Technology transfer-oriented term expansion retrieval method, device, and electronic equipment

Abstract

This invention belongs to the field of computer information retrieval technology and discloses a term expansion retrieval method, apparatus, and electronic device for technology transfer. The method includes: expanding user-input queries with keywords; determining multiple target windows based on the co-occurrence relationship between the expanded keyword set and technical terms within different time windows in a patent database; extracting the Top-N most frequently co-occurring terms from each window and generating a period constraint vector using a word embedding model; generating a set of candidate terms under the constraints of the period constraint vector corresponding to each target window based on the user-input query; calculating the time importance score and time span score of the candidate terms based on their distribution across time windows; and performing a retrieval after filtering the candidate terms based on the time importance score, time span score, and semantic similarity score. This invention solves the problem of temporal semantic discontinuity in existing retrieval methods.

Description

Technical Field

[0001] This invention belongs to the field of computer information retrieval technology, specifically relating to a term expansion retrieval method, device, and electronic device for technology transfer. Background Technology

[0002] In the field of technology transfer and commercialization, one of the core functions of technology manager platforms is to help users accurately match technology supply and demand information. However, technology itself is not static; the meaning of its terminology and the set of related technologies evolve significantly over time. This leads to a temporal semantic discontinuity problem. For example, searching for "blockchain" using existing systems will focus on currently popular terms like "smart contracts" and "DeFi," while core concepts from a few years ago, such as "distributed ledgers" and "cryptocurrencies," may be overlooked. This makes the search results unable to reflect the true development history of the technology and difficult to meet in-depth needs such as technology tracing and trend analysis.

[0003] Therefore, the core problem facing current technology retrieval has evolved from static semantic understanding to the perception of dynamic semantic evolution. There is an urgent need for a solution that can understand technological evolution and perform dynamic terminology conversion and retrieval at low cost and high efficiency. Summary of the Invention

[0004] To address the aforementioned problems in the prior art, this invention provides a term expansion retrieval method, apparatus, and electronic device for technology transfer.

[0005] The technical problem to be solved by this invention is achieved through the following technical solution:

[0006] In a first aspect, the present invention provides a terminology expansion retrieval method for technology transfer, comprising:

[0007] Receive user input queries, preprocess the user input queries and expand the keywords to obtain an expanded keyword set;

[0008] Based on the co-occurrence relationship between the extended keyword set and technical terms in different time windows in the patent database, multiple target windows are determined;

[0009] From the patent data of each target window, extract the Top-N terms that co-occur most frequently with the extended keyword set; based on the Top-N terms, use a word embedding model to generate a period constraint vector corresponding to the target window, so as to constrain the technical context of the target window;

[0010] Based on the user input query, a set of candidate terms is generated under the constraints of the period constraint vector corresponding to each target window, and the candidate terms under each target window are summarized to obtain a candidate term set.

[0011] For each candidate term in the candidate term set, its time importance score and time span score are calculated based on its distribution in each time window.

[0012] The candidate term set is filtered based on the time importance score, time span score, and semantic similarity score between the candidate terms and the user input query, and the retrieval is performed based on the filtered candidate terms.

[0013] Secondly, the present invention provides a terminology expansion retrieval device for technology transfer, comprising:

[0014] The user receiving module is used to receive user input queries, preprocess the user input queries and expand the keywords to obtain an expanded keyword set.

[0015] The window determination module is used to determine multiple target windows based on the co-occurrence relationship between the extended keyword set and technical terms in different time windows in the patent database;

[0016] The constraint generation module is used to extract the Top-N terms that co-occur most frequently with the extended keyword set from the patent data of each target window; based on the Top-N terms, a word embedding model is used to generate a period constraint vector corresponding to the target window to constrain the technical context of the target window.

[0017] The candidate term generation module is used to generate a set of candidate terms under the constraints of the period constraint vector corresponding to each target window according to the user input query, and to summarize the candidate terms under each target window to obtain a candidate term set.

[0018] The score calculation module is used to calculate the time importance score and time span score for each candidate term in the candidate term set, based on its distribution in each time window.

[0019] The filtering and retrieval module is used to filter the candidate term set based on the time importance score, time span score, and semantic similarity score between the candidate terms and the user input query, and to perform a retrieval based on the filtered candidate terms.

[0020] Thirdly, the present invention 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;

[0021] Memory, used to store computer programs;

[0022] When a processor executes a computer program stored in memory, it implements the steps of the aforementioned term expansion retrieval method for technology transfer.

[0023] The terminology expansion retrieval method for technology transfer provided by this invention effectively solves the problem of temporal semantic discontinuity in existing technologies by introducing a dynamic analysis mechanism with multiple time windows. Specifically, this invention locks multiple key time windows based on the co-occurrence relationship between expanded keywords and historical patent data, thereby constructing a dynamically evolving technological context framework. Then, by extracting high-frequency terms from each window and generating period constraint vectors, a precise semantic constraint model is established for each historical stage. This enables the system to recreate the technological language environment of different periods and avoid retrieval bias caused by semantic changes in terms. Next, by generating candidate terms under the constraints of each period, and innovatively introducing two quantitative indicators, time importance and time span, for evaluation, the selected term set can reflect both the explosive impact of technology in a specific period and its continuity and universality throughout the entire development process. Finally, these insights from the time dimension are combined with traditional semantic similarity scores for comprehensive screening, so that the final retrieval results are not only semantically relevant but also clearly outline the complete development trajectory of a technology—including current hot topics and core concepts that were crucial in history. This greatly improves the search quality in in-depth applications such as technology transfer and trend tracing.

[0024] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0025] Figure 1 This is a flowchart of a term expansion retrieval method for technology transfer provided by the present invention;

[0026] Figure 2 This is a technical roadmap for a term expansion retrieval method for technology transfer provided by the present invention;

[0027] Figure 3 This is a schematic diagram of the structure of a terminology expansion retrieval device for technology transfer provided by the present invention;

[0028] Figure 4 This is a schematic diagram of the structure of an electronic device provided by the present invention. Detailed Implementation

[0029] 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.

[0030] To address the core problems currently faced in technology retrieval, this invention provides a terminology expansion retrieval method oriented towards technology transfer. (See [link to relevant documentation]). Figure 1 and Figure 2 As shown, the method includes the following steps:

[0031] S10. Receive user input query, preprocess the user input query and expand the keywords to obtain an expanded keyword set.

[0032] Specifically, the preprocessing of user input queries includes text cleaning, word segmentation, and stop word removal. Subsequently, based on the preprocessed terms, core terms and related words that are semantically related, technically equivalent, or subordinate are expanded through methods such as querying domain knowledge graphs, technical thesaurus, or utilizing existing mature large-scale models, thereby constructing an expanded keyword set. For example, expanded keywords can be obtained by querying domain knowledge graphs such as the keyword table accompanying the International Patent Classification or the Derwent manual code table.

[0033] S20. Based on the co-occurrence relationship between the extended keyword set and technical terms in different time windows in the patent database, determine multiple target windows.

[0034] Specifically, in one implementation, multiple target windows are determined based on the co-occurrence relationship between the extended keyword set and technical terms in different time windows of the patent database, which may include:

[0035] (1) Based on the extended keyword set, the patent data in each time window is retrieved to obtain a patent set containing terms from the extended keyword set.

[0036] (2) Calculate the score for each time window based on the co-occurrence of technical terms between the extended keyword set and the patent set for each time window:

[0037] ;

[0038] In the formula, Indicates time window The score; Indicates an expanded set of keywords; Represents an expanded set of keywords Keywords in; Indicates within the time window The collection of patents; Represents a patent set The collection of all distinct technical terms contained therein. In the patent collection Terms in With keywords The number of times they appear together in the same patent document; Keywords In patent collection The document frequency and co-occurrence count can be obtained by parsing the patent database, constructing a text index, and statistically analyzing the co-occurrence of terms within a specified text range and document frequency.

[0039] (3) Filter multiple target windows based on the scores of each time window.

[0040] Specifically, select the time windows with the highest scores as the target windows.

[0041] In this implementation, normalization effectively reduces the random interference of frequent co-occurrence in a few patents, and more robustly reflects the broad and stable correlation formed by technical terms in different time windows. The results focus more on the mainstream trend of technological development.

[0042] In another implementation, The calculation method can also be:

[0043] .

[0044] Understandably, compared to The calculation formula in the current implementation is more intuitive and easier to calculate than the previous calculation method, and it can also achieve the effect of capturing co-occurrence relationships.

[0045] S30. Extract and expand the Top-N terms that co-occur most frequently from the patent data of each target window; based on the Top-N terms, use a word embedding model to generate the period constraint vector corresponding to the target window, so as to constrain the technical context of the target window.

[0046] Specifically, after extracting and expanding the top-N most frequently co-occurring terms from the patent data of each target window, a pre-trained word embedding model can be used to embed each of these top-N terms into a vector. Based on each embedding vector, a period constraint vector corresponding to the target window is generated. This can be achieved by directly calculating the arithmetic mean of the embedding vectors, or by performing a weighted average based on the TF-IDF score or co-occurrence frequency of the term corresponding to each embedding vector within the target window. In this invention, the word embedding model can be BERT, RoBERTa, or a variant thereof, but is not limited to these.

[0047] Alternatively, in one implementation, the word embedding model can be trained using the patent text within each target window. The resulting model's vector space can more accurately encode and reflect the technical context and terminology usage habits of that specific period, thereby making the generated period constraint vector more accurate in its semantic constraint on that window.

[0048] S40. Based on the user's input query, generate a set of candidate terms under the constraints of the period constraint vector corresponding to each target window, and summarize the candidate terms under each target window to obtain a candidate term set.

[0049] Specifically, based on the user's input query, a set of candidate terms is generated under the constraints of the period constraint vector corresponding to each target window, including:

[0050] (1) Encode the user input query into a query vector and project the period constraint vector into the same vector space as the query vector.

[0051] Understandably, since the query vector and the period constraint vector may originate from different models or training data, their vector spaces are inconsistent, making direct similarity comparison impossible. Therefore, it is necessary to project them into the same space to ensure the accuracy and effectiveness of similarity calculation. Specifically, the user input query is encoded into a query vector using a word embedding model, and the period constraint vector is projected into the same vector space as the query vector using a linear transformation (learning the projection matrix).

[0052] (2) Determine the relevant field based on the user's input query.

[0053] Specifically, by extracting terms from user-input queries and constructing simple search queries to search the patent database, relevant fields can be identified based on the search results. For example, by statistically analyzing the International Patent Classification (IPC) numbers of patents in the search results, the most frequently occurring main classifications or classification groups can be identified as relevant fields.

[0054] (3) Calculate the comprehensive relevance score of each term in the terminology database within the relevant domain to the query vector and the projected period constraint vector:

[0055] ;

[0056] In the formula, Represents the query vector. This represents the encoded vector corresponding to a term in a terminology database within a relevant domain, obtained through word embedding tools. This represents the time constraint vector after projection. This represents a function for calculating similarity scores. and They are respectively and The corresponding weights.

[0057] (4) Based on the comprehensive relevance score, select a set of candidate terms from the terminology database in the relevant field.

[0058] Specifically, terms with high comprehensive relevance scores are selected from a terminology database in the relevant field as candidate terms.

[0059] Alternatively, in one implementation, and It does not have to be a pre-specified value; it can also be determined in the following ways:

[0060] (a) Concatenate the query vector and the period constraint vector to obtain a joint vector;

[0061] (b) Input the joint vector into a pre-trained feedforward neural network to make it predict and The value;

[0062] The feedforward neural network is trained based on the following loss function:

[0063] ;

[0064] In the formula, As a preset positive scalar, , It is the overall relevance score of positive samples. This is the comprehensive relevance score of the negative samples. Positive and negative samples are joint vector samples, constructed in the same way as joint vectors. The difference between positive and negative samples is that the query vector and period constraint vector forming the positive samples both come from the same time window, share the same domain, and contain frequently co-occurring terms. In contrast, the query vector and period constraint vector forming the positive samples come from different time windows with significantly different domains, or, while their domains are similar, they have a low co-occurrence frequency. Furthermore, the calculation... and The comprehensive relevance score used and The positive and negative samples are respectively input into the feedforward neural network during training, and then predicted by the feedforward neural network.

[0065] In this implementation, dynamic prediction is achieved through a feedforward neural network. and The numerical values ​​can adaptively adjust the relative importance of query similarity and period conformity in candidate term generation based on the current query and period context. Compared with a fixed weight strategy, this significantly improves the contextual relevance and flexibility of the candidate term set.

[0066] S50. For each candidate term in the candidate term set, calculate its time importance score and time span score based on its distribution in each time window.

[0067] Specifically, for each candidate term in the candidate term set, its time span score is calculated using the following formula:

[0068] ;

[0069] In the formula, The target window size for the distribution of candidate terms. The target number of windows;

[0070] Furthermore, for each candidate term in the candidate term set, its time importance score is calculated using the following formula:

[0071] ;

[0072] In the formula, The target window set for the distribution of candidate terms. Target window for candidate term distribution The score, This represents the summation of scores over all time windows.

[0073] S60. Filter the candidate term set based on the time importance score, time span score, and semantic similarity score between the candidate terms and the user input query, and perform retrieval based on the filtered candidate terms.

[0074] Specifically, a comprehensive score for candidate terms is calculated based on their temporal importance score, time span score, and semantic similarity score between the candidate terms and the user's input query.

[0075] ;

[0076] In the formula, This represents the semantic similarity score between the candidate terms and the user-input query. This is for , , The preset weighting factor, and For example, =0.5, =0.3, =0.2, but it is not limited to this; in practice, it can be set accordingly based on the actual search purpose. The value.

[0077] Then, the candidate term set is filtered based on the overall score, and the candidate terms with higher overall scores are selected for retrieval.

[0078] As can be seen from the process shown in steps S10 to S60 above, this invention realizes a time-aware term expansion retrieval. It constructs a dynamic technological context through multi-window analysis, generates candidate terms under the constraints of this context, and finally performs multi-dimensional filtering based on the distribution characteristics of terms throughout the entire period. This ensures that the final set of terms used for retrieval can accurately match the user's current intent and fully cover the historical development of the technology, thereby fundamentally overcoming the time-series semantic discontinuity problem described in the background art.

[0079] The effectiveness of the present invention will be further illustrated by the following experiments.

[0080] Experimental setup:

[0081] From the USPTO (United States Patent and Trademark Office) public database, "image compression" was selected as a typical technical field. Patent documents closely related to this field from 2000 to 2020, totaling 1000, were screened and collected as the test set for this experiment, with a time window set to 4. "Image compression" was selected as the core field for user input query execution.

[0082] Comparison Model:

[0083] Model 1 (No Period Constraint): When generating candidate terms, the period constraint vector is removed, and only the semantic similarity between the query vector and the term is relied upon.

[0084] Model 2 (without time filtering): In the final filtering, the time importance score and time span score are removed, and the ranking is based solely on the semantic similarity score.

[0085] Baseline model: The classic retrieval model based on global term frequency (TF-IDF) is used as the baseline for performance comparison.

[0086] The experimental results are shown in Table 1:

[0087] Table 1

[0088]

[0089] In Table 1, retrieval precision refers to the proportion of relevant terms in the first 10 returned results, determined manually by domain experts. Temporal coverage refers to the number of different time windows covered by the first 10 results. This metric is used to conservatively measure the diversity of search results over time.

[0090] As shown in Table 1, the model of this invention achieved relatively optimal performance, indicating that the overall architecture of this invention is effective. Its retrieval results not only have the highest relevance but also better cover different stages of technological development. The performance degradation was most significant after removing the period constraint vector (Model 1). This demonstrates that limiting the search to a specific historical context is crucial for retrieving core terms at that time, directly addressing the problem of temporal semantic discontinuity. After removing the time score filter (Model 2), the temporal coverage decreased significantly. This indicates that semantic similarity alone is insufficient to identify terms of diachronic importance; quantitative evaluation of the time dimension provides an indispensable filtering dimension. Furthermore, Table 1 also shows that the performance of all simplified models of this invention is not significantly different from the baseline model in terms of retrieval accuracy, but they all outperform the baseline model in terms of temporal coverage, demonstrating the advancement of this invention. However, none of them reached the level of the complete model.

[0091] In summary, the terminology expansion retrieval method for technology transfer provided by this invention identifies multiple key time windows based on the co-occurrence relationship between expanded keywords and historical patent data, thereby constructing a dynamically evolving technological context framework. Next, by extracting high-frequency terms from each window and generating period constraint vectors, a precise semantic constraint model is established for each historical stage. This enables the system to recreate the technological language environment of different periods, avoiding retrieval biases caused by semantic changes in terms. Then, by generating candidate terms under the constraints of each period and innovatively introducing two quantitative indicators—time importance and time span—for evaluation, the selected term set reflects both the explosive impact of the technology in a specific period and its continuity and universality throughout its development. Finally, these insights from the time dimension are combined with traditional semantic similarity scores for comprehensive screening, ensuring that the final retrieval results are not only semantically relevant but also clearly outline the complete development trajectory of a technology—including current hot topics and core concepts that were historically crucial. This significantly improves retrieval quality in in-depth applications such as technology transfer and trend tracing.

[0092] The method provided in 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.

[0093] Corresponding to the above-described terminology expansion retrieval method for technology transfer, the present invention also provides a terminology expansion retrieval device for technology transfer, see [link to device]. Figure 3 As shown, the device includes:

[0094] User receiving module 301 is used to receive user input queries, preprocess the user input queries and expand keywords to obtain an expanded keyword set;

[0095] The window determination module 302 is used to determine multiple target windows based on the co-occurrence relationship between the extended keyword set and technical terms in different time windows in the patent database;

[0096] The constraint generation module 303 is used to extract the Top-N terms that co-occur most frequently with the extended keyword set from the patent data of each target window; based on the Top-N terms, a period constraint vector corresponding to the target window is generated using a word embedding model to constrain the technical context of the target window;

[0097] The candidate term generation module 304 is used to generate a set of candidate terms under the constraints of the period constraint vector corresponding to each target window according to the user input query, and to summarize the candidate terms under each target window to obtain a candidate term set.

[0098] The score calculation module 305 is used to calculate the time importance score and time span score for each candidate term in the candidate term set, based on its distribution in each time window.

[0099] The filtering and retrieval module 306 is used to filter the candidate term set based on the time importance score, time span score, and semantic similarity score between the candidate terms and the user input query, and to perform a retrieval based on the filtered candidate terms.

[0100] Optionally, the window determination module 302 is specifically used for:

[0101] Based on the extended keyword set, the patent data within each time window is retrieved to obtain a patent set containing terms from the extended keyword set;

[0102] The score for each time window is calculated based on the co-occurrence of technical terms between the expanded keyword set and the patent set for each time window.

[0103]

[0104] In the formula, Indicates time window The score; This refers to the extended keyword set; Represents an expanded set of keywords Keywords in; Indicates within the time window The collection of patents; Represents a patent set The collection of all distinct technical terms contained therein. In the patent collection Terms in With keywords The number of times they appear together in the same patent document; Keywords In patent collection The frequency of documents appearing in;

[0105] Multiple target windows are filtered based on the scores of each time window.

[0106] Optionally, the constraint generation module 303 generates a period constraint vector corresponding to the target window based on the Top-N terms using a word embedding model, including:

[0107] Using a pre-trained word embedding model, each of the Top-N terms is embedded as a vector, and the period constraint vector corresponding to the target window is generated based on each embedding vector.

[0108] Optionally, the candidate term generation module 304 generates a set of candidate terms based on the user input query, under the constraints of the period constraint vector corresponding to each target window, including:

[0109] The user input query is encoded into a query vector, and the period constraint vector is projected onto the same vector space as the query vector.

[0110] Based on the user's input query, determine the relevant field;

[0111] Calculate the comprehensive relevance score of each term in the terminology library within the relevant domain to the query vector and the projected period constraint vector:

[0112] ;

[0113] In the formula, Represents the query vector. This represents the encoding vector corresponding to a term in a terminology database within a relevant domain. This represents the time constraint vector after projection. This represents a function for calculating similarity scores. and They are respectively and The corresponding weights;

[0114] Based on the comprehensive relevance score, a set of candidate terms is selected from the terminology database in the relevant field.

[0115] Optionally, and Determined in the following manner:

[0116] The query vector and the period constraint vector are concatenated to obtain a joint vector;

[0117] The joint vector is input into a pre-trained feedforward neural network to make predictions. and The value;

[0118] The feedforward neural network is trained based on the following loss function:

[0119] ;

[0120] In the formula, As a preset positive scalar, , It is the overall relevance score of positive samples. It is the comprehensive relevance score of negative samples, calculated. and The comprehensive relevance score used and The positive and negative samples are respectively input into the feedforward neural network during training, and then predicted by the feedforward neural network.

[0121] Optionally, the fraction calculation module 305 is specifically used for:

[0122] For each candidate term in the candidate term set, its time span score is calculated using the following formula:

[0123] ;

[0124] In the formula, The target window size for the distribution of candidate terms. The target number of windows;

[0125] For each candidate term in the candidate term set, its time importance score is calculated using the following formula:

[0126] ;

[0127] In the formula, The target window set for the distribution of candidate terms. Target window for candidate term distribution The score, This represents the summation of scores over all time windows.

[0128] Optionally, the filtering and retrieval module 306 filters the candidate term set based on the candidate terms' time importance score, time span score, and semantic similarity score between the candidate terms and the user-input query, including:

[0129] A comprehensive score for the candidate terms is calculated based on their temporal importance score, time span score, and semantic similarity score to the user-input query.

[0130] ;

[0131] In the formula, This represents the semantic similarity score between the candidate terms and the user-input query. This is for , , The preset weighting factor, and ;

[0132] The candidate term set is filtered based on the comprehensive score.

[0133] The terminology expansion retrieval device for technology transfer provided by this invention effectively solves the problem of temporal semantic discontinuity in existing technologies by introducing a dynamic analysis mechanism with multiple time windows.

[0134] 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 601, a communication interface 602, a memory 603, and a communication bus 604, wherein the processor 601, the communication interface 602, and the memory 603 communicate with each other through the communication bus 604.

[0135] Memory 603 is used to store computer programs;

[0136] When the processor 601 executes the program stored in the memory 603, it implements the steps of any of the above-described term expansion retrieval methods for technology transfer.

[0137] 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.

[0138] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0139] 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.

[0140] 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.

[0141] It should be noted that, for the device / electronic device 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.

[0142] 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.

[0143] 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.

[0144] 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.

[0145] 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.

[0146] 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.

[0147] 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.

[0148] 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.

[0149] 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 terminology expansion retrieval method for technology transfer, characterized in that, include: Receive user input queries, preprocess the user input queries and expand the keywords to obtain an expanded keyword set; Based on the co-occurrence relationship between the extended keyword set and technical terms in different time windows in the patent database, multiple target windows are determined; From the patent data of each target window, extract the Top-N terms that co-occur most frequently with the extended keyword set; based on the Top-N terms, use a word embedding model to generate a period constraint vector corresponding to the target window, so as to constrain the technical context of the target window; Based on the user input query, a set of candidate terms is generated under the constraints of the period constraint vector corresponding to each target window, and the candidate terms under each target window are summarized to obtain a candidate term set. For each candidate term in the candidate term set, its time importance score and time span score are calculated based on its distribution in each time window. The candidate term set is filtered based on the time importance score, time span score, and semantic similarity score between the candidate terms and the user input query, and the retrieval is performed based on the filtered candidate terms. The step involves determining multiple target windows based on the co-occurrence relationship between the expanded keyword set and technical terms within different time windows in the patent database, including: Based on the extended keyword set, the patent data within each time window is retrieved to obtain a patent set containing terms from the extended keyword set; The score for each time window is calculated based on the co-occurrence of technical terms between the extended keyword set and the patent set for each time window. The calculation formula is as follows: ， Or: ; In the formula, Indicates time window The score; This refers to the extended keyword set; Represents an expanded set of keywords Keywords in; Indicates within the time window The collection of patents; Represents a patent set The collection of all distinct technical terms contained therein. In the patent collection Terms in With keywords The number of times they appear together in the same patent document; Keywords In patent collection The frequency of documents appearing in; Multiple target windows are filtered based on the scores of each time window.

2. The terminology expansion retrieval method for technology transfer according to claim 1, characterized in that, The step of generating the period constraint vector corresponding to the target window based on the Top-N terms using a word embedding model includes: Using a pre-trained word embedding model, each of the Top-N terms is embedded as a vector, and the period constraint vector corresponding to the target window is generated based on each embedding vector.

3. The terminology expansion retrieval method for technology transfer according to claim 1, characterized in that, Based on the user input query, a set of candidate terms is generated under the constraints of the period constraint vector corresponding to each target window, including: The user input query is encoded into a query vector, and the period constraint vector is projected onto the same vector space as the query vector. Based on the user's input query, determine the relevant field; Calculate the comprehensive relevance score of each term in the terminology library within the relevant domain to the query vector and the projected period constraint vector: ； In the formula, Represents the query vector. This represents the encoding vector corresponding to a term in a terminology database within a relevant domain. This represents the time constraint vector after projection. This represents a function for calculating similarity scores. and They are respectively and The corresponding weights; Based on the comprehensive relevance score, a set of candidate terms is selected from the terminology database in the relevant field.

4. The terminology expansion retrieval method for technology transfer according to claim 3, characterized in that, and Determined in the following manner: The query vector and the period constraint vector are concatenated to obtain a joint vector; The joint vector is input into a pre-trained feedforward neural network to make predictions. and The value; The feedforward neural network is trained based on the following loss function: ； In the formula, As a preset positive scalar, , It is the overall relevance score of positive samples. It is the comprehensive relevance score of negative samples, calculated. and The comprehensive relevance score used and The positive and negative samples are respectively input into the feedforward neural network during training, and then predicted by the feedforward neural network.

5. The terminology expansion retrieval method for technology transfer according to claim 1, characterized in that, For each candidate term in the candidate term set, its time importance score and time span score are calculated based on its distribution across various time windows, including: For each candidate term in the candidate term set, its time span score is calculated using the following formula: ； In the formula, The target window size for the distribution of candidate terms. The target number of windows; For each candidate term in the candidate term set, its time importance score is calculated using the following formula: ； In the formula, The target window set for the distribution of candidate terms. Target window for candidate term distribution The score, This represents the summation of scores over all time windows.

6. The terminology expansion retrieval method for technology transfer according to claim 1, characterized in that, The process of filtering the candidate term set based on the candidate terms' temporal importance score, time span score, and semantic similarity score between the candidate terms and the user's input query includes: A comprehensive score for the candidate terms is calculated based on their temporal importance score, time span score, and semantic similarity score to the user-input query. ； In the formula, This represents the semantic similarity score between the candidate terms and the user-input query. This is for , , The preset weighting factor, and ; The candidate term set is filtered based on the comprehensive score.

7. A terminology expansion retrieval device for technology transfer, characterized in that, include: The user receiving module is used to receive user input queries, preprocess the user input queries and expand the keywords to obtain an expanded keyword set. The window determination module is used to determine multiple target windows based on the co-occurrence relationship between the extended keyword set and technical terms in different time windows in the patent database; The constraint generation module is used to extract the Top-N terms that co-occur most frequently with the extended keyword set from the patent data of each target window; based on the Top-N terms, a word embedding model is used to generate a period constraint vector corresponding to the target window to constrain the technical context of the target window. The candidate term generation module is used to generate a set of candidate terms under the constraints of the period constraint vector corresponding to each target window according to the user input query, and to summarize the candidate terms under each target window to obtain a candidate term set. The score calculation module is used to calculate the time importance score and time span score for each candidate term in the candidate term set, based on its distribution in each time window. The filtering and retrieval module is used to filter the candidate term set based on the time importance score, time span score, and semantic similarity score between the candidate terms and the user input query, and to perform a retrieval based on the filtered candidate terms. The window determination module is specifically used for: The patent data within each time window is retrieved based on the extended keyword set to obtain a patent set containing terms from the extended keyword set. The score for each time window is calculated based on the co-occurrence of technical terms between the extended keyword set and the patent set within that time window, using the following formula: ， Or: ; In the formula, Indicates time window The score; This refers to the extended keyword set; Represents an expanded set of keywords Keywords in; Indicates within the time window The collection of patents; Represents a patent set The collection of all distinct technical terms contained therein. In the patent collection Terms in With keywords The number of times they appear together in the same patent document; Keywords In patent collection The frequency of documents appearing in; Multiple target windows are filtered based on the scores of each time window.

8. 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; A processor, when executing a computer program stored in memory, implements the steps of the term expansion retrieval method for technology transfer as described in any one of claims 1 to 6.

Images