Patent novelty evaluation method, device, equipment and readable storage medium

Abstract

The present invention provides a patent novelty evaluation method, device, equipment and readable storage medium. The patent novelty evaluation method includes: determining the corresponding fruit fly counting bloom filter based on the category of the patent to be evaluated; inputting the text of the patent to be evaluated into the language model to obtain a first patent vector; reconstructing the first patent vector based on the fruit fly neural network algorithm to obtain a second patent vector; and calculating the novelty score of the patent to be evaluated based on the second patent vector and the fruit fly counting bloom filter. The algorithm for calculating the patent novelty score of the present invention has a fast execution speed and high time efficiency, and adopts a fruit fly counting bloom filter structure to store patent historical data information, and occupies a small memory space and has high space efficiency. The patent novelty score result calculated on this basis is also relatively accurate.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a patent novelty assessment method, device, equipment and readable storage medium. Background Art

[0002] Patent novelty mainly refers to the degree of originality of the invention or utility model technology at the technical level, and is a quantification of patent novelty. With the rapid development of various technologies in different fields, a large number of patent documents have been applied for and authorized. The massive amount of patents has brought new opportunities for technology-based companies, but also brought new challenges to the analysis of patent technologies. For companies, patents with higher novelty often have higher value and are more conducive to the company's technological innovation and development. However, patent data has the characteristics of large data volume and rich semantics. Traditional text novelty evaluation methods often cannot achieve fast and effective patent novelty evaluation.

[0003] Therefore, the purpose of patent novelty assessment is to evaluate the degree of innovation of the technology contained in a patent by comparing the similarity with historical data in patent data. Generally speaking, the novelty of a patent based on similarity is generally defined as follows: given a patent and a set of historical patents in the same category, the novelty of the patent is determined by the technical similarity between the patent and the patents in the historical patent set. The lower the similarity, the higher the patent novelty. In theory, the novelty of a patent can be directly calculated by the similarity between patent vectors, but this approach has a high time complexity, the algorithm execution speed is relatively slow, and it is not feasible when the amount of patent data is large. Summary of the invention

[0004] The main purpose of the present invention is to provide a patent novelty evaluation method, device, equipment and readable storage medium, aiming to solve the technical problem that the existing solution of directly calculating the patent novelty based on the similarity between patent vectors has high time complexity and is not suitable for the situation where the amount of data corresponding to the patent category is large.

[0005] In a first aspect, the present invention provides a patent novelty evaluation method, the patent novelty evaluation method comprising the following steps:

[0006] Determine the corresponding fruit fly counting Bloom filter based on the category of the patent to be evaluated;

[0007] Inputting the text of the patent to be evaluated into a language model to obtain a first patent vector;

[0008] Reconstructing the first patent vector based on the fruit fly neural network algorithm to obtain a second patent vector;

[0009] Based on the second patent vector and the fruit fly counting Bloom filter, the novelty score of the patent to be evaluated is calculated.

[0010] Optionally, the step of determining a corresponding fruit fly counting Bloom filter based on the category of the patent to be evaluated includes:

[0011] A target data set is formed based on the target patent text, wherein the target patent text is taken from the patent history data in the patent database, and the category of the target patent text is consistent with the category of the patent to be evaluated;

[0012] After preprocessing the target patent text in the target data set, the text is input into the language model to obtain a corresponding target patent vector group, wherein each target patent vector in the target patent vector group has the same number of dimensions;

[0013] Reconstructing each target patent vector in the target patent vector group based on a fruit fly neural network algorithm to obtain a first target patent vector group;

[0014] The numerical values ​​of each dimension of each first target patent vector in the first target patent vector group are superimposed on the corresponding positions of the array of the fruit fly counting Bloom filter structure to obtain the fruit fly counting Bloom filter corresponding to the category of the patent to be evaluated, wherein the array length of the fruit fly counting Bloom filter structure is the same as the number of dimensions of each first target patent vector in the first target patent vector group.

[0015] Optionally, the step of reconstructing the first patent vector based on the fruit fly neural network algorithm to obtain the second patent vector includes:

[0016] Normalizing each dimension of the first patent vector to obtain a third patent vector;

[0017] Obtain a target matrix, and perform dimension expansion on the third patent vector based on the target matrix to obtain a fourth patent vector, wherein the number of dimensions of the fourth patent vector is greater than the number of dimensions of the third patent vector;

[0018] The fourth patent vector is subjected to vector binarization processing of the winner-takes-all strategy to obtain the second patent vector.

[0019] Optionally, the step of obtaining the target matrix includes:

[0020] Generate a target matrix, the step of generating the target matrix comprising:

[0021] Generate a random matrix of a preset size, wherein the number of columns of the random matrix is ​​determined based on the amount of data in the target patent text, and the number of rows of the random matrix is ​​the number of dimensions of each target patent vector in the target patent vector group;

[0022] Initializing the values ​​of all positions of the random matrix to zero;

[0023] Each row of the initialized random matrix is ​​randomly sampled, and the positions of the preset number of sampling selections are set to one to obtain a target matrix.

[0024] Optionally, the step of normalizing each dimension of the first patent vector to obtain a third patent vector includes:

[0025] Substituting the value of each dimension of the first patent vector into the normalization formula, a third patent vector is obtained. The normalization formula is:

[0026]

[0027] Among them, V i ' is the third patent vector, V i is the value of each dimension in the first patent vector, V min is the minimum value of each dimension in the first patent vector, avg(V) is the numerical mean of all dimensions in the first patent vector, and mean is a preset value.

[0028] Optionally, the step of calculating the novelty score of the patent to be evaluated based on the second patent vector and the fruit fly count Bloom filter includes:

[0029] Substitute the second patent vector and the array of the fruit fly counting Bloom filter structure into the first formula, and calculate the novelty score of the patent to be evaluated based on the first formula. The first formula is:

[0030]

[0031] Where s is the novelty score of the patent to be evaluated, M[i] is the value of the i-th element of the array of the fruit fly counting bloom filter structure, ξ is the scaling factor, V i ′ is the value of the i-th dimension of the second eigenvector, and ||V′||0 is the number of dimensions with non-zero values ​​in the second eigenvector.

[0032] In a second aspect, the present invention further provides a patent novelty evaluation device, the patent novelty evaluation device comprising:

[0033] A determination module, used for determining a corresponding fruit fly counting Bloom filter based on the category of the patent to be evaluated;

[0034] A vectorization representation module, used for inputting the text of the patent to be evaluated into a language model to obtain a first patent vector;

[0035] A reconstruction module, used for reconstructing the first patent vector based on a fruit fly neural network algorithm to obtain a second patent vector;

[0036] A calculation module is used to calculate the novelty score of the patent to be evaluated based on the second patent vector and the fruit fly counting Bloom filter.

[0037] Optionally, the determining module is specifically used to:

[0038] A target data set is formed based on the target patent text, wherein the target patent text is taken from the patent history data in the patent database, and the category of the target patent text is consistent with the category of the patent to be evaluated;

[0039] After preprocessing the target patent text in the target data set, the text is input into the language model to obtain a corresponding target patent vector group, wherein each target patent vector in the target patent vector group has the same dimension;

[0040] Reconstructing each target patent vector in the target patent vector group based on a fruit fly neural network algorithm to obtain a first target patent vector group;

[0041] The numerical values ​​of each dimension of each first target patent vector in the first target patent vector group are superimposed on the corresponding positions of the array of the fruit fly counting Bloom filter structure to obtain the fruit fly counting Bloom filter corresponding to the category of the patent to be evaluated, wherein the array length of the fruit fly counting Bloom filter structure is the same as the number of dimensions of each first target patent vector in the first target patent vector group.

[0042] Optionally, the reconstruction module is specifically used to:

[0043] Normalizing each dimension of the first patent vector to obtain a third patent vector;

[0044] Obtain a target matrix, and perform dimension expansion on the third patent vector based on the target matrix to obtain a fourth patent vector, wherein the number of dimensions of the fourth patent vector is greater than the number of dimensions of the third patent vector;

[0045] The fourth patent vector is subjected to vector binarization processing of the winner-takes-all strategy to obtain the second patent vector.

[0046] Optionally, the patent novelty evaluation device further includes a generation module, which is specifically used to:

[0047] Generate a target matrix, the step of generating the target matrix comprising:

[0048] Generate a random matrix of a preset size, wherein the number of columns of the random matrix is ​​determined based on the amount of data in the target patent text, and the number of rows of the random matrix is ​​the number of dimensions of each target patent vector in the target patent vector group;

[0049] Initializing the values ​​of all positions of the random matrix to zero;

[0050] Each row of the initialized random matrix is ​​randomly sampled, and the positions of the preset number of sampling selections are set to one to obtain a target matrix.

[0051] Optionally, the patent novelty evaluation device further includes a normalization module, which is specifically used to:

[0052] Substituting the value of each dimension of the first patent vector into the normalization formula, a third patent vector is obtained. The normalization formula is:

[0053]

[0054] Among them, V i ' is the third patent vector, V i is the value of each dimension in the first patent vector, V min is the minimum value of each dimension in the first patent vector, avg(V) is the numerical mean of all dimensions in the first patent vector, and mean is a preset value.

[0055] Optionally, the computing module is specifically used to:

[0056] Substitute the second patent vector and the array of the fruit fly counting Bloom filter structure into the first formula, and calculate the novelty score of the patent to be evaluated based on the first formula. The first formula is:

[0057]

[0058] Where s is the novelty score of the patent to be evaluated, M[i] is the value of the i-th element of the array of the fruit fly counting bloom filter structure, ξ is the scaling factor, V i ′ is the value of the i-th dimension of the second eigenvector, and ||V′||0 is the number of dimensions with non-zero values ​​in the second eigenvector.

[0059] In a third aspect, the present invention also provides a patent novelty assessment device, which includes a processor, a memory, and a patent novelty assessment program stored on the memory and executable by the processor, wherein when the patent novelty assessment program is executed by the processor, the steps of the patent novelty assessment method described above are implemented.

[0060] In a fourth aspect, the present invention further provides a readable storage medium, on which a patent novelty evaluation program is stored, wherein when the patent novelty evaluation program is executed by a processor, the steps of the patent novelty evaluation method as described above are implemented.

[0061] The present invention provides a patent novelty evaluation method, device, equipment and readable storage medium. The patent novelty evaluation method includes: determining the corresponding fruit fly counting bloom filter based on the category of the patent to be evaluated; inputting the text of the patent to be evaluated into the language model to obtain a first patent vector; reconstructing the first patent vector based on the fruit fly neural network algorithm to obtain a second patent vector; and calculating the novelty score of the patent to be evaluated based on the second patent vector and the fruit fly counting bloom filter. The algorithm for calculating the patent novelty score of the present invention has a fast execution speed and high time efficiency, and adopts a fruit fly counting bloom filter structure to store patent historical data information, and occupies a small memory space and has high space efficiency. The patent novelty score result calculated on this basis is also relatively accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A schematic diagram of the hardware structure of a patent novelty assessment device involved in an embodiment of the present invention;

[0063] Figure 2 A schematic diagram of a process flow of an embodiment of a method for evaluating patent novelty of the present invention;

[0064] Figure 3 A flowchart of another embodiment of the patent novelty evaluation method of the present invention;

[0065] Figure 4 A schematic diagram of a flow chart of another embodiment of the patent novelty evaluation method of the present invention;

[0066] Figure 5 Schematic diagram of functional modules of an embodiment of a device for evaluating patent novelty according to the present invention.

[0067] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0068] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0069] In a first aspect, an embodiment of the present invention provides a patent novelty assessment device.

[0070] Reference Figure 1 , Figure 1It is a schematic diagram of the hardware structure of the patent novelty evaluation device involved in the embodiment of the present invention. In the embodiment of the present invention, the patent novelty evaluation device may include a processor 1001 (such as a central processing unit, CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components; the user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard); the network interface 1004 may optionally include a standard wired interface, a wireless interface (such as wireless fidelity WIreless-FIdelity, WI-FI interface); the memory 1005 may be a high-speed random access memory (random access memory, RAM), or a stable memory (non-volatile memory), such as a disk storage, and the memory 1005 may optionally be a storage device independent of the aforementioned processor 1001. Those skilled in the art will understand that Figure 1 The hardware structure shown in the figure does not constitute a limitation of the present invention, and may include more or less components than those shown in the figure, or combine certain components, or arrange the components differently.

[0071] Continue to refer to Figure 1 , Figure 1 The memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a patent novelty evaluation program. The processor 1001 may call the patent novelty evaluation program stored in the memory 1005 and execute the patent novelty evaluation method provided in the embodiment of the present invention.

[0072] In a second aspect, an embodiment of the present invention provides a method for evaluating patent novelty.

[0073] Reference Figure 2 , Figure 2 Schematic diagram of a process of an embodiment of a method for evaluating patent novelty of the present invention.

[0074] In an embodiment of the patent novelty evaluation method of the present invention, the patent novelty evaluation method includes:

[0075] Step S10, determining a corresponding fruit fly counting Bloom filter based on the category of the patent to be evaluated;

[0076] In this embodiment, the novelty of any patent needs to be evaluated in combination with the historical patent data in the field of technology of the patent. Therefore, before evaluating any patent, it is necessary to first search for all historical patents of the same category as the patent to be evaluated. In this embodiment, after searching for all historical patents of the same category as the patent to be evaluated, a fruit fly counting Bloom filter is used to store the data information of the historical patents to represent the technical features of the corresponding category patent set. This method not only occupies less memory space, but the historical patent data in the fruit fly counting Bloom filter can be dynamically added and deleted, which is more flexible than general algorithms. Therefore, in this embodiment, before evaluating the input patent text, the corresponding fruit fly counting Bloom filter is first determined based on the category of the input patent to be evaluated.

[0077] Further, in one embodiment, referring to Figure 3 , the step S10 comprises:

[0078] Step S101, constructing a target data set based on a target patent text, wherein the target patent text is taken from patent history data in a patent database, and the category of the target patent text is consistent with the category of the patent to be evaluated;

[0079] Step S102, preprocessing the target patent text in the target data set and inputting it into the language model to obtain a corresponding target patent vector group, wherein each target patent vector in the target patent vector group has the same dimension;

[0080] Step S103, reconstructing each target patent vector in the target patent vector group based on the fruit fly neural network algorithm to obtain a first target patent vector group;

[0081] Step S104, superimposing the values ​​of each dimension of each first target patent vector in the first target patent vector group to the corresponding position of the array of the fruit fly counting Bloom filter structure to obtain a fruit fly counting Bloom filter corresponding to the category of the patent to be evaluated, wherein the array length of the fruit fly counting Bloom filter structure is the same as the number of dimensions of each first target patent vector in the first target patent vector group.

[0082] In this embodiment, specifically, the step of determining the corresponding fruit fly counting Bloom filter based on the category of the patent to be evaluated includes: based on the category of the patent to be evaluated, searching for a target patent text with the same category as the above-mentioned patent to be evaluated from the patent history data of the patent database, and constructing a corresponding target data set based on the above-mentioned searched target patent text.

[0083] The target patent text searched from the patent database is a semantic text, which needs to be converted into a vectorized representation through a language model such as the Sentence-Bert pre-trained model. Before inputting into the language model, the target patent text needs to be preprocessed to extract the text content and remove some characters to improve the accuracy of the text features. The preprocessing operation includes: splicing the title text and abstract text of the target patent text, and removing the wrong characters and special symbols in the spliced ​​text to obtain a set of text corresponding to the target patent text T = {t1, t2, ... t n}.

[0084] The text set T corresponding to the target patent text is T = {t1, t2, ...t n} is input into the language model, and each target patent text in the text set is trained based on the language model to obtain the corresponding target patent vector group V = {v1, v2, …v n}, wherein the dimensions of each target patent vector in the above target patent vector group are the same.

[0085] After obtaining the target patent vector group, each target patent vector in the target patent vector group is reconstructed based on the fruit fly neural network algorithm to obtain a first target patent vector group. Among them, each target patent vector in the target patent vector group will be converted into a high-dimensional sparse binary vector after reconstruction, and this conversion will hardly change the semantic expression of the patent vector, and it is convenient to insert each reconstructed target patent vector into the fruit fly counting bloom filter.

[0086] Among them, the fruit fly counting bloom filter adopts the form of counting, and the structure is an array, and the length of the array is the same as the number of dimensions of each first target patent vector in the first target patent vector group. After obtaining the first target patent vector group, the values ​​of each dimension of each first target patent vector in the first target patent vector group are superimposed on the corresponding position of the array of the fruit fly counting bloom filter structure to obtain the fruit fly counting bloom filter corresponding to the category of the patent to be evaluated.

[0087] For example, if the first target patent vector group includes three four-dimensional first target patent vectors v1 = (1,0,0,1), v2 = (1,0,1,0), and v3 = (0,1,0,1), the initial fruit fly counting Bloom filter array is [0,0,0,0], and v1, v2, and v3 are inserted into the corresponding positions in the array structure, that is, the first dimension of each vector is superimposed on the first element position in the array structure, and the second dimension of each vector is superimposed on the second element position in the array structure, and so on, to obtain the fruit fly counting Bloom filter corresponding to the input patent to be evaluated, which is a [2,1,1,2] array in structure.

[0088] Step S20, inputting the text of the patent to be evaluated into a language model to obtain a first patent vector;

[0089] In this embodiment, the input patent text to be evaluated is a semantic text, which needs to be converted into a vectorized representation through a language model such as a Sentence-Bert pre-training model. Before inputting into the language model, the text of the above-mentioned patent to be evaluated needs to be preprocessed to extract the text content and remove some characters to improve the accuracy of the text features. The preprocessing operation includes: splicing the title text and abstract text of the input patent text to be evaluated, and removing the wrong characters and special symbols in the spliced ​​text to obtain the first patent vector corresponding to the patent text to be evaluated.

[0090] Step S30, reconstructing the first patent vector based on the fruit fly neural network algorithm to obtain a second patent vector;

[0091] In this embodiment, the first patent vector is reconstructed based on the fruit fly neural network algorithm to obtain the second patent vector. The process of reconstructing the first patent vector based on the fruit fly neural network algorithm is the same as the process of reconstructing each target patent vector in the target patent vector group based on the fruit fly neural network algorithm. The reconstruction process includes: vector normalization, vector dimension expansion, and vector binarization. After reconstruction, the first patent vector will be converted into a high-dimensional sparse binary vector, namely the second patent vector, and this conversion will hardly change the semantic expression of the patent vector. The first patent vector is processed in accordance with the target patent vector group, so that the first patent vector can be directly calculated with the fruit fly counting Bloom filter obtained based on the target patent vector group to obtain a novelty score.

[0092] Further, in one embodiment, referring to Figure 4 , the step S30 comprises:

[0093] Step S301, normalizing each dimension of the first patent vector to obtain a third patent vector;

[0094] Step S302, obtaining a target matrix, and performing dimension expansion on the third patent vector based on the target matrix to obtain a fourth patent vector, wherein the number of dimensions of the fourth patent vector is greater than the number of dimensions of the third patent vector;

[0095] Step S303, performing vector binarization processing of the winner-takes-all strategy on the fourth patent vector to obtain a second patent vector.

[0096] In this embodiment, each dimension of the first patent vector is normalized to obtain a third patent vector. A target matrix is ​​obtained, and the third patent vector is dimensionally expanded based on the target matrix to obtain a fourth patent vector, that is, the fourth patent vector is obtained by performing a dot multiplication of the target matrix and the third eigenvector, and the dimension of the fourth patent vector is greater than the dimension of the third patent vector.

[0097] After the fourth eigenvector is obtained by dimensional expansion, the fourth patent vector is subjected to a vector binarization process of the winner-takes-all strategy to obtain a high-dimensional sparse binary vector, namely, the second patent vector. Among them, the vector binarization of the winner-takes-all strategy is adopted, that is, the numerical order of each dimension in the fourth patent vector is first determined, the first n dimensions of the numerical order are set to 0, and the other dimensions are set to 1, so as to obtain a high-dimensional sparse binary vector. Among them, the value of n is much smaller than the number of dimensions of the fourth patent vector, which is related to the error rate of the fruit fly counting Bloom filter, and is usually around 10.

[0098] Furthermore, in one embodiment, the step before step S302 includes:

[0099] Generate a target matrix, the step of generating the target matrix comprising:

[0100] Generate a random matrix of a preset size, wherein the number of columns of the random matrix is ​​determined based on the amount of data in the target patent text, and the number of rows of the random matrix is ​​the number of dimensions of each target patent vector in the target patent vector group;

[0101] Initializing the values ​​of all positions of the random matrix to zero;

[0102] Each row of the initialized random matrix is ​​randomly sampled, and the positions of the preset number of sampling selections are set to one to obtain a target matrix.

[0103] In this embodiment, before obtaining the target matrix, it is necessary to generate the target matrix first. The target matrix is ​​generated in the process of reconstructing each target patent vector in the target patent vector group based on the fruit fly neural network algorithm. Since the target matrix when reconstructing the first patent vector based on the fruit fly neural network algorithm is the same as the target matrix when reconstructing each target patent vector in the target patent vector group based on the fruit fly neural network algorithm, the corresponding target matrix can be directly obtained at this time.

[0104] The step of generating the target matrix includes: generating a random matrix of a preset size, wherein the random matrix is ​​to increase the dimension of each target patent vector in the target patent vector group, so the number of rows of the random matrix is ​​the number of dimensions of each target patent vector in the target patent vector group. The number of columns of the random matrix is ​​determined based on the data volume of the target patent text. Considering the error rate of the fruit fly bloom filter, the number of columns of the random matrix should be greater than 10 times the data volume of the target patent text.

[0105] After generating a random matrix of a preset size, the values ​​of all positions of the random matrix are initialized to zero. Then, each row of the initialized random matrix is ​​randomly sampled, and the positions of the preset number of samplings are set to one to obtain a target matrix. The preset number is generally 6.

[0106] Furthermore, in one embodiment, step S301 includes:

[0107] Substituting the value of each dimension of the first patent vector into the normalization formula, a third patent vector is obtained. The normalization formula is:

[0108]

[0109] Among them, V i ' is the third patent vector, V i is the value of each dimension in the first patent vector, V min is the minimum value of each dimension in the first patent vector, avg(V) is the numerical mean of all dimensions in the first patent vector, and mean is a preset value.

[0110] In this embodiment, the value of each dimension of the first patent vector is substituted into the normalization formula to obtain the third patent vector, and the normalization formula is:

[0111]

[0112] Among them, V i ' is the third patent vector, V i is the value of each dimension in the first patent vector, V min is the minimum value of each dimension in the first patent vector, avg(V) is the numerical mean of all dimensions in the first patent vector, and mean is a preset value, which represents the mean of each dimension in the first patent vector after normalization.

[0113] Step S40, based on the second patent vector and the fruit fly count Bloom filter, calculate the novelty score of the patent to be evaluated.

[0114] In this embodiment, the novelty score of the patent to be evaluated is calculated based on the second patent vector and the fruit fly counting Bloom filter. Compared with the prior art, the novelty score is obtained by directly calculating the similarity between the second patent vector and the target patent vector. When the amount of patent data is large, the execution speed of the algorithm is faster, and due to the inherent characteristics of the fruit fly counting Bloom filter, the memory space occupied by the present invention is also smaller. In addition, the reconstruction process of the patent vector based on the fruit fly neural network algorithm will not change the semantic expression of the patent vector. At the same time, the fruit fly counting Bloom filter is used to represent the technical features of the historical patent data set corresponding to the category of the patent to be evaluated. The corresponding historical patent data can be dynamically deleted and inserted, and the features are more representative, which improves the accuracy of the patent novelty evaluation.

[0115] Further, in one embodiment, the step S40 includes:

[0116] Substitute the second patent vector and the array of the fruit fly counting Bloom filter structure into the first formula, and calculate the novelty score of the patent to be evaluated based on the first formula. The first formula is:

[0117]

[0118] Where s is the novelty score of the patent to be evaluated, M[i] is the value of the i-th element of the array of the fruit fly counting bloom filter structure, ξ is the scaling factor, V i ′ is the value of the i-th dimension of the second eigenvector, and ||V′||0 is the number of dimensions with non-zero values ​​in the second eigenvector.

[0119] In this embodiment, the second patent vector and the array of the fruit fly counting Bloom filter structure are substituted into the first formula, and the novelty score of the patent to be evaluated is calculated based on the first formula. The first formula is:

[0120]

[0121] Where s is the novelty score of the patent to be evaluated, M[i] is the value of the i-th element of the array of the fruit fly counting bloom filter structure, ξ is the scaling factor, V i ′ is the value of the i-th dimension of the second eigenvector, and ||V′||0 is the number of dimensions with non-zero values ​​in the second eigenvector. Among them, ξ is a scaling factor with a value range of (0, 1), which represents the influence of the corresponding value of each position in the fruit fly counting Bloom filter on the final novelty score. This value is a preset value, which is determined according to the influence of the value on the Bloom filter when executing the algorithm. The larger the value of ξ, the greater the influence on the value of the corresponding array element on the fruit fly counting Bloom filter.

[0122] In this embodiment, a patent novelty evaluation method is provided, which includes: determining a corresponding fruit fly counting bloom filter based on the category of the patent to be evaluated; inputting the text of the patent to be evaluated into a language model to obtain a first patent vector; reconstructing the first patent vector based on a fruit fly neural network algorithm to obtain a second patent vector; and calculating the novelty score of the patent to be evaluated based on the second patent vector and the fruit fly counting bloom filter. The algorithm for calculating the patent novelty score of the present invention has a fast execution speed and high time efficiency, and uses a fruit fly counting bloom filter structure to store patent historical data information, which occupies a small memory space and has high space efficiency. The patent novelty score result calculated on this basis is also relatively accurate.

[0123] In a third aspect, an embodiment of the present invention also provides a patent novelty assessment device.

[0124] Reference Figure 5 , a functional module diagram of an embodiment of a patent novelty assessment device.

[0125] In this embodiment, the patent novelty evaluation device includes:

[0126] A determination module 10, for determining a corresponding fruit fly counting Bloom filter based on the category of the patent to be evaluated;

[0127] A vectorization representation module 20, used for inputting the text of the patent to be evaluated into a language model to obtain a first patent vector;

[0128] A reconstruction module 30, used to reconstruct the first patent vector based on a fruit fly neural network algorithm to obtain a second patent vector;

[0129] The calculation module 40 is used to calculate the novelty score of the patent to be evaluated based on the second patent vector and the fruit fly counting Bloom filter.

[0130] Further, in one embodiment, the determining module 10 is specifically configured to:

[0131] A target data set is formed based on the target patent text, wherein the target patent text is taken from the patent history data in the patent database, and the category of the target patent text is consistent with the category of the patent to be evaluated;

[0132] After preprocessing the target patent text in the target data set, the text is input into the language model to obtain a corresponding target patent vector group, wherein each target patent vector in the target patent vector group has the same dimension;

[0133] Reconstructing each target patent vector in the target patent vector group based on a fruit fly neural network algorithm to obtain a first target patent vector group;

[0134] The numerical values ​​of each dimension of each first target patent vector in the first target patent vector group are superimposed on the corresponding positions of the array of the fruit fly counting Bloom filter structure to obtain the fruit fly counting Bloom filter corresponding to the category of the patent to be evaluated, wherein the array length of the fruit fly counting Bloom filter structure is the same as the number of dimensions of each first target patent vector in the first target patent vector group.

[0135] Furthermore, in one embodiment, the reconstruction module 30 is specifically used to:

[0136] Normalizing each dimension of the first patent vector to obtain a third patent vector;

[0137] Obtain a target matrix, and perform dimension expansion on the third patent vector based on the target matrix to obtain a fourth patent vector, wherein the number of dimensions of the fourth patent vector is greater than the number of dimensions of the third patent vector;

[0138] The fourth patent vector is subjected to vector binarization processing of the winner-takes-all strategy to obtain the second patent vector.

[0139] Furthermore, in one embodiment, the patent novelty evaluation device further includes a generation module, which is specifically used to:

[0140] Generate a target matrix, the step of generating the target matrix comprising:

[0141] Generate a random matrix of a preset size, wherein the number of columns of the random matrix is ​​determined based on the amount of data in the target patent text, and the number of rows of the random matrix is ​​the number of dimensions of each target patent vector in the target patent vector group;

[0142] Initializing the values ​​of all positions of the random matrix to zero;

[0143] Each row of the initialized random matrix is ​​randomly sampled, and the positions of the preset number of sampling selections are set to one to obtain a target matrix.

[0144] Furthermore, in one embodiment, the patent novelty evaluation device further includes a normalization module, which is specifically used to:

[0145] Substituting the value of each dimension of the first patent vector into the normalization formula, a third patent vector is obtained. The normalization formula is:

[0146]

[0147] Among them, V i ' is the third patent vector, V i is the value of each dimension in the first patent vector, V minis the minimum value of each dimension in the first patent vector, avg(V) is the numerical mean of all dimensions in the first patent vector, and mean is a preset value.

[0148] Furthermore, in one embodiment, the calculation module 40 is specifically used for:

[0149] Substitute the second patent vector and the array of the fruit fly counting Bloom filter structure into the first formula, and calculate the novelty score of the patent to be evaluated based on the first formula. The first formula is:

[0150]

[0151] Where s is the novelty score of the patent to be evaluated, M[i] is the value of the i-th element of the array of the fruit fly counting bloom filter structure, ξ is the scaling factor, V i ′ is the value of the i-th dimension of the second eigenvector, and ||V′||0 is the number of dimensions with non-zero values ​​in the second eigenvector.

[0152] Among them, the functional implementation of each module in the above-mentioned patent novelty assessment device corresponds to the various steps in the above-mentioned patent novelty assessment method embodiment, and its functions and implementation processes will not be repeated here one by one.

[0153] In a fourth aspect, an embodiment of the present invention further provides a readable storage medium.

[0154] The readable storage medium of the present invention stores a patent novelty evaluation program, wherein when the patent novelty evaluation program is executed by a processor, the steps of the patent novelty evaluation method as described above are implemented.

[0155] Among them, the method implemented when the patent novelty evaluation program is executed can refer to the various embodiments of the patent novelty evaluation method of the present invention, and will not be repeated here.

[0156] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0157] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0158] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device to execute the methods described in each embodiment of the present invention.

[0159] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A patent novelty evaluation method, characterized in that: The patent novelty evaluation method includes: Determine the corresponding fruit fly counting Bloom filter based on the category of the patent to be evaluated; Inputting the text of the patent to be evaluated into a language model to obtain a first patent vector; Reconstructing the first patent vector based on the fruit fly neural network algorithm to obtain a second patent vector; Based on the second patent vector and the fruit fly counting Bloom filter, calculate the novelty score of the patent to be evaluated; The step of determining the corresponding fruit fly counting Bloom filter based on the category of the patent to be evaluated includes: A target data set is formed based on the target patent text, wherein the target patent text is taken from the patent history data in the patent database, and the category of the target patent text is consistent with the category of the patent to be evaluated; After preprocessing the target patent text in the target data set, the text is input into the language model to obtain a corresponding target patent vector group, wherein each target patent vector in the target patent vector group has the same number of dimensions; Reconstructing each target patent vector in the target patent vector group based on a fruit fly neural network algorithm to obtain a first target patent vector group; The numerical values ​​of each dimension of each first target patent vector in the first target patent vector group are superimposed on the corresponding positions of the array of the fruit fly counting Bloom filter structure to obtain the fruit fly counting Bloom filter corresponding to the category of the patent to be evaluated, wherein the array length of the fruit fly counting Bloom filter structure is the same as the number of dimensions of each first target patent vector in the first target patent vector group.

2. The patent novelty evaluation method according to claim 1, characterized in that: The step of reconstructing the first patent vector based on the fruit fly neural network algorithm to obtain the second patent vector includes: Normalizing each dimension of the first patent vector to obtain a third patent vector; Obtain a target matrix, and perform dimension expansion on the third patent vector based on the target matrix to obtain a fourth patent vector, wherein the number of dimensions of the fourth patent vector is greater than the number of dimensions of the third patent vector; The fourth patent vector is subjected to vector binarization processing of the winner-takes-all strategy to obtain the second patent vector.

3. The patent novelty evaluation method according to claim 2, characterized in that: The step of obtaining the target matrix includes: Generate a target matrix, the step of generating the target matrix comprising: Generate a random matrix of a preset size, wherein the number of columns of the random matrix is ​​determined based on the amount of data in the target patent text, and the number of rows of the random matrix is ​​the number of dimensions of each target patent vector in the target patent vector group; Initializing the values ​​of all positions of the random matrix to zero; Each row of the initialized random matrix is ​​randomly sampled, and the positions of the preset number of sampling selections are set to one to obtain a target matrix.

4. The patent novelty evaluation method according to claim 2, characterized in that: The step of normalizing each dimension of the first patent vector to obtain the third patent vector comprises: Substituting the value of each dimension of the first patent vector into the normalization formula, a third patent vector is obtained. The normalization formula is: Among them, V i ′ is the third patent vector, V i is the value of each dimension in the first patent vector, V min is the minimum value of each dimension in the first patent vector, avg(V) is the numerical mean of all dimensions in the first patent vector, and mean is a preset value.

5. The patent novelty evaluation method according to claim 1, characterized in that: The step of calculating the novelty score of the patent to be evaluated based on the second patent vector and the fruit fly counting Bloom filter comprises: Substitute the second patent vector and the array of the fruit fly counting Bloom filter structure into the first formula, and calculate the novelty score of the patent to be evaluated based on the first formula. The first formula is: Where s is the novelty score of the patent to be evaluated, M[i] is the value of the i-th element of the array of the fruit fly counting bloom filter structure, ξ is the scaling factor, V i ′ is the value of the i-th dimension of the second eigenvector, and V′0 is the number of dimensions with non-zero values ​​in the second eigenvector.

6. A patent novelty evaluation device, characterized in that: The patent novelty evaluation device comprises: A determination module, used for determining a corresponding fruit fly counting Bloom filter based on the category of the patent to be evaluated; A vectorization representation module, used for inputting the text of the patent to be evaluated into a language model to obtain a first patent vector; A reconstruction module, used for reconstructing the first patent vector based on a fruit fly neural network algorithm to obtain a second patent vector; A calculation module, configured to calculate the novelty score of the patent to be evaluated based on the second patent vector and the fruit fly counting Bloom filter; The determining module is specifically used for: A target data set is formed based on the target patent text, wherein the target patent text is taken from the patent history data in the patent database, and the category of the target patent text is consistent with the category of the patent to be evaluated; After preprocessing the target patent text in the target data set, the text is input into the language model to obtain a corresponding target patent vector group, wherein each target patent vector in the target patent vector group has the same dimension; Reconstructing each target patent vector in the target patent vector group based on a fruit fly neural network algorithm to obtain a first target patent vector group; The numerical values ​​of each dimension of each first target patent vector in the first target patent vector group are superimposed on the corresponding positions of the array of the fruit fly counting Bloom filter structure to obtain the fruit fly counting Bloom filter corresponding to the category of the patent to be evaluated, wherein the array length of the fruit fly counting Bloom filter structure is the same as the number of dimensions of each first target patent vector in the first target patent vector group.

7. A patent novelty assessment device, characterized in that: The patent novelty evaluation device includes a processor, a memory, and a patent novelty evaluation program stored in the memory and executable by the processor, wherein when the patent novelty evaluation program is executed by the processor, the steps of the patent novelty evaluation method as described in any one of claims 1 to 5 are implemented.

8. A readable storage medium, characterized in that: The readable storage medium stores a patent novelty evaluation program, wherein when the patent novelty evaluation program is executed by a processor, the steps of the patent novelty evaluation method as described in any one of claims 1 to 5 are implemented.

Images