Construction method and device of hierarchical term set, computer equipment, storage medium and computer program product

Abstract

The invention relates to a hierarchical term set construction method and device, computer equipment, a storage medium and a computer program product, and the method comprises the steps: carrying out the term extraction of a target text according to at least three term extraction strategies, and obtaining a plurality of terms under each term extraction strategy, and then constructing term sets of different levels according to a plurality of terms under each term extraction strategy, and finally classifying all the terms under the term sets of different levels to obtain classification results of all the terms. According to the method, the target text content can be deeply analyzed from multiple different dimensions by adopting multiple term extraction strategies, so that the terms are extracted more comprehensively, meanwhile, the term coverage range is increased, the mutual verification and supplement effects can be realized, and the term extraction accuracy is improved.

Description

Technical Field

[0001] The present application relates to the technical field of natural language processing, and in particular to a method, apparatus, computer equipment, storage medium and computer program product for constructing a hierarchical term set. Background Art

[0002] With the rapid development of Internet technology, Natural Language Processing (NLP) has been widely used in text mining, large language model (LLM) and patent intelligent analysis due to its powerful semantic understanding and information extraction capabilities. In the field of patent intelligent analysis, due to the high knowledge density and terminology density of patents and the unstructured characteristics of text, it is particularly important to classify the terms in a large number of patent documents into target fields.

[0003] At present, when classifying terms in patent documents, it is often based on a single rule or deep learning technology to extract and remove duplicate terms to achieve term classification. However, the above information classification method has the problem of inaccuracy. Summary of the invention

[0004] Based on this, it is necessary to provide a method, apparatus, computer device, storage medium and computer program product for constructing a hierarchical term set that can improve classification accuracy in response to the above technical problems.

[0005] In a first aspect, the present application provides a method for constructing a hierarchical term set, the method comprising:

[0006] According to at least three term extraction strategies, term extraction is performed on the target text to obtain multiple terms under each term extraction strategy;

[0007] Constructing term sets of different levels according to multiple terms under each term extraction strategy; term sets of different levels include sub-word term sets, full-word term sets and cluster term sets;

[0008] All terms under different levels of term sets are classified to obtain classification results of all terms; the classification results include chip screen patent terms, non-chip screen patent terms and uncertainties.

[0009] In one embodiment, at least three term extraction strategies include a word segmentation processing strategy, a language model extraction strategy, and a neural network extraction strategy. According to the at least three term extraction strategies, term extraction is performed on the target text to obtain multiple terms under each term extraction strategy, including:

[0010] Performing word segmentation processing on the target text according to the word segmentation processing strategy to obtain multiple terms under the word segmentation processing strategy;

[0011] According to the language model extraction strategy, the target text is input into the large language model for term extraction, and multiple terms of the language model extraction strategy are obtained;

[0012] According to the neural network extraction strategy, the target text is input into a pre-trained neural network model for term extraction, and multiple terms under the neural network extraction strategy are obtained.

[0013] In one embodiment, term sets of different levels are constructed according to multiple terms under each term extraction strategy, including:

[0014] Perform secondary word segmentation and de-duplication processing on multiple terms to obtain a sub-word term set;

[0015] De-duplicate multiple terms to obtain a full-word term set;

[0016] Clustering is performed on multiple terms to obtain a clustered term set.

[0017] In one embodiment, all terms under the term sets at different levels are classified to obtain classification results of all terms, including:

[0018] Construct a predictive term base based on all terms under different levels of term sets;

[0019] According to the predicted term base and the preset term base, all terms under the term sets at different levels are classified to obtain the classification results of all terms.

[0020] In one embodiment, all terms under different levels of term sets are classified according to the predicted term base and the preset term base to obtain classification results of all terms, including:

[0021] Input the prediction term base into the large language model, output the probability that each term in the prediction term base belongs to the target field, and score each term according to the probability to obtain the evaluation score corresponding to each term;

[0022] Traversing each term in the term sets at different levels, calculating the similarity between each term and all the preset terms in the preset term base, and determining the maximum similarity between each term and all the preset terms;

[0023] According to the evaluation score corresponding to each term and the maximum similarity between each term and all preset terms, all terms under the term sets at different levels are classified to obtain the classification results of all terms.

[0024] In one embodiment, all terms under different levels of term sets are classified according to the evaluation score corresponding to each term and the maximum similarity between each term and all preset terms, and the classification results of all terms are obtained, including:

[0025] For any term, the evaluation score and the maximum similarity are weighted to obtain the comprehensive similarity corresponding to each term;

[0026] According to the comprehensive similarity and category double thresholds, all terms under different levels of term sets are classified to obtain the classification results of all terms.

[0027] In a second aspect, the present application also provides a device for constructing a hierarchical term set, the device comprising:

[0028] An extraction module, used for performing term extraction on the target text according to at least three term extraction strategies to obtain a plurality of terms under each term extraction strategy;

[0029] A construction module is used to construct term sets of different levels according to multiple terms under each term extraction strategy; the term sets include sub-word term sets, full-word term sets and cluster term sets;

[0030] The classification module is used to classify all terms under different levels of term sets to obtain classification results of all terms; the classification results include chip screen patent terms, non-chip screen patent terms and uncertainties.

[0031] In a third aspect, the present application further provides a computer device, the computer device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the following steps when executing the computer program:

[0032] According to at least three term extraction strategies, term extraction is performed on the target text to obtain multiple terms under each term extraction strategy;

[0033] Construct term sets of different levels according to multiple terms under each term extraction strategy; the term sets include sub-word term sets, full-word term sets and cluster term sets;

[0034] All terms under different levels of term sets are classified to obtain classification results of all terms; the classification results include chip screen patent terms, non-chip screen patent terms and uncertainties.

[0035] In a fourth aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0036] According to at least three term extraction strategies, term extraction is performed on the target text to obtain multiple terms under each term extraction strategy;

[0037] Construct term sets of different levels according to multiple terms under each term extraction strategy; the term sets include sub-word term sets, full-word term sets and cluster term sets;

[0038] All terms under different levels of term sets are classified to obtain classification results of all terms; the classification results include chip screen patent terms, non-chip screen patent terms and uncertainties.

[0039] In a fifth aspect, the present application further provides a computer program product, the computer program product comprising a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0040] According to at least three term extraction strategies, term extraction is performed on the target text to obtain multiple terms under each term extraction strategy;

[0041] Construct term sets of different levels according to multiple terms under each term extraction strategy; the term sets include sub-word term sets, full-word term sets and cluster term sets;

[0042] All terms under different levels of term sets are classified to obtain classification results of all terms; the classification results include chip screen patent terms, non-chip screen patent terms and uncertainties.

[0043] The method, device, computer equipment, storage medium and computer program product for constructing the above-mentioned hierarchical term set, the method extracts terms from the target text according to at least three term extraction strategies, obtains multiple terms under each term extraction strategy, then constructs term sets of different levels based on the multiple terms under each term extraction strategy, and finally classifies all terms under the term sets of different levels to obtain the classification results of all terms. Among them, the classification results include core screen patent terms, non-core screen patent terms and uncertainties, and the term sets include sub-word term sets, full-word term sets and clustered term sets. In the above method, the use of multiple term extraction strategies can deeply analyze the content of the target text from multiple different dimensions, so as to extract terms more comprehensively, while increasing the coverage of terms, and can achieve the effect of mutual verification and complementation, thereby improving the accuracy of term extraction. Moreover, by constructing term sets of different levels, the terms can be refined and processed according to different levels, so that when processing the text later, the category and context level to which the term belongs can be more accurately identified, thereby greatly improving the accuracy of classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is an internal structure diagram of a computer device in one embodiment;

[0045] Figure 2A flowchart of a method for constructing a hierarchical term set in one embodiment;

[0046] Figure 3 is a flowchart of a method for constructing a hierarchical term set in another embodiment;

[0047] Figure 4 is a flowchart of a method for constructing a hierarchical term set in another embodiment;

[0048] Figure 5 is a flowchart of a method for constructing a hierarchical term set in another embodiment;

[0049] Figure 6 is a flowchart of a method for constructing a hierarchical term set in another embodiment;

[0050] Figure 7 is a flowchart of a method for constructing a hierarchical term set in another embodiment;

[0051] Figure 8 The structure block diagram of a device for constructing a hierarchical term set in one embodiment. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0053] With the rapid development of Internet technology, Natural Language Processing (NLP) has been widely used in text mining, large language models (LLM) and patent intelligent analysis due to its powerful semantic understanding and information extraction capabilities. In the field of patent intelligent analysis, due to the high knowledge density and terminology density of patents and the unstructured characteristics of text, it is particularly important to classify the terms in a large number of patent documents into target fields. At present, when classifying terms in patent documents, it is often based on a single rule or deep learning technology to extract and deduplicate terms to achieve term classification. However, the above information classification method has the problem of inaccuracy. The present application provides a method for constructing a hierarchical term set, aiming to solve the above technical problems. The following embodiments will specifically illustrate the method for constructing a hierarchical term set described in the present application.

[0054] The method for constructing a hierarchical term set provided in the embodiment of the present application can be applied to Figure 1 The computer device shown in the figure can be a terminal or a server, and its internal structure diagram can be as shown in Figure 1 As shown, the computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for constructing a hierarchical terminology set is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0055] Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0056] In one embodiment, Figure 2 As shown in the figure, a method for constructing a hierarchical term set is provided. Figure 1 The computer device in the example is used to illustrate, including the following steps:

[0057] S201, extracting terms from a target text according to at least three term extraction strategies to obtain a plurality of terms under each term extraction strategy.

[0058] Among them, term extraction strategies include word segmentation processing strategy, language model extraction strategy, neural network extraction strategy, etc. The target text is the patent text in the field of chip screens. The term is the term in the patent text in the field of chip screens.

[0059] In an embodiment of the present application, a computer device can obtain a target text through a preset path. Optionally, a display interface can be set on the computer device, and then respond to the operation of the target text input by the user on the display interface, and then obtain the target text. Optionally, the computer device can respond to the text acquisition path operation triggered by the user on the display interface, and then obtain the target text according to the text acquisition path. A variety of different term extraction strategies can be stored in advance in a storage path in the computer device, and then after the computer device obtains the target text, it can call a variety of different term extraction strategies from the storage path, and then perform term extraction on the target text through a variety of term extraction strategies to obtain multiple terms under each term extraction strategy. For example, the target text is subjected to term extraction through the first term extraction strategy to obtain multiple terms under the first term extraction strategy, and the target text is subjected to term extraction through the second keyword extraction strategy to obtain multiple terms under the second term extraction strategy. In the above manner, multiple terms under all term extraction strategies can be obtained.

[0060] S202: construct term sets of different levels according to the multiple terms under each term extraction strategy.

[0061] The term set is the patent term set of patent core screen. Different levels of term sets include sub-word term sets, full-word term sets and cluster term sets.

[0062] In the embodiment of the present application, a plurality of different word library construction strategies can be stored in advance in a storage path in a computer device, and then after the computer device obtains a plurality of terms under different term extraction strategies based on the above steps, a plurality of different word library construction strategies can be retrieved from the storage path, and then different categories of term libraries can be constructed according to different word library construction strategies. Specifically, a first term library can be constructed by a first word library construction strategy, and a second term library can be constructed by a second word library construction strategy. In the above manner, different categories of term libraries can be constructed.

[0063] S203, classify all terms under the term sets at different levels to obtain classification results of all terms.

[0064] The classification results include chip-screen patent terms, non-chip-screen patent terms and uncertain terms. Among them, uncertain means that it neither belongs to the chip-screen patent terms nor to the chip-screen patent terms.

[0065] In the embodiment of the present application, after the computer device obtains the term sets of different levels based on the above steps, it can classify all the terms in each term set to obtain the classification results of all the terms in each term set. Specifically, the similarity between all the terms in the term set and all the terms in the existing chip screen field can be determined first, and then all the terms under the term sets of different levels can be classified according to the similarity to obtain the classification results of all the terms.

[0066] The method for constructing a hierarchical term set provided in the embodiment of the present application extracts terms from the target text according to at least three term extraction strategies to obtain multiple terms under each term extraction strategy, then constructs term sets of different levels based on the multiple terms under each term extraction strategy, and finally classifies all terms under the term sets of different levels to obtain classification results of all terms. Among them, the classification results include core screen patent terms, non-core screen patent terms and uncertainties, and the term sets include sub-word term sets, full-word term sets and clustered term sets. In the above method, the use of multiple term extraction strategies can deeply analyze the content of the target text from multiple different dimensions, so as to extract terms more comprehensively, while increasing the coverage of terms, and can achieve the effect of mutual verification and complementation, thereby improving the accuracy of term extraction. Moreover, by constructing term sets of different levels, the terms can be refinedly classified and processed according to different levels, so that when processing the text later, the category and context level to which the term belongs can be more accurately identified, thereby greatly improving the accuracy of classification.

[0067] In one embodiment, the at least three term extraction strategies include a word segmentation processing strategy, a language model extraction strategy, and a neural network extraction strategy. Based on the at least three term extraction strategies, a specific implementation method for extracting terminology from a target text is provided, such as Figure 3 As shown, the above step S201 of "performing term extraction on the target text according to at least three term extraction strategies to obtain multiple terms under each term extraction strategy" includes:

[0068] S301, performing word segmentation processing on the target text according to the word segmentation processing strategy to obtain multiple terms under the word segmentation processing strategy.

[0069] In an embodiment of the present application, after the computer device obtains the target text, it can first analyze the term formation rules or characteristics in the target text, specifically, it can analyze the common prefixes and suffixes of the terms, the part-of-speech distribution of the terms, etc., and then decompose the text into vocabulary units or phrases according to the term formation rules or characteristics to obtain multiple terms under the word segmentation processing strategy. For example, on the basis of traditional word segmentation technology, the results of term feature analysis can be introduced, and the word segmentation algorithm can be optimized by adjusting the word segmentation threshold or introducing a term-specific dictionary to improve the recall rate of the term. And after obtaining a preliminary set of terms, stop word filtering is performed to remove words that have no practical meaning in a specific context, so as to obtain a more accurate set of terms.

[0070] S302: Input the target text into a large language model to extract terms according to the language model extraction strategy, and obtain a plurality of terms of the language model extraction strategy.

[0071] Among them, the language model is a large language model (Large Language Model, referred to as LLM), such as GPT-4.

[0072] In an embodiment of the present application, after the computer device obtains the target text, it can first input the target text into a large language model, conduct question and answer by designing appropriate prompts, implement term extraction, and obtain multiple terms of the language model extraction strategy. For example, the prompt is set to "You are now a senior patent expert, extract the terms or terms in the target text below, separated by semicolons." Such prompts can guide the model to generate a term list that meets the requirements. Since the model may generate some false terms or irrelevant terms, the illusion elimination module can be further used to remove terms that are irrelevant to the target content or inconsistent with reality.

[0073] S303, inputting the target text into a pre-trained neural network model to extract terms according to the neural network extraction strategy, and obtaining multiple terms under the neural network extraction strategy.

[0074] Among them, the neural network model can adopt the Phase-Bert named entity recognition model.

[0075] In an embodiment of the present application, the computer can pre-train the initial neural network model based on a large amount of sample data to obtain a trained neural network model. Specifically, the computer device can first obtain a large amount of sample text, and then manually annotate the sample text, or use word segmentation technology and / or a large language model to extract term samples, and then use the term samples to annotate the sample text to obtain a large amount of sample data, which includes the sample text and the corresponding term annotation results, and finally use the large amount of sample data to train the initial neural network model to obtain a trained neural network model. After the computer device obtains the target text, it can input the target text into the trained neural network model to extract terms and obtain multiple terms under the neural network extraction strategy.

[0076] The method described in the embodiments of the present application can achieve efficient and accurate extraction of terms in the target text by combining the above-mentioned multiple term extraction strategies and making comprehensive use of word segmentation technology, large language model technology and deep learning technology.

[0077] In one embodiment, the above-mentioned different categories of term libraries include sub-word term libraries, full-word term libraries and cluster term libraries. On this basis, a specific implementation method for constructing different categories of term libraries is provided, such as Figure 4 As shown, the "constructing term sets of different levels according to multiple terms under each term extraction strategy" in the above step S202 includes:

[0078] S401, performing secondary word segmentation and deduplication processing on multiple terms to obtain a sub-word term set.

[0079] Among them, the subword term set is a term set with a finer segmentation granularity.

[0080] In an embodiment of the present application, after obtaining multiple terms under different term extraction strategies, the computer device can perform secondary word segmentation and deduplication processing on the multiple terms to obtain a sub-term set. For example, the content of the sub-term set is: two-phase, static, coordinate system, circuit board, design, field, voltage, signal, transmission, structure, afterimage, elimination, device, shock wave, technology, inversion, light, doping, epitaxy, layer, base region, resistance, scale, invariant, feature, transformation, sift, data, mapping, relationship, information, slit, nozzle, resistance, material, voltage, control, signal.

[0081] S402, performing deduplication processing on multiple terms to obtain a full-word term set.

[0082] Among them, the full-word term set is a more complete set of terms.

[0083] In the embodiment of the present application, after obtaining multiple terms under different term extraction strategies, the computer device can perform deduplication processing on the multiple terms to obtain a full-word term set. For example, the content of the full-word term set is: two-phase stationary coordinate system, circuit board design field, voltage signal transmission structure, afterimage elimination device, shock wave technology, inverted lightly doped epitaxial layer, base resistance scale invariant feature transformation SIFT, data mapping relationship information, slit nozzle, resistance material, voltage control signal, positive phase clock signal, super lens array, measurement device, deposition process, heavily doped source region, conductive region, safety factor, multi-threaded port scanning program, self-reset function, acoustic short circuit.

[0084] S403, clustering the multiple terms to obtain a clustered term set.

[0085] Among them, the clustered term set is that the similarity of terms corresponding to different categories is high.

[0086] In an embodiment of the present application, after obtaining multiple terms under different term extraction strategies, the computer device can use a clustering algorithm to cluster the multiple terms to obtain a clustered term set. Specifically, the kmeans algorithm can be used to cluster terms with high word similarity into different clusters, and clean up the noise points to obtain a clustered term set. For example, words such as "etching process, etching process technology, etching process, etching technology, dry etching, vapor phase etching" are classified into one category and represented by "etching technology". Optionally, deduplication can be performed before clustering multiple terms.

[0087] The method described in the embodiment of the present application can identify more fine-grained sub-words by performing secondary segmentation on the terms, can better understand the detailed information in the target text, can remove irrelevant information by deduplicating the terms, and can discover the potential relationship and similarity between the terms by clustering the terms, which can further improve the classification accuracy.

[0088] In one embodiment, a specific implementation method for classifying all terms under different levels of term sets is also provided, such as Figure 5 As shown, the "classifying all terms under the term sets at different levels to obtain classification results of all terms" in the above step S203 includes:

[0089] S501, constructing a prediction term base according to all terms under term sets at different levels.

[0090] The prediction term base includes all or part of the terms in term sets at different levels.

[0091] In the embodiment of the present application, after the computer device obtains the term sets at different levels, it can build a predictive term library according to the term sets at different levels. Specifically, the predictive term library can be built according to all the terms in all the term sets. In this case, the predictive term library includes all sub-word terms in the sub-word term set, all full-word terms in the full-word term set, and all cluster terms in the cluster term set. Optionally, the term sets at different levels can be batched to build multiple predictive term libraries according to actual needs or the information processing capabilities of the computer device.

[0092] S502, classifying all terms under term sets at different levels according to the predicted term base and the preset term base, and obtaining classification results of all terms.

[0093] Among them, the preset terminology library is the existing patent chip screen patent terminology library.

[0094] In an embodiment of the present application, the computer device can obtain the existing terminology library in the field of chip screens in advance by means of web crawling, and use it as the preset terminology library. After the computer device obtains the predicted terminology library based on the above steps, it can classify all terms under the term sets of different levels according to the predicted terminology library and the preset terminology library to obtain the classification results. Specifically, the sub-word terms and the preset terminology library in the predicted terminology library can be classified first, and then the full-word terms and the preset terminology library in the predicted terminology library can be classified, and finally the clustered terms and the preset terminology library in the predicted terminology library can be classified to obtain the classification results.

[0095] Optional, such as Figure 6 As shown, the above step S502 of “classifying all terms under the term sets at different levels according to the predicted term base and the preset term base to obtain classification results of all terms” includes:

[0096] S601, input the prediction term library into the large language model, output the probability that each term in the prediction term library belongs to the target field, and score each term according to the probability to obtain the evaluation score corresponding to each term.

[0097] Among them, the target area is the chip and screen field.

[0098] In the embodiment of the present application, after the computer device obtains the prediction term library based on the above steps, the prediction term library can be input into the large language model, and then the question and answer can be performed by designing appropriate prompts. For example, the prompt is "output the probability that all the words in the above text belong to the core screen field term", and the probability of each term in the prediction term library belonging to the target field is obtained, and each term is scored according to the probability to obtain the evaluation score corresponding to each term. For example, score each term in the preset term library according to the probability that each term belongs to the target field, evaluate the probability that each term belongs to the core screen field term, and output it in json format: <context> {{ }}< / context> , analyze and get the score Optionally, to speed up the above process, multiple terms can be taken at once to construct prompts for question and answer.

[0099] S602, traversing each term in the term sets at different levels, calculating the similarity between each term and all the preset terms in the preset term library, and determining the maximum similarity between each term and all the preset terms.

[0100] In the embodiment of the present application, after obtaining the term sets at different levels, the computer device can traverse each term in the term sets at different levels, calculate the similarity between each term in the term sets at different levels and all the preset terms in the preset term library, and determine the maximum similarity between each term and all the preset terms from the similarities between each term in the term sets at different levels and all the preset terms in the preset term library. For example, to traverse the term bases of different categories Each word in , and the preset term base Calculate the cosine similarity for each preset term in and take the maximum value as .

[0101] S603, classifying all terms under the term sets at different levels according to the evaluation score corresponding to each term and the maximum similarity between each term and all preset terms, to obtain classification results for all terms.

[0102] In an embodiment of the present application, after the computer device obtains the evaluation score and the maximum similarity corresponding to each term based on the above steps, the evaluation score and the maximum similarity can be summed, and the sum value can be compared with a preset value, and then all terms under different levels of term sets can be classified based on the comparison result to obtain the classification results of all terms.

[0103] Optional, such as Figure 7 As shown, the above step S603 of “classifying all terms under the term sets at different levels according to the evaluation score corresponding to each term and the maximum similarity between each term and all preset terms to obtain the classification results of all terms” includes:

[0104] S701, for any term, weighted calculation is performed on the evaluation score and the maximum similarity to obtain the comprehensive similarity corresponding to each term.

[0105] In the embodiment of the present application, for any term, after obtaining the evaluation score and the maximum similarity, the computer device can perform weighted calculation on the evaluation score and the maximum similarity to obtain the comprehensive similarity corresponding to each term. For example, by the formula Calculate the comprehensive similarity corresponding to each term, where , For the fusion weights, they can be set to 0.35 and 0.65 respectively.

[0106] S702, classifying all terms under the term sets at different levels according to the comprehensive similarity and category double thresholds, and obtaining classification results of all terms.

[0107] Among them, the category double threshold includes the target first threshold and target second threshold .

[0108] In the embodiment of the present application, the category dual threshold can be determined according to the Dynamic Bivariate Thresholding (DBVT) method, and the constraint condition is , the specific method for determining the category double threshold is:

[0109] (1) Determine the initial first threshold according to the first quartile method :The formula is ,in, Representation term The probability distribution function of Represents the first quartile.

[0110] (2) Determine the initial second threshold using the method of maximizing the inter-class variance : In the interval Traverse , calculate the inter-class variance caused by each pair of thresholds, and select the threshold pair that maximizes the inter-class variance. The formula is ,in and They represent the between-class variance under the initial first threshold and the initial second threshold respectively.

[0111] (3) The initial first threshold and the initial second threshold are jointly optimized according to the joint optimization function, and the formula is: ,in, It is an optimization criterion function that jointly considers two thresholds and can be defined as the weighted sum of the variance between classes: . and is the weight coefficient used to balance the importance of the two thresholds.

[0112] (4) Under the constraints Next, traverse the search parameter space and find The maximized threshold pair is the optimal threshold pair. , which is the first target threshold and target second threshold .

[0113] In the embodiment of the present application, the computer device can determine the category double threshold according to the DBVT method, and then after obtaining the comprehensive similarity, it can classify all terms under the term sets of different levels according to the comprehensive similarity and the category double threshold to obtain the classification result. Specifically, if the comprehensive similarity is between 0 and the target first threshold If the comprehensive similarity is between the target second threshold, the classification result is that the term belongs to a non-chip screen patent term; if the comprehensive similarity is between the target second threshold to 1, the classification result indicates that the term belongs to the chip screen patent term; if the comprehensive similarity is within the target first threshold To the second target threshold , the classification result indicates that the category of the term is uncertain. That is, if , the classification result indicates that the term belongs to non-chip screen patent term; if 1, the classification result indicates that the term belongs to the chip-screen patent term; if , the classification result indicates that the category of the term is uncertain.

[0114] Based on all the above embodiments, a method for constructing a hierarchical term set is also provided, and the method includes:

[0115] S801, performing word segmentation processing on the target text according to the word segmentation processing strategy to obtain multiple terms under the word segmentation processing strategy.

[0116] S802: Input the target text into a large language model to extract terms according to the language model extraction strategy, and obtain a plurality of terms of the language model extraction strategy.

[0117] S803, inputting the target text into a pre-trained neural network model to extract terms according to the neural network extraction strategy, and obtaining multiple terms under the neural network extraction strategy.

[0118] S804, performing secondary word segmentation and deduplication processing on the multiple terms to obtain a sub-word term set.

[0119] S805, performing deduplication processing on multiple terms to obtain a full-word term set.

[0120] S806: Clustering the multiple terms to obtain a clustered term set.

[0121] S807, constructing a prediction term base according to all terms under the term sets at different levels.

[0122] S808, input the prediction term library into the large language model, output the probability that each term in the prediction term library belongs to the target field, and score each term according to the probability to obtain the evaluation score corresponding to each term.

[0123] S809, traversing each term in the term sets at different levels, calculating the similarity between each term and all the preset terms in the preset term library, and determining the maximum similarity between each term and all the preset terms.

[0124] S810: For any term, weighted calculation is performed on the evaluation score and the maximum similarity between each term and all preset terms to obtain a comprehensive similarity corresponding to each term.

[0125] S811, classify all terms under the term sets at different levels according to the comprehensive similarity and category double thresholds to obtain the classification results of all terms.

[0126] The method described in the embodiments of this application reduces the time of subsequent intelligent analysis tasks, improves the efficiency of the entire analysis process, and supports subsequent intelligent analysis tasks such as clustering, technical theme identification, and patent classification from a unified perspective through automated term recognition and hierarchical term set construction. Using large-scale language model technology and deep learning and other technologies, a three-stage method is used to create a hierarchical term set construction system for chip screen patents. The system aims to improve the hierarchical and structured representation of patent features and promote the effectiveness of intelligent analysis in field tasks such as clustering, classification, and technical theme mining.

[0127] The methods described in the above steps are all described in the above embodiments. Please refer to the above description for details and will not be repeated here.

[0128] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0129] Based on the same inventive concept, the embodiment of the present application also provides a hierarchical term set construction device for implementing the hierarchical term set construction method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiments of one or more hierarchical term set construction devices provided below can refer to the limitations of the hierarchical term set construction method above, and will not be repeated here.

[0130] In one embodiment, Figure 8 As shown, a device for constructing a hierarchical term set is provided, comprising:

[0131] The extraction module 11 is used to extract terms from the target text according to at least three term extraction strategies to obtain multiple terms under each term extraction strategy.

[0132] The construction module 12 is used to construct term sets of different levels according to multiple terms under each term extraction strategy; the term sets of different levels include sub-word term sets, full-word term sets and cluster term sets.

[0133] The classification module 13 is used to classify all terms under the term sets at different levels to obtain the classification results of all terms; the classification results include chip screen patent terms, non-chip screen patent terms and uncertainties.

[0134] In one embodiment, the extraction module 11 includes:

[0135] The first extraction unit is used to perform word segmentation processing on the target text according to the word segmentation processing strategy to obtain multiple terms under the word segmentation processing strategy.

[0136] The second extraction unit is used to input the target text into the large language model to extract terms according to the language model extraction strategy, and obtain multiple terms of the language model extraction strategy.

[0137] The third extraction unit is used to input the target text into a pre-trained neural network model to extract terms according to the neural network extraction strategy, so as to obtain multiple terms under the neural network extraction strategy.

[0138] In one embodiment, the building block 12 includes:

[0139] The first processing unit is used to perform secondary word segmentation and de-duplication processing on the multiple terms to obtain a sub-word term set.

[0140] The second processing unit is used to perform deduplication processing on the multiple terms to obtain a full-word term set.

[0141] The third processing unit is used to perform clustering processing on the multiple terms to obtain a cluster term set.

[0142] In one embodiment, the classification module 13 includes:

[0143] An extraction unit is constructed to construct a prediction term base according to all terms under term sets at different levels.

[0144] The classification unit is used to classify all terms under different levels of term sets according to the predicted term base and the preset term base to obtain the classification results of all terms.

[0145] In one embodiment, the above classification unit includes:

[0146] The output subunit is used to input the prediction term library into the large language model, output the probability that each term in the prediction term library belongs to the target field, and score each term according to the probability to obtain the evaluation score corresponding to each term.

[0147] The calculation subunit is used to traverse each term in the term sets of different levels, calculate the similarity between each term and all the preset terms in the preset term library, and determine the maximum similarity between each term and all the preset terms.

[0148] The classification subunit is used to classify all terms under the term sets of different levels according to the evaluation score corresponding to each term and the maximum similarity between each term and all preset terms, so as to obtain the classification results of all terms.

[0149] In one embodiment, the above-mentioned classification subunit is specifically used to perform weighted calculation on the evaluation score and the maximum similarity for any term to obtain the comprehensive similarity corresponding to each term; and classify all terms under the term sets at different levels according to the comprehensive similarity and category double thresholds to obtain the classification results of all terms.

[0150] Each module in the above-mentioned hierarchical terminology construction device can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0151] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0152] According to at least three term extraction strategies, term extraction is performed on the target text to obtain multiple terms under each term extraction strategy;

[0153] Constructing term sets of different levels according to multiple terms under each term extraction strategy; term sets of different levels include sub-word term sets, full-word term sets and cluster term sets;

[0154] All terms under different levels of term sets are classified to obtain classification results of all terms; the classification results include chip screen patent terms, non-chip screen patent terms and uncertainties.

[0155] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0156] Performing word segmentation processing on the target text according to the word segmentation processing strategy to obtain multiple terms under the word segmentation processing strategy;

[0157] According to the language model extraction strategy, the target text is input into the large language model for term extraction, and multiple terms of the language model extraction strategy are obtained;

[0158] According to the neural network extraction strategy, the target text is input into a pre-trained neural network model for term extraction, and multiple terms under the neural network extraction strategy are obtained.

[0159] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0160] Perform secondary word segmentation and de-duplication processing on multiple terms to obtain a sub-word term set;

[0161] De-duplicate multiple terms to obtain a full-word term set;

[0162] Clustering is performed on multiple terms to obtain a clustered term set.

[0163] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0164] Construct a predictive term base based on all terms under different levels of term sets;

[0165] According to the predicted term base and the preset term base, all terms under the term sets at different levels are classified to obtain the classification results of all terms.

[0166] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0167] Input the prediction term base into the large language model, output the probability that each term in the prediction term base belongs to the target field, and score each term according to the probability to obtain the evaluation score corresponding to each term;

[0168] Traversing each term in the term sets at different levels, calculating the similarity between each term and all the preset terms in the preset term base, and determining the maximum similarity between each term and all the preset terms;

[0169] According to the evaluation score corresponding to each term and the maximum similarity between each term and all preset terms, all terms under the term sets at different levels are classified to obtain the classification results of all terms.

[0170] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0171] For any term, the evaluation score and the maximum similarity are weighted to obtain the comprehensive similarity corresponding to each term;

[0172] According to the comprehensive similarity and category double thresholds, all terms under different levels of term sets are classified to obtain the classification results of all terms.

[0173] The above embodiment provides a computer device, whose implementation principle and technical effect are similar to those of the above method embodiment, and will not be repeated here.

[0174] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0175] According to at least three term extraction strategies, term extraction is performed on the target text to obtain multiple terms under each term extraction strategy;

[0176] Constructing term sets of different levels according to multiple terms under each term extraction strategy; term sets of different levels include sub-word term sets, full-word term sets and cluster term sets;

[0177] All terms under different levels of term sets are classified to obtain classification results of all terms; the classification results include chip screen patent terms, non-chip screen patent terms and uncertainties.

[0178] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0179] Performing word segmentation processing on the target text according to the word segmentation processing strategy to obtain multiple terms under the word segmentation processing strategy;

[0180] According to the language model extraction strategy, the target text is input into the large language model for term extraction, and multiple terms of the language model extraction strategy are obtained;

[0181] According to the neural network extraction strategy, the target text is input into a pre-trained neural network model for term extraction, and multiple terms under the neural network extraction strategy are obtained.

[0182] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0183] Perform secondary word segmentation and de-duplication processing on multiple terms to obtain a sub-word term set;

[0184] De-duplicate multiple terms to obtain a full-word term set;

[0185] Clustering is performed on multiple terms to obtain a clustered term set.

[0186] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0187] Construct a predictive term base based on all terms under different levels of term sets;

[0188] According to the predicted term base and the preset term base, all terms under the term sets at different levels are classified to obtain the classification results of all terms.

[0189] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0190] Input the prediction term base into the large language model, output the probability that each term in the prediction term base belongs to the target field, and score each term according to the probability to obtain the evaluation score corresponding to each term;

[0191] Traversing each term in the term sets at different levels, calculating the similarity between each term and all the preset terms in the preset term base, and determining the maximum similarity between each term and all the preset terms;

[0192] According to the evaluation score corresponding to each term and the maximum similarity between each term and all preset terms, all terms under the term sets at different levels are classified to obtain the classification results of all terms.

[0193] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0194] For any term, the evaluation score and the maximum similarity are weighted to obtain the comprehensive similarity corresponding to each term;

[0195] According to the comprehensive similarity and category double thresholds, all terms under different levels of term sets are classified to obtain the classification results of all terms.

[0196] The above embodiment provides a computer-readable storage medium, whose implementation principle and technical effect are similar to those of the above method embodiment, and will not be repeated here.

[0197] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0198] According to at least three term extraction strategies, term extraction is performed on the target text to obtain multiple terms under each term extraction strategy;

[0199] Constructing term sets of different levels according to multiple terms under each term extraction strategy; term sets of different levels include sub-word term sets, full-word term sets and cluster term sets;

[0200] All terms under different levels of term sets are classified to obtain classification results of all terms; the classification results include chip screen patent terms, non-chip screen patent terms and uncertainties.

[0201] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0202] Performing word segmentation processing on the target text according to the word segmentation processing strategy to obtain multiple terms under the word segmentation processing strategy;

[0203] According to the language model extraction strategy, the target text is input into the large language model for term extraction, and multiple terms of the language model extraction strategy are obtained;

[0204] According to the neural network extraction strategy, the target text is input into a pre-trained neural network model for term extraction, and multiple terms under the neural network extraction strategy are obtained.

[0205] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0206] Perform secondary word segmentation and de-duplication processing on multiple terms to obtain a sub-word term set;

[0207] De-duplicate multiple terms to obtain a full-word term set;

[0208] Clustering is performed on multiple terms to obtain a clustered term set.

[0209] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0210] Construct a predictive term base based on all terms under different levels of term sets;

[0211] According to the predicted term base and the preset term base, all terms under the term sets at different levels are classified to obtain the classification results of all terms.

[0212] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0213] Input the prediction term base into the large language model, output the probability that each term in the prediction term base belongs to the target field, and score each term according to the probability to obtain the evaluation score corresponding to each term;

[0214] Traversing each term in the term sets at different levels, calculating the similarity between each term and all the preset terms in the preset term base, and determining the maximum similarity between each term and all the preset terms;

[0215] According to the evaluation score corresponding to each term and the maximum similarity between each term and all preset terms, all terms under the term sets at different levels are classified to obtain the classification results of all terms.

[0216] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0217] For any term, the evaluation score and the maximum similarity are weighted to obtain the comprehensive similarity corresponding to each term;

[0218] According to the comprehensive similarity and category double thresholds, all terms under different levels of term sets are classified to obtain the classification results of all terms.

[0219] The above embodiment provides a computer program product, whose implementation principle and technical effect are similar to those of the above method embodiment, and will not be repeated here.

[0220] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0221] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0222] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for constructing a hierarchical term set, characterized in that: The method comprises: Extracting terms from the target text according to at least three term extraction strategies to obtain a plurality of terms under each of the term extraction strategies; Constructing term sets at different levels according to the multiple terms under each of the term extraction strategies; the term sets at different levels include sub-word term sets, full-word term sets and cluster term sets; All the terms under the term sets at different levels are classified to obtain classification results of all the terms; the classification results include chip-screen patent terms, non-chip-screen patent terms and uncertainties.

2. The method according to claim 1, characterized in that The at least three term extraction strategies include a word segmentation processing strategy, a language model extraction strategy, and a neural network extraction strategy. The term extraction is performed on the target text according to the at least three term extraction strategies to obtain a plurality of terms under each of the term extraction strategies, including: Performing word segmentation processing on the target text according to the word segmentation processing strategy to obtain a plurality of terms under the word segmentation processing strategy; Inputting the target text into a large language model to extract terms according to the language model extraction strategy, and obtaining a plurality of terms of the language model extraction strategy; According to the neural network extraction strategy, the target text is input into a pre-trained neural network model to extract terms, thereby obtaining a plurality of terms under the neural network extraction strategy.

3. The method according to claim 2, characterized in that The constructing of term sets at different levels according to the multiple terms under each term extraction strategy includes: Performing secondary word segmentation and deduplication processing on the multiple terms to obtain a sub-word term set; Deduplication processing is performed on the multiple terms to obtain a full-word term set; The multiple terms are clustered to obtain a clustered term set.

4. The method according to claim 1, characterized in that: The classifying of all the terms under the term sets at different levels to obtain the classification results of all the terms includes: Building a prediction term base according to all the terms under the term sets at different levels; According to the predicted term library and the preset term library, all terms under the term sets of different levels are classified to obtain classification results of all the terms.

5. The method according to claim 4, characterized in that The step of classifying all the terms under the term sets at different levels according to the predicted term base and the preset term base to obtain the classification results of all the terms includes: Inputting the predicted term library into a large language model, outputting the probability that each term in the predicted term library belongs to the target domain, and scoring each term according to the probability to obtain an evaluation score corresponding to each term; Traversing each term in the term sets of different levels, calculating the similarity between each term and all the preset terms in the preset term library, and determining the maximum similarity between each term and all the preset terms; According to the evaluation score corresponding to each term and the maximum similarity between each term and all the preset terms, all the terms under the term sets of different levels are classified to obtain the classification results of all the terms.

6. The method according to claim 5, characterized in that The step of classifying all the terms under the term sets at different levels according to the evaluation score corresponding to each term and the maximum similarity between each term and all the preset terms to obtain the classification results of all the terms includes: For any of the terms, weighted calculation is performed on the evaluation score and the maximum similarity between each of the terms and all the preset terms to obtain a comprehensive similarity corresponding to each of the terms; According to the comprehensive similarity and category double thresholds, all terms under the term sets at different levels are classified to obtain classification results of all terms.

7. A device for constructing a hierarchical term set, characterized in that: The device comprises: An extraction module, used to extract terms from a target text according to at least three term extraction strategies, and obtain a plurality of terms under each of the term extraction strategies; A construction module, used for constructing term sets of different levels according to the multiple terms under each of the term extraction strategies; the term sets include sub-word term sets, full-word term sets and cluster term sets; A classification module is used to classify all terms under the term sets of different levels to obtain classification results of all the terms; the classification results include chip screen patent terms, non-chip screen patent terms and uncertainties.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images