A patent achievement transformation recommendation method and device based on self-learning strategy

Abstract

A patent achievement transformation recommendation method and device based on a self-learning strategy, the method comprising: obtaining a patent data set; extracting one patent from each enterprise's patents to form a training set; using multiple methods to calculate the similarity between the abstract text of each patent in the training set and the abstract text of the remaining patents in the data set; according to the similarity calculated by a preset method, arranging the remaining patents in the data set in descending order to obtain a sample group consisting of the first m patents, and the sample feature of each sample is a plurality of similarities calculated by multiple methods; training a support vector machine sorting model based on the sample group; obtaining patent texts to be transformed, inputting the patent texts to be transformed into the support vector machine sorting model, and obtaining recommended enterprise results; the method does not require manual data annotation, and uses multiple methods to calculate similarities, and the recommendation result has a high accuracy rate.

Description

Technical Field

[0001] The present invention relates to the technical field of intellectual property recommendation algorithms. Background Art

[0002] Patent conversion refers to the conversion of patented technology into specific commercial products or processes to realize its economic value and social benefits. The current patent conversion rate is low, and most patents cannot be directly converted into real productivity. The reasons include poor dissemination of patent information and the inability of universities with many invention patents to find suitable enterprises to cooperate and transform their results in a timely manner.

[0003] Many scholars have studied the transformation and technology transfer of university patents. First, machine learning algorithms are used to measure the optimal transformation prediction scheme, identify and predict the influencing factors of patent achievement transformation; or the AdaBoost algorithm is used to predict the probability of university patent transformation. However, the above two schemes only use patent data in one field, and the patent samples are relatively small; there are also studies that find that the transformation mode of university patent achievements, patent technology characteristics, patent maturity, etc. affect the transformation mode of university patents; or sort out the typical mode and evolution path of university patent transformation, and through the comparative analysis between typical modes and the characteristic analysis of mode evolution path, the influencing mechanism and model of university patent transformation mode selection are constructed, and the conclusion that cooperative innovation can promote patent transformation is drawn, but it only stays at the stage of suggestions and concepts; there are also related studies that use the internal text features of patents to qualitatively estimate the patent value, and then use the external transformation conditions of patents to explore the opportunities for patent transformation, but no specific enterprise is given to transform the patent.

[0004] In summary, the existing patent achievement transformation methods require manual data annotation, or only analyze the factors that affect the transformation, without analyzing how to improve the transformation rate and recommend suitable companies for cooperation to universities. However, in actual operation, users need to recommend suitable companies based on patent achievements. At the same time, the cost of manually annotating a large amount of patent data is too high, and it is difficult to implement.

[0005] Therefore, how to provide a feasible method for recommending the transformation of patent achievements has become a technical problem that needs to be urgently solved in this field. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides a patent achievement transformation recommendation method and device based on a self-learning strategy. The method is based on the enterprise information contained in the patent text, does not require manual annotation of data, and only requires the patent collection of the enterprise. It can assist universities in making decisions on which enterprise to cooperate with, and adopts multiple methods to calculate similarity, and the recommendation results are highly accurate.

[0007] Based on the same inventive concept, the present invention has four independent technical solutions:

[0008] 1. A patent achievement transformation recommendation method based on self-learning strategy, including:

[0009] Acquire a patent text dataset, where the patent text dataset comes from different companies;

[0010] One patent is extracted from each company’s patents to form a training set;

[0011] Using a variety of methods to calculate the similarity between the abstract text of each patent in the training set and the abstract text of the remaining patents in the data set;

[0012] For each patent in the training set, according to the similarity calculated by a preset method, the remaining patents in the data set are arranged in descending order to obtain a sample group consisting of the first m patents, and the sample feature of each sample is a plurality of similarities calculated by a plurality of methods;

[0013] Training a support vector machine ranking model based on the sample group;

[0014] The patent text to be converted is obtained, and the patent text to be converted is input into the support vector machine sorting model to obtain recommended enterprise results.

[0015] Furthermore, the multiple methods include: cosine similarity, Dice coefficient, fuzzywuzzy algorithm, bert model algorithm, Bleu algorithm and edit distance algorithm.

[0016] Furthermore, the preset method is the Bleu algorithm.

[0017] Furthermore, before training the support vector machine ranking model based on the sample group, the following steps of marking the sample relevance are included:

[0018] In the sample group, the correlation of patent text A with the highest similarity to the patent samples from the same enterprise in the training set is set to 4, the correlation of other patent texts from the same enterprise is set to 3, the correlation of patents in different enterprises with higher similarity than patent text A is set to 2, and the correlation of patents in different enterprises with lower similarity than patent text A is set to 1.

[0019] Further, training a support vector machine ranking model based on the sample group includes:

[0020] Patent texts with a relevance of 4 are used as test samples, and patent texts with a relevance of 1-3 are used as training samples.

[0021] Furthermore, during training, the C value of the support vector machine sorting model is 500, and the kernel function is a polynomial kernel function.

[0022] Furthermore, during training, the C value of the support vector machine sorting model is 100, and the kernel function is a linear kernel function.

[0023] 2. A patent achievement transformation recommendation device based on a self-learning strategy, comprising:

[0024] A data acquisition module, used to acquire patent text datasets, where the patent text datasets come from different enterprises;

[0025] The training set building module is used to extract one patent from each enterprise’s patents to form a training set;

[0026] A similarity calculation module, used to calculate the similarity between the abstract text of each patent in the training set and the abstract text of the remaining patents in the data set using multiple methods;

[0027] A sample establishment module is used to, for each patent in the training set, arrange the remaining patents in the data set in descending order according to the similarity calculated by a preset method to obtain a sample group consisting of the first m patents, wherein the sample feature of each sample is a plurality of similarities calculated by a plurality of methods;

[0028] A training module, used for training a support vector machine ranking model based on the sample group;

[0029] The recommendation module is used to obtain the patent text to be converted, input the patent text to be converted into the support vector machine ranking model, and obtain the recommended enterprise results.

[0030] 3. A computer-readable storage medium storing a computer program, wherein the computer program implements the above method when executed by a processor.

[0031] 4. An electronic device, comprising a processor and a storage device, wherein the storage device stores a plurality of instructions, and the processor is used to read the plurality of instructions in the storage device and execute the above method.

[0032] The patent achievement transformation recommendation method and device based on the self-learning strategy provided by the present invention have at least the following beneficial effects:

[0033] (1) This method uses a self-learning strategy to help universities find enterprises that are suitable for transforming invention patents and putting them into practical application, promote cooperation between universities and enterprises, and improve the conversion rate of results. The patent results transformation model is obtained by using the self-learning method. There is no need to manually annotate data. Only the enterprise patent data set is required, which is low-cost and easy to operate.

[0034] (2) After experimental comparison and verification training, it can be seen that this method uses multiple methods to calculate similarity, which has an improved performance compared to a single similarity method. The patent achievement transformation prediction model achieves better performance when the linear kernel function C value is 100, and the accuracy rate of predicting enterprises suitable for transformation achievements reaches 69.4%. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0036] Figure 1 A flowchart of an embodiment of a patent achievement transformation recommendation method based on a self-learning strategy provided by the present invention;

[0037] Figure 2 A schematic diagram of calculating similarity based on the BERT model;

[0038] Figure 3 A flowchart of the experimental verification process of the patent achievement transformation recommendation method based on the self-learning strategy provided by the present invention. DETAILED DESCRIPTION

[0039] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0040] It should be understood that when used in this specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0041] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, the singular forms "a", "an" and "the" are intended to include plural forms unless the context clearly indicates otherwise.

[0042] The following is a clear and complete description of the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only 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 creative work are within the scope of protection of this application.

[0043] In the following description, many specific details are set forth to facilitate a full understanding of the present application, but the present application may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.

[0044] Embodiment 1:

[0045] See also Figure 1 In some embodiments, a patent achievement transformation recommendation method based on a self-learning strategy is provided, comprising:

[0046] S1. Obtain a patent text dataset, where the patent text dataset comes from different companies;

[0047] S2, extract one patent from each enterprise's patents to form a training set;

[0048] S3, using a variety of methods to calculate the similarity between the abstract text of each patent in the training set and the abstract text of the remaining patents in the data set;

[0049] S4. For each patent in the training set, according to the similarity calculated by a preset method, the remaining patents in the data set are arranged in descending order to obtain a sample group consisting of the first m patents, and the sample feature of each sample is a plurality of similarities calculated by a plurality of methods;

[0050] S5, training a support vector machine ranking model based on the sample group;

[0051] S6. Obtain the patent text to be converted, input the patent text to be converted into the support vector machine ranking model, and obtain recommended enterprise results.

[0052] Specifically, in step S1, patent data from different companies are obtained, and L patents from L companies are randomly selected as training set patents. One patent is selected from each company, and used to compare text similarity with the abstracts of all patent samples in subsequent steps. Then, from the remaining data, one patent is selected from each company, and they are combined together as a test set for subsequent testing of the accuracy of model recommendation.

[0053] In step S3, the multiple methods include: cosine similarity, Dice coefficient, fuzzywuzzy algorithm, bert model algorithm, Bleu algorithm and edit distance algorithm.

[0054] Specifically, six similarity methods are used: text2vec similarity function, Dice coefficient, ratio() function of fuzzywuzzy library, bert model, bleu parameter, edit distance, to calculate the similarity between L test patents and all patent abstract texts from different companies. We assume that the company with the highest similarity is the company that best matches the test patent, and calculate the accuracy of patent and company matching using the six methods. The implementation principles of the six similarity methods are:

[0055] (1) Cosine similarity method, the calculation formula is as follows:

[0056]

[0057] (2) Dice coefficient, the calculation steps are as follows:

[0058] First use jieba to segment words, and then calculate the Dice coefficient of the two text word sets. The Dice coefficient calculation formula is as follows:

[0059] Dice(s1,s2)=2*comm(s1,s2) / (leng(s1)+leng(s2));

[0060] (3) fuzzywuzzy algorithm:

[0061] (1-number of revisions / total number of words in two sentences (without deduplication))*100;

[0062] (4) See Figure 2 , Figure 2 The process of calculating similarity based on the BERT model is shown;

[0063] (5) Bleu calculation similarity:

[0064] First, calculate N-gram. The N-gram calculation formula is as follows:

[0065]

[0066] Then introduce the length penalty factor, the length penalty factor formula is as follows:

[0067]

[0068] Finally, the evaluation formula is obtained:

[0069]

[0070] (6) Calculate the edit distance: Using the idea of ​​dynamic programming, the calculation formula is as follows:

[0071]

[0072] As a preferred implementation, the preset method is the Bleu algorithm.

[0073] In step S4, the data set is first divided, and the similarity calculation results are divided into L groups according to the patents in the training set. The data in one group includes the results of comparing the abstract similarity of all patents and test patents using six methods, and the first m similar patents sorted in descending order by the Bleu algorithm are taken. The Bleu algorithm is selected as the preset algorithm because the Bleu algorithm has the highest accuracy and the best effect in calculating similarity. Selecting this algorithm as the main sorting algorithm plays an important role in recommending accurate companies. The feature of each sample is the abstract similarity calculated by different methods.

[0074] Step S5, before training the support vector machine ranking model based on the sample group, includes the following steps of marking the sample relevance:

[0075] In the sample group, the relevance of patent text A, which has the highest similarity with the test set text in the patent samples from the same enterprise, is set to 4, the relevance of other patent texts from the same enterprise is set to 3, the relevance of patents in different enterprises with higher similarity than patent text A is set to 2, and the relevance of patents in different enterprises with lower similarity than patent text A is set to 1.

[0076] As a preferred implementation method, each set of samples includes a positive example set and a negative example set. The positive example set is a set of patents from the same company as the test patent, which is subdivided into the most similar patent A and other patents from the same company except A; the negative example set is a set of patents from different companies from the test patent, which is subdivided into patents with higher similarity than A but different companies and patents with lower similarity than A but different companies.

[0077] In step S5, training a support vector machine ranking model based on the sample group includes:

[0078] The samples with a correlation of 4 are used as test samples, and the samples with a correlation of 1-3 are used as training samples.

[0079] As a preferred implementation, during training, the C value of the support vector machine sorting model is 500, and the kernel function is a polynomial kernel function. The training set accuracy performs well under this parameter setting condition.

[0080] As a preferred implementation, during training, the C value of the support vector machine ranking model is 100, and the kernel function is a linear kernel function. The test set accuracy performs well under this parameter setting condition.

[0081] This method uses the entire patent collection of each enterprise, randomly selects a patent as a learning sample, obtains the patent feature matrix by calculating the abstract similarity, divides it into positive and negative examples for training, and adjusts the training parameters to improve the training effect. Use the trained model to sort the test set patents, input a patent to get the most suitable enterprise for achievement transformation, and also list the enterprises that are more suitable for transformation.

[0082] See also Figure 3 The experimental process shown in the figure is further illustrated by a set of specific experiments to illustrate the beneficial effects of the method provided in this embodiment:

[0083] 1. Select the experimental data range: 19,595 invention patents of 7,336 enterprises in a certain place were screened. There were 99 enterprises with more than 20 patents, and 1.3% of enterprises had 34% of patents. Therefore, 6,646 patents corresponding to 99 enterprises were selected for the experiment. Two patents were selected for each enterprise, one was added to the training set and the other was added to the test set, for a total of 198 patents as the training set patents and the test set patents.

[0084] The 198 patent numbers are arranged in ascending order according to the company numbers, such as patents No. 9 and 10 come from company No. 5.

[0085] 2. Calculate the similarity of patent texts: Calculate whether the most similar patents calculated by the 6 similarity methods for the 99 training set patents are in the same company as the test patent. The counting results are as follows. Correctness = the number of companies where the most similar patents and the test patents are located / the total number of test patents. If one of the 6 methods is calculated correctly, the result is correct. The calculation results of the six methods are shown in Table 1. It can be verified from Table 1 that the similarity accuracy obtained by the Bleu algorithm is the highest, and the accuracy of the final union result can be raised to 82.8%.

[0086] Table 1

[0087]

[0088] 3. Dataset settings:

[0089] (1) Divide the test patent data into four parts, perform cross-validation, and increase the training set validation. Observe whether the accuracy rate is improved after increasing the amount of training data. Each sample in the data set file has six features, corresponding to the calculation results of the six similarity methods.

[0090] (2) Setting up positive and negative examples for each tested patent

[0091] The top 30 most similar patents calculated by the Bleu method are listed, and those in the same company are set as positive examples, and those in different companies are set as negative examples. Set a relevance feature value for each sample. The relevance of the most similar patent A in the same company is set to 4, the relevance of other patents in the same company is set to 3, the relevance of patents in different companies with higher similarity than A is set to 2, and the relevance of patents in different companies with lower similarity than a is set to 1. The sample features are the calculation results of the six similarities.

[0092] The data set is then used for training and testing respectively. The division method of this experiment is: keep the test sample unchanged and gradually increase the data volume of the training sample. The patent achievement transformation prediction model is trained according to the input training sample data, and the training parameters are adjusted to optimize the effect. The training parameters include the type of kernel function and the C value.

[0093] Specifically, data set 4 is always used as a test sample, and the model is trained when the training samples are 1, 1+2, 1+3, and 1+2+3, respectively.

[0094] 4. Experimental parameter settings

[0095] (1) C value: The C value is 1, 2, and the default value is 3, 4, 5, 10, 15, 20, 30, 40, 50, 100, 200, 300, 400, 500, 1000.

[0096] (2) There are four types of kernel functions: linear function, polynomial function, radial basis function, and sigmoid tanh (sa*b+c).

[0097] (3) Take k=1, L=198, N=99, m=30.

[0098] 5. Experimental Procedure

[0099] During the experiment, the input is the patent to be recommended and the patent information involved in all candidate companies, and the output is the company where the most similar patent is located as the recommendation result.

[0100] Step 1: Obtain summary prediction scores. Use the trained patent achievement transformation prediction model to predict the prediction score of each sample in the test set. The higher the score, the more similar the sample and the test patent are according to the current self-learning strategy method.

[0101] Step 2: Take the k patent samples with the highest prediction scores in each group. If one of the k patent samples has a correct prediction result, that is, it comes from the same company as the test patent, then the calculation result of this group is correct. Count the results of L groups. The more patents from the same company, the higher the accuracy of the ranking results.

[0102] Input data: Patent achievement transformation prediction model predicts the similarity ranking of patents;

[0103] Output data: the number of correctly matched enterprises.

[0104] Step 3: Compare whether the calculation results of the patent achievement transformation prediction model are more accurate than those of a single similarity method, compare the accuracy of the patent achievement transformation prediction model under different C values ​​and kernel function types, observe the accuracy of the training set and test set, and analyze whether the model effect is improved after adding training data.

[0105] 6. Experimental Results

[0106] The training set accuracy results are shown in Table 2. The training set accuracy will increase when the amount of training data increases, that is, the more training data, the better the effect. The performance is better when using linear kernel function and polynomial kernel function. The performance is best when the C value is 500 and the kernel function is polynomial kernel function. The accuracy rate of matching enterprises can reach up to 76%.

[0107] Table 2

[0108]

[0109] The test set accuracy results are shown in Table 3. The test set accuracy generally improves when the amount of training data increases, that is, the more training data, the better the effect. The performance is better when using linear kernel functions and polynomial kernel functions. The performance is best when the C value is 100 and the kernel function is a linear kernel function. The accuracy of matching enterprises can reach up to 69.4%.

[0110] Table 3

[0111]

[0112] Embodiment 2:

[0113] In some embodiments, a patent achievement transformation recommendation device based on a self-learning strategy is provided, comprising:

[0114] A data acquisition module, used to acquire patent text datasets, where the patent text datasets come from different enterprises;

[0115] The training set building module is used to extract one patent from each enterprise’s patents to form a training set;

[0116] A similarity calculation module, used to calculate the similarity between the abstract text of each patent in the training set and the abstract text of the remaining patents in the data set using multiple methods;

[0117] A sample establishment module is used to, for each patent in the training set, arrange the remaining patents in the data set in descending order according to the similarity calculated by a preset method to obtain a sample group consisting of the first m patents, wherein the sample feature of each sample is a plurality of similarities calculated by a plurality of methods;

[0118] A training module, used for training a support vector machine ranking model based on the sample group;

[0119] The recommendation module is used to obtain the patent text to be converted, input the patent text to be converted into the support vector machine ranking model, and obtain the recommended enterprise results.

[0120] Embodiment three:

[0121] In some embodiments, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0122] The patent achievement transformation recommendation method, device, storage medium and electronic device based on the self-learning strategy provided in the above embodiments adopt the self-learning strategy to help colleges and universities find enterprises suitable for transformation and practical application of invention patents, promote cooperation between colleges and enterprises, and improve the achievement transformation rate. The patent achievement transformation model is obtained by the self-learning method, without manual data annotation, only the enterprise patent data set is needed, the cost is low and the operation is convenient; after experimental comparison and verification training, it can be known that this method uses multiple methods to calculate similarity, which has an improved effect compared with the performance of a single similarity method. The patent achievement transformation prediction model uses a linear kernel function C value of 100 to achieve better performance, and the accuracy rate of predicting enterprises suitable for transformation achievements reaches 69.4%.

[0123] It should be understood that in the embodiments of the present application, the processor referred to may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0124] The memory may include a read-only memory, a flash memory, and a random access memory, and provides instructions and data to the processor. A part or all of the memory may also include a nonvolatile random access memory.

[0125] It should be understood that the above-mentioned integrated module / unit, if implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on such an understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The above-mentioned computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the above-mentioned computer program includes computer program code, and the above-mentioned computer program code can be in source code form, object code form, executable file or some intermediate form. The above-mentioned computer-readable medium may include: any entity or device capable of carrying the above-mentioned computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the above-mentioned computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

[0126] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. 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 application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0127] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0128] In the embodiments provided in the present application, it should be understood that the disclosed apparatus / terminal equipment and method can be implemented in other ways. For example, the apparatus / equipment embodiments described above are only schematic, for example, the division of the above modules or units is only a logical function division, and in actual implementation, other division methods can be used, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0129] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A patent achievement transformation recommendation method based on self-learning strategy, characterized in that: include: Acquire a patent text dataset, where the patent text dataset comes from different companies; Extract two patents from each company, add one to the training set and one to the test set; Using a variety of methods to calculate the similarity between the abstract text of each patent in the training set and the abstract text of the remaining patents in the data set; For each patent in the training set, according to the similarity calculated by a preset method, the remaining patents in the data set are arranged in descending order to obtain a sample group consisting of the first m patents, and the sample feature of each sample is a plurality of similarities calculated by a plurality of methods; Training a support vector machine ranking model based on the sample group; Obtaining the patent text to be converted, inputting the patent text to be converted into the support vector machine ranking model, and obtaining the recommended enterprise results; Before training the support vector machine ranking model based on the sample group, the following steps of marking the sample relevance are included: In the sample group, the relevance of the patent text A with the highest similarity to the patents in the training set from the same enterprise is set to 4, the relevance of other patent texts from the same enterprise is set to 3, the relevance of patents in different enterprises with higher similarity than patent text A is set to 2, and the relevance of patents in different enterprises with lower similarity than patent text A is set to 1; The calculation method of the patent text A with the highest similarity is: Calculate whether the most similar patent calculated by the six similarity methods for the training set patents is in the same enterprise as the test set patent. The accuracy rate = the number of companies where the most similar patents and the test set patents are in the same enterprise / the total number of test set patents. The similarity calculated by the similarity method with the highest accuracy rate is selected as the similarity of patent text A.

2. The method according to claim 1, characterized in that: The multiple methods include: cosine similarity, Dice coefficient, fuzzywuzzy algorithm, bert model algorithm, Bleu algorithm and edit distance algorithm.

3. The method according to claim 1, characterized in that Training a support vector machine ranking model based on the sample group includes: Patent texts with a relevance of 4 are used as test samples, and patent texts with a relevance of 1-3 are used as training samples.

4. The method according to claim 1, characterized in that During training, the C value of the support vector machine sorting model is 500, and the kernel function is a polynomial kernel function.

5. The method according to claim 1, characterized in that During training, the C value of the support vector machine sorting model is 100, and the kernel function is a linear kernel function.

6. A patent achievement transformation recommendation device based on self-learning strategy, characterized in that: include: A data acquisition module, used to acquire patent text datasets, where the patent text datasets come from different enterprises; The training set building module is used to extract two patents from each enterprise's patents, one to be added to the training set and the other to be added to the test set; A similarity calculation module, used to calculate the similarity between the abstract text of each patent in the training set and the abstract text of the remaining patents in the data set using multiple methods; A sample establishment module is used to, for each patent in the training set, arrange the remaining patents in the data set in descending order according to the similarity calculated by a preset method to obtain a sample group consisting of the first m patents, wherein the sample feature of each sample is a plurality of similarities calculated by a plurality of methods; A training module, used for training a support vector machine ranking model based on the sample group; A recommendation module, used to obtain the patent text to be converted, input the patent text to be converted into the support vector machine ranking model, and obtain the recommended enterprise results; Before training the support vector machine ranking model based on the sample group, the following steps of marking the sample relevance are included: In the sample group, the relevance of the patent text A with the highest similarity to the patents in the training set from the same enterprise is set to 4, the relevance of other patent texts from the same enterprise is set to 3, the relevance of patents in different enterprises with higher similarity than patent text A is set to 2, and the relevance of patents in different enterprises with lower similarity than patent text A is set to 1; The calculation method of the patent text A with the highest similarity is: Calculate whether the most similar patent calculated by the six similarity methods for the training set patents is in the same enterprise as the test set patent. The accuracy rate = the number of companies where the most similar patents and the test set patents are in the same enterprise / the total number of test set patents. The similarity calculated by the similarity method with the highest accuracy rate is selected as the similarity of patent text A.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

8. An electronic device comprising a processor and a storage device, characterized in that: The storage device stores a plurality of instructions, and the processor is used to read the plurality of instructions in the storage device and execute the method according to any one of claims 1 to 5.

Images