A search method and device applied to a technical manager platform

Abstract

The application discloses a retrieval method and device applied to a technical manager platform, and the method comprises the following steps: receiving a user input query; judging whether a non-standard term is contained in the user input query; when the non-standard term is contained in the user input query, generating a constraint condition based on the non-standard term and the relevance of nodes in a field knowledge graph; wherein the constraint condition is used for constraining a technical attribute of a term, the attributes of the nodes in the field knowledge graph comprise the technical attribute, and the technical attribute at least comprises a patent correlation degree, a subject cross degree and a technical maturity; converting the non-standard term into at least one standard term by using a generative adversarial term conversion engine according to the non-standard term and the constraint condition; and performing retrieval according to the at least one standard term. The application solves the semantic fault problem between industry terms and general retrieval strategies faced by current technical managers during retrieval.

Description

Technical Field

[0001] This invention belongs to the field of computer information retrieval technology, specifically relating to a retrieval method and apparatus applied to a technology manager platform. Background Technology

[0002] In the field of technology transfer and commercialization, one of the core functions of a technology manager platform is to help users quickly and accurately match technology supply and demand information. However, due to the specialization and industry specificity of the technical field, technology managers often face a semantic gap between industry terminology and general search strategies when conducting searches. Specifically, when technical experts input search keywords, they usually use industry-standard terms or abbreviations, while traditional intelligent search systems, built on general corpora or standard thesaurus, cannot effectively identify these non-standardized but widely used expressions, leading to search results that deviate significantly from user expectations.

[0003] Currently, mainstream technical retrieval solutions address this issue primarily through the following methods:

[0004] Keyword expansion search: This method expands the user-input keywords using a thesaurus or technical dictionary, but the scope is limited and it is difficult to cover dynamic terminology or regional differences in expression within the industry.

[0005] Semantic vector retrieval: It uses pre-trained models such as BERT to map text into vectors for similarity matching, but its training data mostly comes from general domains and is not sensitive enough to the "jargon" of specific industries.

[0006] Therefore, existing technologies do not yet have a good solution to the semantic gap between industry terminology and general search strategies that current technology managers face when searching. Summary of the Invention

[0007] To address the aforementioned problems in the prior art, this invention provides a retrieval method and apparatus for use on a technology manager platform.

[0008] The technical problem to be solved by this invention is achieved through the following technical solution:

[0009] In a first aspect, the present invention provides a retrieval method applied to a technology manager platform, comprising:

[0010] Receive user input queries;

[0011] Determine whether the user-input query contains non-standard terms;

[0012] When the user input query contains non-standard terms, constraints are generated based on the relevance between the non-standard terms and nodes in the domain knowledge graph; wherein, the constraints are used to constrain the technical attributes of the terms, and the attributes of the nodes in the domain knowledge graph include the technical attributes, which at least include patent relevance, interdisciplinary degree, and technology maturity.

[0013] Based on the non-standard terminology and the constraints, the non-standard terminology is converted into at least one standard term using a generative adversarial terminology conversion engine.

[0014] Search according to at least one of the standard terms;

[0015] The generative adversarial terminology conversion engine includes a generator, a discriminator, and an adjuster. The generator generates gated attention based on the constraints and generates at least one candidate term based on the gated attention and the non-standard terminology. The discriminator determines whether all candidate terms satisfy the constraints. When all constraints are satisfied, the at least one candidate term is output as the at least one standard term. The adjuster adjusts the constraints and triggers the generator to regenerate at least one candidate term when the discriminator determines that a candidate term does not satisfy the constraints.

[0016] Optionally, determining whether the user-input query contains non-standard terms includes:

[0017] The user-input query is subjected to terminology extraction, and it is determined whether the extracted terms contain non-standard terms.

[0018] If the extracted terms do not contain non-standard terms, it is further determined whether the non-standard term matches the non-standard term-standard term mapping table. If not, it is determined that the user input query contains non-standard terms.

[0019] Optionally, the technical attributes are encoded as technical attribute vectors;

[0020] The constraints generated based on the relevance of nodes in the non-standard terminology and domain knowledge graph include:

[0021] From the non-standard terminology-standard terminology mapping table, obtain the target standard terminology that is semantically and / or structurally similar to the non-standard terminology;

[0022] From the domain knowledge graph, standard terms with a node similarity greater than a threshold with the target standard term are selected to obtain the first group of candidate nodes;

[0023] Cluster the first group of candidate nodes, and further filter the first group of candidate nodes based on the clustering results to obtain the second group of candidate nodes;

[0024] The constraint vector is constructed based on the second group of candidate nodes as the constraint condition.

[0025] Optionally, the generator generates gated attention based on the constraints, including:

[0026] The non-standard terms are embedded into non-standard term encoding vectors using a word embedding tool, and the non-standard term encoding vectors and the constraint vectors are concatenated into a joint vector.

[0027] ;

[0028] in, This represents the encoding vector of the non-standard terminology. Represents the constraint vector;

[0029] The joint vector is linearly transformed using the first weight matrix and the first bias matrix to obtain the gated vector:

[0030] ;

[0031] in, This is the first weight matrix used to control the feature mixing ratio; It is the first bias matrix used to control the gating sensitivity. and All were obtained through pre-training;

[0032] Based on the gate vector, the gate weights are generated using the Sigmoid function:

[0033] ;

[0034] in, This represents the Sigmoid function.

[0035] Optionally, the generator generates at least one candidate term based on the gated attention and the non-standard terminology, including:

[0036] Projecting the non-standard term encoding vector yields the term projection vector:

[0037] ;

[0038] in, This is the second weight matrix used to filter term noise. It is the second bias vector that controls the activation threshold of term features. and All were obtained through pre-training;

[0039] Projecting the constraint vector yields the constraint projection vector:

[0040]

[0041] in, This represents the third weight matrix used to reinforce the domain preset. It is the third bias vector used to control the intensity of domain knowledge injection. and All were obtained through pre-training; for Activation function;

[0042] Based on the term projection vector and the constraint projection vector, the candidate term feature vector is output using the gating weights:

[0043]

[0044] in, It represents the Hadamardi (or Hadama) stack;

[0045] Based on the feature vectors of the candidate terms, a domain knowledge graph is searched to obtain at least one candidate term.

[0046] Optionally, the discriminator determines whether the at least one candidate term satisfies the constraint condition, including:

[0047] For each candidate term, calculate its connectivity score:

[0048] ;

[0049] in, Candidate terms obtained from the domain knowledge graph The set of k-hop neighbor nodes, This represents the second group of candidate nodes; Indicates taking and The size of the intersection;

[0050] For each candidate term, calculate its technical attribute score:

[0051] ;

[0052] in, This refers to the non-standard terminology. This is the technical attribute score vector for the non-standard terminology. This is a pre-trained technology attribute prediction model, trained based on the domain knowledge graph, used to predict the patent relevance of terms. Interdisciplinary degree and technology maturity ; This represents a vector composed of the patent relevance, interdisciplinary nature, and technological maturity of candidate terms. The formula for calculating vector similarity;

[0053] Based on the connectivity score and technical field score of each candidate term, determine whether the selected term satisfies the constraints:

[0054] ;

[0055] in, and The preset scaling factor, and , This indicates the result of determining whether a candidate term meets the constraints. This indicates that the judgment result is yes. This indicates that the judgment result is negative.

[0056] Optionally, the adjuster adjusts the constraint conditions by:

[0057] Calculate the adversarial vector based on candidate terms:

[0058] ;

[0059] in, , , The subset of candidate terms that satisfy the constraint clause from the at least one candidate term. The subset of candidate terms that do not satisfy the constraint clause among the at least one candidate term. The subset of candidate terms that satisfy the constraint clause from the at least one candidate term. This is the mean vector of the technical attribute vectors of all candidate terms. This is the mean vector of the technical attribute vectors of the second group of candidate nodes;

[0060] Based on the adversarial vector, reconstruct the constraint vector:

[0061] ;

[0062] in, For the amplitude limiting function, The threshold value for the amplitude limit. The reconstructed constraint vector.

[0063] Secondly, the present invention provides a retrieval device applied to a technology manager platform, comprising:

[0064] The user input module is used to receive user input queries;

[0065] The judgment module is used to determine whether the user input query contains non-standard terms;

[0066] A generation module is used to generate constraints based on the relevance between the non-standard terms and nodes in the domain knowledge graph when the user input query contains non-standard terms; wherein, the constraints are used to constrain the technical attributes of the terms, and the attributes of the nodes in the domain knowledge graph include the technical attributes, which at least include patent relevance, interdisciplinary degree, and technology maturity.

[0067] A conversion module is used to convert the non-standard terms into at least one standard term based on the non-standard terms and the constraints using a generative adversarial term conversion engine.

[0068] A retrieval module is used to retrieve data based on the at least one standard term.

[0069] The generative adversarial terminology conversion engine includes a generator, a discriminator, and an adjuster. The generator generates gated attention based on the constraints and generates at least one candidate term based on the gated attention and the non-standard terminology. The discriminator determines whether all candidate terms satisfy the constraints. When all constraints are satisfied, the at least one candidate term is output as the at least one standard term. The adjuster adjusts the constraints and triggers the generator to regenerate at least one candidate term when the discriminator determines that a candidate term does not satisfy the constraints.

[0070] Thirdly, the present invention also provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0071] Memory, used to store computer programs;

[0072] When the processor executes a program stored in memory, it implements the method steps of any of the above-mentioned retrieval methods applied to the technology manager platform.

[0073] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the method steps of any of the above-described retrieval methods applied to a technology manager platform.

[0074] The retrieval method provided by this invention for a technology manager platform, by combining the technical attributes of a domain knowledge graph with a generative adversarial terminology conversion engine, can intelligently and dynamically convert non-standard terms in user queries into standard terms that conform to technical attribute constraints such as patent relevance and interdisciplinary integration, significantly improving the accuracy and coverage of technology retrieval. The gating attention mechanism of the generator and the condition verification of the discriminator form a closed-loop optimization, which, together with the dynamic constraint adjustment of the adjuster, effectively solves the semantic gap problem between industry terms and general retrieval strategies currently faced by technology managers during retrieval.

[0075] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0076] Figure 1 This is a flowchart illustrating a retrieval method applied to a technology manager platform according to an embodiment of the present invention;

[0077] Figure 2 This is a schematic diagram illustrating the working principle of a generative adversarial terminology conversion engine provided in an embodiment of the present invention;

[0078] Figure 3 This is a schematic diagram of the structure of a retrieval device applied to a technology manager platform provided in an embodiment of the present invention;

[0079] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0080] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0081] To address the semantic gap between industry terminology and general search strategies faced by technology managers during retrieval, this invention provides a retrieval method applicable to technology manager platforms, such as... Figure 1 As shown, the method includes the following steps:

[0082] S10, Receive user input query.

[0083] Here, users can enter a query that consists of one or more words, or a sentence.

[0084] S20. Determine whether the user's input query contains non-standard terms.

[0085] Specifically, determining whether the user's input query contains non-standard terms includes:

[0086] (1) Extract terms from user input queries and determine whether the extracted terms contain non-standard terms.

[0087] Here, if the user's input query consists of one or more words, terms can be extracted directly from them. If the user's input query is a sentence, it can be segmented into words first, and then terms can be extracted.

[0088] (2) If the extracted terms do not contain non-standard terms, further determine whether the non-standard terms match the non-standard term-standard term mapping table. If not, determine that the user input query contains non-standard terms.

[0089] Here, the non-standard terminology-standard terminology mapping table is pre-built. This table stores the mapping relationships between non-standard terms and standard terms. A lookup of a non-standard term in the table will retrieve its corresponding standard term. If a non-standard term does not match the non-standard terminology-standard terminology mapping table, it means that the table does not contain the non-standard term and its corresponding standard term. In this case, it can be determined that the user's input query contains a non-standard term. If a non-standard term matches the non-standard terminology-standard terminology mapping table, the retrieval can be performed directly using the standard term mapped to the non-standard term.

[0090] S30. When the user input query contains non-standard terms, generate constraint conditions based on the relevance between the non-standard terms and nodes in the domain knowledge graph; wherein, the constraint conditions are used to constrain the technical attributes of the terms, and the attributes of the nodes in the domain knowledge graph include technical attributes, which at least include patent relevance, interdisciplinary degree and technology maturity.

[0091] Nodes in a domain knowledge graph are all standard terms, and their attributes include the aforementioned technical attributes as well as their semantic attributes. In practice, the semantic attributes of nodes in a domain knowledge graph can be represented by standard term encoding vectors. These standard term encoding vectors can be obtained by embedding standard terms using word embedding tools, such as BERT, but not limited to this. The technical attributes of nodes in a domain knowledge graph can be obtained through the following methods:

[0092] Specifically, regarding patent relevance, some patent search platforms provide citation relationships and technical similarity scores between patents, which can approximate patent relevance. For standard terms, their patent relevance can be set based on the patent text in which they appear. For example, patent texts often contain keywords, which are also standard terms; therefore, the patent relevance of the patent text can be directly cited based on these keywords. Regarding interdisciplinary relevance, the coexistence of multiple classification codes can indicate the interdisciplinary relevance of a patent. Correspondingly, the interdisciplinary relevance of keywords in a patent can be set based on the patent's interdisciplinary relevance. Alternatively, the interdisciplinary relevance of a patent can be determined by statistically analyzing the number and span of classification codes involved, and then using a subject modeling approach (LDA) to analyze cross-disciplinary terms in the abstract / claims, thereby setting the interdisciplinary relevance of these terms as the patent's interdisciplinary relevance. Regarding technology maturity, some patent search platforms provide technology development stage curves (emergence / growth / maturity / decline), which can indicate the technology maturity of a patent. Correspondingly, the technology maturity of keywords in a patent can be set based on the patent's technology maturity.

[0093] In step S30, constraints are generated based on the relevance of nodes in the non-standard terminology and domain knowledge graph, including:

[0094] (i) Obtain target standard terms that are semantically and / or structurally similar to non-standard terms from the non-standard term-standard term mapping table.

[0095] Specifically, non-standard terms are embedded using word embedding tools to obtain non-standard term encoding vectors. Then, based on the similarity of the encoded vectors, terms that are semantically similar to the non-standard terms are obtained from the non-standard term-standard term mapping table. These terms include both non-standard and standard terms from the table. If a semantically similar term is also a non-standard term, its corresponding standard term is then obtained. Similarly, terms that are structurally similar to the non-standard terms are obtained from the non-standard term-standard term mapping table. Structural similarity refers to combinations containing the same characters. If a structurally similar term is also a non-standard term, its corresponding standard term is then obtained.

[0096] (ii) Select standard terms from the domain knowledge graph whose node similarity to the target standard term is greater than the threshold to obtain the first group of candidate nodes.

[0097] Specifically, node similarity includes both semantic and technical attribute similarity. Semantic similarity can be achieved by calculating vector similarity. Technical attribute similarity can be achieved by calculating the similarity of vectors composed of three technical attributes, but it is not limited to this. Correspondingly, the aforementioned thresholds include thresholds for both semantic and technical attribute similarity dimensions.

[0098] Understandably, this step is equivalent to a coarse screening of nodes in the domain knowledge graph, so the threshold used in this step should not be too high.

[0099] (iii) Cluster the first group of candidate nodes, and further filter the first group of candidate nodes based on the clustering results to obtain the second group of candidate nodes.

[0100] Here, the K-Means method can be used to cluster the first group of candidate nodes. Then, the top-ranked clusters are further selected from the clustering results as the second group of candidate nodes.

[0101] (iv) Construct constraint vectors based on the second group of candidate nodes as constraint conditions.

[0102] Specifically, the technical attribute vectors of the second group of candidate nodes are weighted and summed to obtain the constraint vector. When weighting and summing the technical attribute vectors of these nodes, the weights can be set according to the ranking of the clustering group to which the node belongs in the clustering results, with higher-ranked clustering groups assigned greater weights. Furthermore, when assigning weights, it is ensured that the sum of the weights of each technical attribute vector is 1.

[0103] S40. Based on the non-standard terms and constraints, use a generative adversarial terminology conversion engine to convert the non-standard terms into at least one standard term.

[0104] Among them, see Figure 2 As shown, the generative adversarial terminology conversion engine includes a generator, a discriminator, and an adjuster. The generator is used to generate gated attention based on constraints and to generate at least one candidate term based on the gated attention and non-standard terms. The discriminator is used to determine whether at least one candidate term satisfies the constraints. When all constraints are satisfied, at least one candidate term is output as at least one standard term. The adjuster is used to adjust the constraints and trigger the generator to regenerate at least one candidate term when the discriminator determines that a candidate term does not satisfy the constraints.

[0105] Specifically, the generator generates gated attention based on constraints, including:

[0106] (a) Use word embedding tools to embed non-standard terms into non-standard term encoding vectors (already implemented), and concatenate the non-standard term encoding vectors and constraint vectors into a joint vector:

[0107] ;

[0108] in, Represents the encoding vector for non-standard terms. Represents the constraint vector;

[0109] (b) The joint vector is linearly transformed using the first weight matrix and the first bias matrix to obtain the gated vector:

[0110] ;

[0111] in, This is the first weight matrix used to control the feature mixing ratio; It is the first bias matrix used to control the gating sensitivity. and All were obtained through pre-training;

[0112] (c) Generate gate weights using the Sigmoid function based on the gate vector:

[0113] ;

[0114] in, This represents the Sigmoid function.

[0115] Then, the generator generates at least one candidate term based on gated attention and non-standard terminology, including:

[0116] (1) Project the non-standard term encoding vector to obtain the term projection vector:

[0117] ;

[0118] in, This is the second weight matrix used to filter term noise. It is the second bias vector that controls the activation threshold of term features. and All were obtained through pre-training;

[0119] (2) Project the constraint vector to obtain the constraint projection vector:

[0120]

[0121] in, This represents the third weight matrix used to reinforce the domain preset. It is the third bias vector used to control the intensity of domain knowledge injection. and All were obtained through pre-training; for Activation function;

[0122] (3) Based on the term projection vector and the constraint projection vector, output the candidate term feature vector using gating weights:

[0123] ;

[0124] in, It represents the Hadamardi (or Hadama) stack.

[0125] (4) Search the domain knowledge graph based on the feature vectors of the candidate terms to obtain at least one candidate term.

[0126] It is understandable that the candidate term feature vector essentially contains information about the candidate term in both semantic and technical attributes. Therefore, it is input into a pre-trained neural network model, which then predicts the encoding vector and technical attribute vector of the candidate term based on the candidate term feature vector. These two vectors are then used to search the domain knowledge graph (based on vector similarity search).

[0127] After at least one candidate term is found from the domain knowledge graph, the discriminator further judges whether these candidate terms all meet the constraints.

[0128] Specifically, the discriminator determines whether at least one candidate term found in the search satisfies the constraints, including:

[0129] (i) For each candidate term, calculate the connectivity score for that candidate term:

[0130] ;

[0131] in, Candidate terms obtained from the domain knowledge graph The set of k-hop neighbor nodes, Indicates the second group of candidate nodes; Indicates taking and The size of the intersection;

[0132] (ii) For each candidate term, calculate the technical attribute score for that candidate term:

[0133] ;

[0134] in, Indicates non-standard terms, This is a score vector for the technical attributes of non-standard terms. This is a pre-trained technology attribute prediction model, trained based on a domain knowledge graph, used to predict the patent relevance of terms. Interdisciplinary degree and technology maturity ; This represents a vector composed of the patent relevance, interdisciplinary nature, and technological maturity of candidate terms. The formula for calculating vector similarity;

[0135] (iv) Based on the connectivity score and technical field score of each candidate term, determine whether the selected term meets the constraints:

[0136] ;

[0137] in, and The preset scaling factor, and ,For example It can be set to 0.7. It can be set to 0.3, but it is not limited to this. This indicates the result of determining whether a candidate term meets the constraints. This indicates that the judgment result is yes. This indicates that the judgment result is negative.

[0138] When the decision region determines that all candidate terms found satisfy the constraints, it outputs at least one candidate term as the final converted (i.e., the standard term used for retrieval). Conversely, if the decision is negative, meaning that not all candidate terms satisfy the constraints, the regulator is further triggered. Specifically, the regulator adjusts the constraints and triggers the generator to regenerate at least one candidate term when the discriminator determines that a candidate term does not satisfy the constraints.

[0139] Among them, the adjuster adjusts the constraints, including:

[0140] (a) Calculate the adversarial vector based on candidate terms:

[0141] ;

[0142] in, , , A subset of candidate terms that satisfy the constraint clauses among at least one candidate term. A subset of candidate terms that do not satisfy the constraint conditions, among at least one candidate term. A subset of candidate terms that satisfy the constraint clauses among at least one candidate term. This is the mean vector of the technical attribute vectors of all candidate terms. This is the mean vector of the technical attribute vectors of the second group of candidate nodes;

[0143] (b) Reconstruct the constraint vector based on the adversarial vector:

[0144] ;

[0145] in, For the amplitude limiting function, This is the threshold value for the amplitude limit, which can be defaulted to 0.1 in practice, but it is not limited to this. The reconstructed constraint vector.

[0146] Then, The input is re-input into the generator, which generates gated attention based on the new constraints, until all candidate terms generated by the generator satisfy the constraints. At this point, these candidate terms can be used as at least one standard term at the final transformation to continue the retrieval.

[0147] S50. Perform a search based on at least one standard term derived from the conversion.

[0148] The specific retrieval method used here is the same as that used in the prior art to perform retrieval using multiple keywords, and will not be described in detail here.

[0149] In one implementation, before performing a search based on at least one converted standard term, the at least one converted standard term can be output to the user for viewing. The user can then agree to or filter the search based on the at least one converted standard term.

[0150] The retrieval method provided by this invention for a technology manager platform, by combining the technical attributes of a domain knowledge graph with a generative adversarial terminology conversion engine, can intelligently and dynamically convert non-standard terms in user queries into standard terms that conform to technical attribute constraints such as patent relevance and interdisciplinary integration, significantly improving the accuracy and coverage of technology retrieval. The gating attention mechanism of the generator and the condition verification of the discriminator form a closed-loop optimization, which, together with the dynamic constraint adjustment of the adjuster, effectively solves the semantic gap problem between industry terms and general retrieval strategies currently faced by technology managers during retrieval.

[0151] To verify the effectiveness of the method of this invention, 100 patents in the field of new energy batteries were used as source data to construct a knowledge graph covering materials and electrochemistry with 200 nodes. Fifteen non-standard terms were also set to verify the effectiveness of the method. The performance was compared with traditional synonym expansion and BERT semantic retrieval methods. The experimental results are as follows:

[0152] Table 1. Comparison of Terminology Conversion Accuracy:

[0153]

[0154] Table 2 Generative Adversarial Optimization Efficiency

[0155]

[0156] The experimental results above demonstrate that, compared to existing technologies, this invention significantly improves terminology conversion accuracy, effectively addresses the semantic gap between industry jargon and standard terminology, and the technical constraints make the results more aligned with industry needs. Furthermore, the generative adversarial mechanism employed in this invention can increase the candidate term qualification rate to 96.3% within two iterations.

[0157] The retrieval method provided by this invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc. No limitation is made herein; any electronic device that can implement this invention falls within the protection scope of this invention.

[0158] Corresponding to the above-described retrieval method applied to a technology manager platform, the present invention also provides a retrieval device applied to a technology manager platform, see [link to relevant documentation]. Figure 3 As shown, the retrieval device includes:

[0159] User input module 301 is used to receive user input queries;

[0160] Decision module 302 is used to determine whether the user input query contains non-standard terms;

[0161] The generation module 303 is used to generate constraints based on the relevance between non-standard terms and nodes in the domain knowledge graph when the user input query contains non-standard terms; wherein, the constraints are used to constrain the technical attributes of the terms, and the attributes of the nodes in the domain knowledge graph include technical attributes, which at least include patent relevance, interdisciplinary degree and technology maturity.

[0162] The conversion module 304 is used to convert non-standard terms into at least one standard term based on non-standard terms and constraints using a generative adversarial term conversion engine.

[0163] The retrieval module 305 is used for retrieving data based on at least one standard term.

[0164] The generative adversarial terminology conversion engine includes a generator, a discriminator, and an adjuster. The generator generates gated attention based on constraints and generates at least one candidate term based on the gated attention and non-standard terms. The discriminator determines whether at least one candidate term satisfies the constraints. When all constraints are satisfied, at least one candidate term is output as at least one standard term. The adjuster adjusts the constraints and triggers the generator to regenerate at least one candidate term when the discriminator determines that a candidate term does not satisfy the constraints.

[0165] Based on the same inventive concept, embodiments of the present invention also provide an electronic device, such as... Figure 4 As shown, it includes a processor 401, a communication interface 402, a memory 403, and a communication bus 404, wherein the processor 401, the communication interface 402, and the memory 403 communicate with each other through the communication bus 404.

[0166] Memory 403 is used to store computer programs;

[0167] When the processor 401 executes the computer program stored in the memory 403, it implements the steps of the retrieval method applied to the technology manager platform described above.

[0168] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus.

[0169] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0170] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0171] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0172] The present invention also provides a computer-readable storage medium. A computer program is stored in this computer-readable storage medium, and when executed by a processor, the computer program implements the steps of the above-described retrieval method applied to a technology manager platform.

[0173] Optionally, the computer-readable storage medium may be non-volatile memory (NVM), such as at least one disk storage device.

[0174] Optionally, the computer-readable storage medium may also be at least one storage device located remotely from the aforementioned processor.

[0175] It should be noted that, for the device / electronic device / storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and relevant parts can be referred to in the description of the method embodiments.

[0176] It should be noted that the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention.

[0177] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0178] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings and the disclosure in carrying out the claimed invention. In the description of the invention, the word "comprising" does not exclude other components or steps, "a" or "an" does not exclude a plurality, and "a plurality" means two or more, unless otherwise explicitly specified. Furthermore, while different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce good results.

[0179] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus (devices), or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects, all of which are collectively referred to herein as "modules" or "systems." Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The computer program may be stored / distributed in a suitable medium, provided with or as part of other hardware, or may take other distribution forms, such as via the Internet or other wired or wireless telecommunications systems.

[0180] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0181] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0182] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0183] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A retrieval method applied to a technology manager platform, characterized in that, include: Receive user input queries; Determine whether the user-input query contains non-standard terms; When the user input query contains non-standard terms, constraints are generated based on the relevance between the non-standard terms and nodes in the domain knowledge graph; wherein, the constraints are used to constrain the technical attributes of the terms, and the attributes of the nodes in the domain knowledge graph include the technical attributes, which at least include patent relevance, interdisciplinary degree, and technology maturity. Based on the non-standard terminology and the constraints, the non-standard terminology is converted into at least one standard term using a generative adversarial terminology conversion engine. Search according to at least one of the standard terms; The generative adversarial terminology conversion engine includes a generator, a discriminator, and an adjuster. The generator generates gated attention based on the constraints and generates at least one candidate term based on the gated attention and the non-standard terminology. The discriminator determines whether all candidate terms satisfy the constraints. When all constraints are satisfied, the at least one candidate term is output as the at least one standard term. The adjuster adjusts the constraints and triggers the generator to regenerate at least one candidate term when the discriminator determines that a candidate term does not satisfy the constraints. The determination of whether the user-input query contains non-standard terms includes: The user-input query is subjected to terminology extraction, and it is determined whether the extracted terms contain non-standard terms. If the extracted terms do not contain non-standard terms, further determine whether the non-standard term matches the non-standard term-standard term mapping table. If not, determine that the user input query contains non-standard terms. The technical attributes are encoded as technical attribute vectors; the constraints generated based on the relevance of nodes in the non-standard terminology and domain knowledge graph include: From the non-standard terminology-standard terminology mapping table, obtain the target standard terminology that is semantically and / or structurally similar to the non-standard terminology; From the domain knowledge graph, standard terms with a node similarity greater than a threshold with the target standard term are selected to obtain the first group of candidate nodes; Cluster the first group of candidate nodes, and further filter the first group of candidate nodes based on the clustering results to obtain the second group of candidate nodes; A constraint vector is constructed based on the second group of candidate nodes as the constraint condition; The generator generates gated attention based on the constraints, including: The non-standard terms are embedded into non-standard term encoding vectors using a word embedding tool, and the non-standard term encoding vectors and the constraint vectors are concatenated into a joint vector. ； in, This represents the encoding vector of the non-standard terminology. Represents the constraint vector; The joint vector is linearly transformed using the first weight matrix and the first bias matrix to obtain the gated vector: ； in, This is the first weight matrix used to control the feature mixing ratio; It is the first bias matrix used to control the gating sensitivity. and All were obtained through pre-training; Based on the gate vector, the gate weights are generated using the Sigmoid function: ； in, This represents the Sigmoid function.

2. The retrieval method according to claim 1, characterized in that, The generator generates at least one candidate term based on the gated attention and the non-standard terminology, including: Projecting the non-standard term encoding vector yields the term projection vector: ； in, This is the second weight matrix used to filter term noise. It is the second bias vector that controls the activation threshold of term features. and All were obtained through pre-training; Projecting the constraint vector yields the constraint projection vector: in, This represents the third weight matrix used to reinforce the domain preset. It is the third bias vector used to control the intensity of domain knowledge injection. and All were obtained through pre-training; for Activation function; Based on the term projection vector and the constraint projection vector, the candidate term feature vector is output using the gating weights: in, It represents the Hadamardi (or Hadama) stack; Based on the feature vectors of the candidate terms, a domain knowledge graph is searched to obtain at least one candidate term.

3. The retrieval method according to claim 1, characterized in that, The discriminator determines whether the at least one candidate term satisfies the constraint condition, including: For each candidate term, calculate its connectivity score: ； in, Candidate terms obtained from the domain knowledge graph The set of k-hop neighbor nodes, This represents the second group of candidate nodes; Indicates taking and The size of the intersection; For each candidate term, calculate its technical attribute score: ； in, This refers to the non-standard terminology. This is the technical attribute score vector for the non-standard terminology. This is a pre-trained technology attribute prediction model, trained based on the domain knowledge graph, used to predict the patent relevance of terms. Interdisciplinary degree and technology maturity ; This represents a vector composed of the patent relevance, interdisciplinary nature, and technological maturity of candidate terms. The formula for calculating vector similarity; Based on the connectivity score and technical field score of each candidate term, determine whether the selected term satisfies the constraints: ； in, and The preset scaling factor, and , This indicates the result of determining whether a candidate term meets the constraints. This indicates that the judgment result is yes. This indicates that the judgment result is negative.

4. The retrieval method according to claim 3, characterized in that, The adjuster adjusts the constraints, including: Calculate the adversarial vector based on candidate terms: ； in, , , The subset of candidate terms that satisfy the constraint clause from the at least one candidate term. The subset of candidate terms that do not satisfy the constraint clause among the at least one candidate term. The subset of candidate terms that satisfy the constraint clause from the at least one candidate term. This is the mean vector of the technical attribute vectors of all candidate terms. This is the mean vector of the technical attribute vectors of the second group of candidate nodes; Based on the adversarial vector, reconstruct the constraint vector: ； in, For the amplitude limiting function, The threshold value for the amplitude limit. The reconstructed constraint vector.

5. A retrieval device applied to a technology manager platform, characterized in that, include: The user input module is used to receive user input queries; The judgment module is used to determine whether the user input query contains non-standard terms; A generation module is used to generate constraints based on the relevance between the non-standard terms and nodes in the domain knowledge graph when the user input query contains non-standard terms; wherein, the constraints are used to constrain the technical attributes of the terms, and the attributes of the nodes in the domain knowledge graph include the technical attributes, which at least include patent relevance, interdisciplinary degree, and technology maturity. A conversion module is used to convert the non-standard terms into at least one standard term based on the non-standard terms and the constraints using a generative adversarial term conversion engine. A retrieval module is used to retrieve data based on the at least one standard term. The generative adversarial terminology conversion engine includes a generator, a discriminator, and an adjuster. The generator generates gated attention based on the constraints and generates at least one candidate term based on the gated attention and the non-standard terminology. The discriminator determines whether all candidate terms satisfy the constraints. When all constraints are satisfied, the at least one candidate term is output as the at least one standard term. The adjuster adjusts the constraints and triggers the generator to regenerate at least one candidate term when the discriminator determines that a candidate term does not satisfy the constraints. The determination of whether the user-input query contains non-standard terms includes: The user-input query is subjected to terminology extraction, and it is determined whether the extracted terms contain non-standard terms. If the extracted terms do not contain non-standard terms, further determine whether the non-standard term matches the non-standard term-standard term mapping table. If not, determine that the user input query contains non-standard terms. The technical attributes are encoded as technical attribute vectors; the constraints generated based on the relevance of nodes in the non-standard terminology and domain knowledge graph include: From the non-standard terminology-standard terminology mapping table, obtain the target standard terminology that is semantically and / or structurally similar to the non-standard terminology; From the domain knowledge graph, standard terms with a node similarity greater than a threshold with the target standard term are selected to obtain the first group of candidate nodes; Cluster the first group of candidate nodes, and further filter the first group of candidate nodes based on the clustering results to obtain the second group of candidate nodes; A constraint vector is constructed based on the second group of candidate nodes as the constraint condition; The generator generates gated attention based on the constraints, including: The non-standard terms are embedded into non-standard term encoding vectors using a word embedding tool, and the non-standard term encoding vectors and the constraint vectors are concatenated into a joint vector. ； in, This represents the encoding vector of the non-standard terminology. Represents the constraint vector; The joint vector is linearly transformed using the first weight matrix and the first bias matrix to obtain the gated vector: ； in, This is the first weight matrix used to control the feature mixing ratio; It is the first bias matrix used to control the gating sensitivity. and All were obtained through pre-training; Based on the gate vector, the gate weights are generated using the Sigmoid function: ； in, This represents the Sigmoid function.

6. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of the retrieval method applied to a technology manager platform as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the retrieval method for a technology manager platform as described in any one of claims 1 to 4.

Images