The value of C can be set by those skilled in the art according to actual needs and application scenarios, and is not limited here. For example, C can be 5.

[0145] The method described above for determining topics and keywords can determine the corresponding first and second feature vectors based on the acquired keyword data and abstract text, using BERT and LDA models. The topics and keywords of key technology patents are then determined based on these first and second feature vectors. The process involves determining the topics and keywords in key technology patents in vector form through topic clustering and importance quantification. No manual intervention is required, thus improving the efficiency and accuracy of topic and keyword identification for key technology patents.

[0146] Assuming there are 5 topics, the identification results obtained by using the key technology identification method in the embodiments of this application are shown in Tables 2 to 8:

[0147] Table 2. Identification results of Chinese data keywords

[0148]

[0149]

[0150] Table 3. Identification results of English data keywords

[0151]

[0152] Table 4. Recognition Results of Chinese Data Subject Names

[0153]

[0154]

[0155] Table 5. Recognition Results of English Data Subject Names

[0156] Subject number English data subject name 1 Universal Donor Stem Cells 2 Promoting Bone Marrow Regeneration Control 3 Cell-Specific Protein Expression 4 Prostaglandin E2 for Muscle Regeneration 5 Pharmaceutical Compositions for Human Embryonic Stem Cells

[0157] Table 6. Identification results of Chinese data topic summaries:

[0158]

[0159]

[0160] Table 7. Results of English Data Topic Summary Recognition:

[0161]

[0162]

[0163] Table 8. Identification Results of Some Key Technology Patents

[0164] Serial Number Chinese data key technology patents English data key technology patents 1 CN108473961B US10705091B2 2 CN109735497B US10301593B2 3 CN103898045B US10301594B1 4 CN111440696B US10669528B2 5 CN111836888B US10925902B2

[0165] Figure 3 This is a schematic diagram of the structure of a key technology identification device provided in an embodiment of this application, such as... Figure 3 As shown, a key technology identification device specifically includes: an acquisition module 100, a filtering module 200, a determination module 300, a statistics module 400, and an identification module 500;

[0166] The acquisition module 100 is used to acquire the patent data to be identified;

[0167] The filtering module 200 is used to filter key technology patents that meet the key prediction conditions based on the patent data to be identified. The key prediction conditions are used to determine the keyness of the patent data to be identified.

[0168] The determining module 300 is used to determine the target keywords of the key technology patent based on the BERT model;

[0169] The statistics module 400 is used to count the number of target keywords in the key technology patents, and to combine the abstracts and patent topics of key technology patents whose numbers meet a preset threshold to obtain a target document.

[0170] The recognition module 500 is used to recognize the target document through a text generation model and generate recognition results for key technologies.

[0171] In one possible implementation, the key prediction conditions include a score of the patent data to be identified, and the apparatus further includes a description generation module, which is used to:

[0172] Determine the scores and rank the key technology patents;

[0173] The first k key technology patents in the sorted list are taken as target patents, where k>1;

[0174] Extract the abstract from the target patent;

[0175] Based on the extracted summary, a description of the key technology is generated using a large language model (LLM).

[0176] In one possible implementation, the determining module 300 is configured to: obtain the subject and keywords of the key technology patent, wherein the subject is used to indicate the technical direction of the key technology patent, and the keywords are used to indicate the technical application of the key technology patent;

[0177] The topic vector is obtained by encoding the topic based on the BERT model, and the topic word is obtained by encoding the topic word.

[0178] Calculate the cosine similarity between the topic vector and the topic word vector, and determine the topic words whose cosine similarity meets the preset conditions as target topic words.

[0179] In one possible implementation, the apparatus further includes a vocabulary building module, the vocabulary building module being used for:

[0180] Construct a stop word list, which includes words that are not related to the key technology;

[0181] The key technology patent includes an abstract text, and the determining module 300 is used for:

[0182] The abstract text of the key technology patent is used as the text to be processed, which includes Chinese text and / or English text.

[0183] The Chinese text and / or the English text are segmented and tagged with parts of speech.

[0184] Based on the results of the word segmentation and part-of-speech tagging, the Chinese text and / or the English text are filtered using the stop word list to obtain keyword data;

[0185] The subject and keywords of the key technology patent are determined based on the keyword data.

[0186] In one possible implementation, the determining module 300 is used to:

[0187] Based on the keyword data, a first distribution and a second distribution are initialized by the LDA model. The first distribution indicates the distribution between the keyword data and the topic, and the second distribution indicates the distribution between the topic and the term.

[0188] The topic allocation is adjusted based on Gibbs sampling, and the first distribution and the second distribution are updated according to the allocated topics;

[0189] The subject of the key technology patent is determined based on the updated first distribution, and the subject terms of the key technology patent are determined based on the updated second distribution.

[0190] In one possible implementation, the determining module 300 is used to:

[0191] Obtain the keyword data and the summary text;

[0192] The first feature vector is obtained by encoding the summary text based on the BERT model;

[0193] Based on the keyword data, the first distribution obtained by using the LDA model is used to determine the probability distribution of the topic corresponding to the document, and the probability distribution is used as the second feature vector. The first distribution indicates the distribution between the keyword data and the topic.

[0194] The subject and keywords of the key technology patent are determined based on the first feature vector and the second feature vector.

[0195] In one possible implementation, the apparatus further includes a lexicon building module, the lexicon building module being used for:

[0196] Establish a lexicon of attributive terms;

[0197] The identification module 500 is used for:

[0198] The target document is input into the text generation model to obtain the title of the target document;

[0199] The title is processed by the adjective lexicon, and the title that has undergone the adjective processing is determined as the identification result of the key technology.

[0200] This embodiment proposes a key technology identification device, which includes an acquisition module, a filtering module, a determination module, a statistics module, and an identification module. The acquisition module acquires patent data to be identified; the filtering module filters key technology patents that meet key prediction conditions based on the patent data, whereby the key prediction conditions determine the degree of keyness of the patent data; the determination module determines target keywords for the key technology patents based on a BERT model; the statistics module counts the number of target keywords in the key technology patents, and concatenates the abstracts and patent topics corresponding to key technology patents with a number meeting a preset threshold to obtain a target document; the identification module identifies the target document using a text generation model, generating key technology identification results. Thus, key technology patents are filtered out using key prediction conditions, target keywords in the key technology patents are determined using a BERT model, the number of target keywords in the key technology patents is counted, and key technology patents with a number meeting a preset threshold are identified. The key technology patent obtained at this time is a patent with a certain number of target subject words. This patent is highly likely to contain key technologies. The target document is obtained by splicing the abstract and subject of the key technology patent. The key technology can be obtained by using a text generation model to identify the target document. The whole process does not rely on human experience, but adopts an automated method to identify key technologies. The final key technology identification result can meet the current identification needs, ensure the accuracy of key technology identification, and improve the efficiency of key technology identification.

[0201] Figure 4 A schematic diagram of another key technology identification device provided in the embodiments of this application is shown below. Figure 4 The device may include: a processor 201, a memory 202, an implementation program 203 for the key technology identification method, a communication interface 204, and a communication bus 205. The memory 202 is used to store the implementation program 203 for the key technology identification method. The computer program is loaded and executed by the processor 201 to implement the relevant steps of the aforementioned key technology identification method based on artificial intelligence.

[0202] The processor 201 may be a high-performance microprocessor, central processing unit (CPU) or graphics processing unit (GPU) or a processor specifically designed for machine learning tasks, with high-speed computing power and multi-threaded processing capabilities, for executing key technology identification methods based on artificial intelligence.

[0203] The memory 202 includes, but is not limited to, random access memory (RAM), read-only memory (ROM), and persistent memory (e.g., flash memory or hard disk drive). The memory 202 is used to store the implementation program 203 of the key technology identification method, as well as the data generated by the processor 201 executing the implementation program 203 of the key technology identification method.

[0204] The implementation program 203 for the key technology identification method is a computer program responsible for implementing the key technology identification method mentioned in the foregoing embodiments.

[0205] The communication interface 204 can use wired or wireless communication methods, such as Ethernet, serial port, Bluetooth, Wi-Fi, etc., to realize data interaction and communication with external devices or other systems. For example, data to be identified can be obtained through the communication interface 204.

[0206] The communication bus 205 is a pathway connecting various components, responsible for transmitting data and control signals between the processor 201, memory 202, implementation program 203 of key technology identification method, and communication interface 204.

[0207] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatuses and methods according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0208] This application also provides corresponding devices and computer-readable storage media for implementing the solutions provided in this application.

[0209] The device includes a memory and a processor. The memory stores instructions or code, and the processor executes the instructions or code to enable the device to perform a key technology identification method according to any embodiment of this application.

[0210] In practical applications, the computer-readable storage medium can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0211] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0212] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0213] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0214] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0215] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying key technologies, characterized in that, The method includes: Obtain the patent data to be identified; Obtain feature data from the patent data to be identified; According to the preset feature data scoring criteria, the feature data in the patent data to be identified is scored to obtain the score of the patent data to be identified. Based on the score of the patent data to be identified, it is determined whether the patent data to be identified is a key technology patent; Obtain the subject and keywords of the key technology patent, wherein the subject is used to indicate the technical direction of the key technology patent and the keywords are used to indicate the technical application of the key technology patent; The topic is encoded using the BERT model to obtain a topic vector, and the topic words are encoded to obtain a topic word vector. Calculate the cosine similarity between the topic vector and the topic word vector, and determine the topic words whose cosine similarity meets the preset conditions as target topic words; The number of target keywords in the key technology patents is counted, and the abstracts and patent topics corresponding to the key technology patents whose numbers meet a preset threshold are concatenated to obtain the target document; The target document is identified using a text generation model to generate identification results for key technologies.

2. The method according to claim 1, characterized in that, The method further includes: Construct a stop word list, which includes words that are not related to the key technology; The key technology patent includes an abstract text, and obtaining the subject and keywords of the key technology patent includes: The abstract text of the key technology patent is used as the text to be processed, which includes Chinese text and / or English text. The Chinese text and / or the English text are segmented and tagged with parts of speech. Based on the results of the word segmentation and part-of-speech tagging, the Chinese text and / or the English text are filtered using the stop word list to obtain keyword data; The subject and keywords of the key technology patent are determined based on the keyword data.

3. The method according to claim 2, characterized in that, The process of determining the subject and keywords of the key technology patent based on the keyword data includes: Based on the keyword data, a first distribution and a second distribution are initialized by the LDA model. The first distribution indicates the distribution between the keyword data and the topic, and the second distribution indicates the distribution between the topic and the term. The topic allocation is adjusted based on Gibbs sampling, and the first distribution and the second distribution are updated according to the allocated topics; The subject of the key technology patent is determined based on the updated first distribution, and the subject terms of the key technology patent are determined based on the updated second distribution.

4. The method according to claim 2, characterized in that, The process of determining the subject and keywords of the key technology patent based on the keyword data includes: Obtain the keyword data and the summary text; The first feature vector is obtained by encoding the summary text based on the BERT model; Based on the keyword data, the first distribution obtained by using the LDA model is used to determine the probability distribution of the topic corresponding to the document, and the probability distribution is used as the second feature vector. The first distribution indicates the distribution between the keyword data and the topic. The subject and keywords of the key technology patent are determined based on the first feature vector and the second feature vector.

5. The method according to claim 1, characterized in that, The method further includes: Establish a lexicon of attributive terms; The step of recognizing the target document using a text generation model to generate recognition results for key technologies includes: The target document is input into the text generation model to obtain the title of the target document; The title is processed by the adjective lexicon, and the title that has undergone the adjective processing is determined as the identification result of the key technology.

6. A key technology identification device, characterized in that, The device includes: an acquisition module, a filtering module, a determination module, a statistics module, and an identification module; The acquisition module is used to acquire the patent data to be identified; The filtering module is used to acquire feature data from the patent data to be identified; score the feature data in the patent data to be identified according to a preset feature data scoring standard to obtain a score for the patent data to be identified; and determine whether the patent data to be identified is a key technology patent based on the score of the patent data to be identified. The determining module is used to obtain the topic and keywords of the key technology patent, wherein the topic indicates the technical direction of the key technology patent and the keywords indicate the technical application of the key technology patent; the topic is encoded into a topic vector based on the BERT model, and the keywords are encoded into a keyword vector; the cosine similarity between the topic vector and the keyword vector is calculated, and the keywords whose cosine similarity meets the preset conditions are determined as target keywords; The statistics module is used to count the number of target keywords in the key technology patents, and to combine the abstracts and patent topics of key technology patents whose counts meet a preset threshold to obtain the target document. The recognition module is used to recognize the target document through a text generation model and generate recognition results for key technologies.

7. An electronic device, characterized in that, The device includes: a processor, a memory, and a system bus; The processor and the memory are connected via the system bus; The memory is used to store one or more programs, the one or more programs including instructions, which, when executed by the processor, cause the processor to perform the key technology identification method according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an implementation program for the key technology identification method, which, when executed by a processor, implements the steps of the method as described in any one of claims 1-5.

Images