Technical competitor identification method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a technical competitor identification method and device, electronic equipment and a storage medium, and relates to the technical field of data mining. The method comprises the steps that a text set is obtained, the text set comprises a plurality of first texts, and the first texts are technical texts of a first object in a target technical field; obtaining a first semantic vector of each first text according to a pre-trained first model; calculating the similarity between each first semantic vector and each second semantic vector, and obtaining a target semantic vector from each second semantic vector according to the similarity; for each second object, obtaining the competitive strength of the second object and the first object according to the similarity corresponding to each target semantic vector corresponding to the second object; and determining the technical competitor of the first object from the second objects according to the competition intensity of the second objects, so that the identification of the technical competitor can cross the limitation of languages and countries under the condition of not depending on artificial participation.

Description

Technical Field

[0001] The present disclosure relates to the field of data mining technology. Specifically, the present application relates to a method, device, electronic device and storage medium for identifying a technology competitor. Background Art Identification of technology competitors refers to the identification of enterprises, universities, research institutions or individuals that are in technology competition in a specific technology field through systematic analysis and evaluation. These technology competitors are in direct or indirect competition with the competing entities in terms of technology innovation, patent application, market development, etc. Identification of technology competitors is of great significance to the strategic planning of enterprises, adjustment of technology research and development direction, and improvement of market competitiveness.

[0002] Existing technologies usually use artificial intelligence algorithms to identify technology competitors based on industry data (including patent data, product data, user review data, web page data, etc.). At present, whether it is based on the enterprise perspective, customer perspective, enterprise and customer mixed perspective, or Internet perspective, the competitor identification methods are all based on monolingual data, which is difficult to adapt to the current trend of technological globalization and cannot identify technology competitors across languages ​​and countries. Summary of the invention

[0003] The embodiments of the present disclosure provide a method, device, electronic device and storage medium for identifying a technology competitor, which are used to solve the technical problem that the technology competitor identification method in the prior art cannot identify technology competitors across languages ​​and countries.

[0004] According to one aspect of an embodiment of the present disclosure, a method for identifying a technology competitor is provided, comprising: Acquire a text set, the text set including a plurality of first texts, the first text being a technical text of a first object in the target technical field; Obtaining, according to a pre-trained first model, a first semantic vector of each first text, the first model being obtained by retraining a pre-trained multilingual first embedding model by a contrastive learning mechanism based on a total text set, wherein the total text set includes multilingual technical texts in the target technical field; Calculating the similarity between each first semantic vector and each second semantic vector, and obtaining a target semantic vector from each second semantic vector according to the similarity, where the second semantic vector is a semantic vector of a technical text of a second object in a target technical field; For each second object, according to the similarities corresponding to each target semantic vector corresponding to the second object, the competition intensity between the second object and the first object is obtained, where the competition intensity represents the degree of technical similarity between the first object and the second object; According to the competition intensity of the second objects, technical competitors of the first object are determined from the second objects.

[0005] According to another aspect of an embodiment of the present disclosure, a device for identifying a technology competitor is provided, comprising: A technical text acquisition module, used to acquire a text set, the text set includes a plurality of first texts, the first text being a technical text of a first object in the target technical field; A semantic vector acquisition module, configured to obtain a first semantic vector of each first text according to a pre-trained first model, the first model being obtained by retraining a pre-trained multilingual first embedding model according to a total text set using a contrastive learning mechanism, wherein the total text set includes multilingual technical texts in the target technical field; A similarity comparison module, used for calculating the similarity between each first semantic vector and each second semantic vector, and obtaining a target semantic vector from each second semantic vector according to the similarity, wherein the second semantic vector is a semantic vector of a technical text of a second object in a target technical field; A competition intensity acquisition module, for obtaining, for each second object, the competition intensity between the second object and the first object according to the similarity corresponding to each target semantic vector corresponding to the second object, wherein the competition intensity represents the degree of technical similarity between the first object and the second object; The competitor determination module determines the technical competitors of the first object from each second object according to the competition intensity of the second object.

[0006] According to another aspect of an embodiment of the present disclosure, an electronic device is provided. The electronic device includes a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps of the method provided in any of the above embodiments.

[0007] According to another aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method provided in any of the above embodiments are implemented.

[0008] According to one aspect of an embodiment of the present disclosure, a computer program product is provided, including a computer program, and the computer program is executed by a processor to perform the steps of the method provided in any of the above embodiments.

[0009] The technical solution provided by the embodiments of the present disclosure has the following beneficial effects: The technical solution provided by the embodiments of the present disclosure can construct a total text set based on a large number of multilingual technical texts in the target technical field, and retrain the multilingual first embedding model through a comparative learning mechanism to obtain a first model, which can fine-tune and optimize the first embedding model's ability to express multilingual text vectors in the target technical field. By using the trained first model, for a determined first object, each first text in the text set of the first object can be converted into a first semantic vector without relying on any human intervention, and a target semantic vector can be determined based on the similarity between the first semantic vector and each second semantic vector in the target technical field, and the second object can be determined by the target semantic vector and the competitive intensity of the second object can be calculated, and the technical competitors of the first object can be determined from each second object by the competitive intensity, so that the identification of technical competitors can transcend the restrictions of language and country, thereby greatly saving the human resource cost for judging technical competitors, and providing huge support for analysis scenarios such as industry analysis and technology layout analysis on a global scale. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings required for describing the embodiments of the present disclosure are briefly introduced below.

[0011] Figure 1 A flowchart of a method for identifying a technology competitor provided by an embodiment of the present disclosure; Figure 2 A flowchart of an unsupervised SimCSE training method provided in an embodiment of the present disclosure; Figure 3 A flowchart of a supervised SimCSE training method provided in an embodiment of the present disclosure; Figure 4 A flowchart of a method for identifying a technology competitor of a patentee provided in an embodiment of the present disclosure; Figure 5 A schematic diagram of the structure of a device for identifying a technology competitor provided by an embodiment of the present disclosure; Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0012] The embodiments of the present disclosure are described below in conjunction with the drawings in the present disclosure. It should be understood that the implementation methods described below in conjunction with the drawings are exemplary descriptions for explaining the technical solutions of the embodiments of the present disclosure and do not constitute a limitation on the technical solutions of the embodiments of the present disclosure.

[0013] It will be understood by those skilled in the art that, unless specifically stated, the singular forms "one", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the terms "including" and "comprising" used in the embodiments of the present disclosure refer to that the corresponding features can be implemented as the presented features, information, data, steps, operations, elements and / or components, but do not exclude the implementation as other features, information, data, steps, operations, elements, components and / or combinations thereof supported by the technical field. It should be understood that when we say that an element is "connected" or "coupled" to another element, the one element may be directly connected or coupled to the other element, or it may refer to that the one element and the other element establish a connection relationship through an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein indicates at least one of the items defined by the term, for example, "A and / or B" may be implemented as "A", or as "B", or as "A and B".

[0014] In order to make the objectives, technical solutions and advantages of the present disclosure more clear, the embodiments of the present disclosure will be further described in detail below with reference to the accompanying drawings.

[0015] The following describes several exemplary embodiments to illustrate the technical solutions of the embodiments of the present disclosure and the technical effects produced by the technical solutions of the present disclosure. It should be noted that the following embodiments can refer to, draw on or combine with each other, and the same terms, similar features and similar implementation steps in different embodiments will not be described repeatedly.

[0016] The following is an explanation of the terms and related technologies involved in this application: Existing technologies usually use artificial intelligence algorithms to identify technological competitors based on industry data (including patent data, product data, user review data, web page data, etc.).

[0017] The current related technologies mainly include: Related technology 1: Competitor identification method based on the enterprise perspective, which identifies competitors through the basic characteristics of the enterprise itself, such as enterprise technology, products and strategies. Competitor identification methods based on the enterprise perspective are mainly represented by identification standard method, patent analysis method, strategic group method and product information clustering method.

[0018] Related technology 2: Competitor identification method based on customer perspective, which takes customers as the main body and determines whether there is a competitive relationship between enterprises by judging the degree of customer overlap. Common methods include questionnaire surveys and brand conversion analysis. With the development of Internet information, the research on text mining methods of online comments has received more and more attention.

[0019] Related technology 3: The identification method of competitors from the mixed perspective of enterprises and customers usually combines structured and unstructured data such as enterprise product data and customer review data, which is more comprehensive and accurate than the identification from a single perspective. For example, based on the customer value leadership strategy, the enterprise financial statements and consumer reviews on e-commerce platforms are selected as information sources, and the competitor evaluation models based on financial characteristics and comprehensive characteristics are established respectively based on the back propagation (BP) neural network.

[0020] Related technology 4: Competitor identification method based on the Internet perspective. This type of method is an identification method that has emerged with the development of the Internet. Competitor identification methods based on the Internet perspective can be roughly divided into two methods: based on web page content and based on web page link structure.

[0021] Among them, based on the webpage content, the support vector machine (SVM) can be used to identify the competitive intention in news articles, which can effectively identify files with competitive intentions. Based on the webpage link structure, the target industry can be selected as the research object, and the Uniform Resource Locator (URL) co-occurrence analysis method can be used to display the competitive landscape of the target industry through a multi-dimensional scaling analysis (MDS) diagram, and the competitive level of competitors can be divided from the perspective of the relevant webpages of the target industry, and the main competitors can be analyzed and identified.

[0022] At present, competitor identification methods based on the enterprise perspective, customer perspective, enterprise and customer mixed perspective, and Internet perspective are all based on monolingual data. There are few studies on identifying technology competitors across languages ​​and countries based on multilingual data. It is difficult to adapt to the current trend of technological globalization and it is impossible to identify technology competitors across languages ​​and countries.

[0023] Based on this, the embodiments of the present disclosure provide a method for identifying technology competitors, which is used to solve the problem that the above related technologies cannot identify technology competitors across languages ​​and countries to a certain extent.

[0024] The following describes several exemplary embodiments to illustrate the technical solutions of the embodiments of the present disclosure and the technical effects produced by the technical solutions of the present disclosure. It should be noted that the following embodiments can refer to, draw on or combine with each other, and the same terms, similar features and similar implementation steps in different embodiments will not be described repeatedly.

[0025] It can be understood that in the method for identifying a technology competitor provided by the embodiment of the present disclosure, any method step can be executed by an electronic device and / or a server, and all steps in the method can be executed independently by the electronic device or the server, or jointly by the electronic device and the server.

[0026] The server may be an independent physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides cloud computing services. The electronic device may be a smart phone, a tablet computer, a laptop computer, a desktop computer, an intelligent voice interaction device (such as a smart speaker), a wearable electronic device (such as a smart watch), a vehicle terminal, a smart home appliance (such as a smart TV), an AR / VR device, etc., but is not limited thereto.

[0027] The following describes the embodiment of the present disclosure with the server as the execution subject, however, this does not constitute a limitation on the embodiment of the present disclosure. In addition to identifying technical competitors, the method provided in the embodiment of the present disclosure can also be applied to calculate and compare text similarities corresponding to various objects in other scenarios.

[0028] Figure 1 A flowchart of a method for identifying a technology competitor provided by an embodiment of the present disclosure is shown in FIG. Figure 1 As shown, the technical solution provided by the embodiment of the present disclosure includes the following steps: Step S101 , obtaining a text set, the text set including a plurality of first texts, the first text being a technical text of a first object in a target technical field.

[0029] Among them, the objects refer to enterprises, universities, research institutions, organizations or individuals.

[0030] Technical text refers to text with technical expression capabilities, such as paper texts, patent texts, fund project texts, book texts, research report texts, etc. In actual application, the type and quantity of technical texts can be determined according to actual needs.

[0031] In addition, considering the length of the technical text and the amount of information that each part can represent, the input (and training sample) of each model in the embodiments of the present disclosure can be a complete technical text or a preprocessed technical text (or preprocessed by the model), and the preprocessed technical text can be the text or text representation corresponding to one or more of the title, abstract, keywords, conclusion, etc. corresponding to the complete technical text.

[0032] Specifically, taking the identification of a technology competitor of any first object as an example, the method for identifying a technology competitor provided by the embodiment of the present disclosure is described in detail.

[0033] It can be understood that there is a corresponding relationship between the object and the technical field involved. For example, when the object is an enterprise, the technical field can be the technical field related to the industry involved in the enterprise. When the object is a university or research institution, the technical field can be the technical field of the subject studied by the university or research institution, etc.

[0034] When determining the first object, the target technical field corresponding to the first object may be determined simultaneously. The target technical field may refer to multiple technical fields involved in the first object, or may be set to a part or a single technical field involved in the first object according to actual needs.

[0035] For example, when determining that the first object is an enterprise with cross-domain industries, the first object involves multiple technical fields at the same time. If the goal is to identify the technical competitors of the first object in the field of communication technology, the communication field is determined as the target technical field.

[0036] After determining the first object (i.e., the target research object), in step S101, among the technical texts in the target field, the technical texts belonging to the first object are obtained as the first texts, and a text set is constructed by the first texts. The number of first texts in the first text set is determined according to the actual situation of the first object, and the embodiment of the present disclosure does not limit this.

[0037] It can be understood that the first text is the technical text of the first object in the target technical field. Considering the influence of the data volume of the technical text in the target field on the recognition, the technical text of the first object can be screened when determining the first text. For example, for patent texts, the texts of authorized patents within a certain period of time can be screened as the first text. For papers, the texts within a certain period of time and with qualified impact factors can be screened as the first text. The method of determining the first text can be adjusted according to actual needs.

[0038] Step S102, obtaining a first semantic vector of each first text according to a pre-trained first model, wherein the first model is obtained by retraining a pre-trained multilingual first embedding model according to a total text set using a contrastive learning mechanism, and the total text set includes multilingual technical texts in the target technical field.

[0039] Specifically, the first model is an embedding model that converts text into semantic vectors. Before applying the first model, it is necessary to pre-train the first model. The first model is obtained by retraining the pre-trained multilingual first embedding model based on the total text set using a contrastive learning mechanism. The goal of contrastive learning is to enable the first embedding model to distinguish between positive examples (similar samples) and negative examples (dissimilar samples), and learn more meaningful feature representations therefrom, thereby improving the first embedding model's ability to grasp the semantics of multilingual texts.

[0040] The multilingual first embedding model is a pre-trained model (large model) with multilingual text representation capabilities (i.e., converting text into semantic vectors). Multilingual refers to at least two languages. The specific types and number of languages ​​involved, as well as the specific type of the multilingual first embedding model, can be determined according to actual needs. For example, when the languages ​​are Chinese and English, the specific type of the first embedding model can be OpenAI, BGE-Large, BCE, GTE-Large, etc.

[0041] It can be understood that the total text set includes multilingual technical texts in the target technical field. When obtaining the total text set, the multilingual technical texts in the target technical field can also be screened to limit the size of the total text set. The specific screening method can be determined based on actual needs, and can be the same or similar to the method of screening the technical text of the first object.

[0042] In step S102, each first text is input into a pre-trained first model, and the first model converts the text into a semantic vector to obtain a first semantic vector of each first text output by the first model.

[0043] It is understandable that in the embodiments of the present disclosure, the semantic vector converted from the text may be a low-dimensional semantic vector or a high-dimensional semantic vector. Considering that the embodiments of the present disclosure are applied to large-scale and complex text data, converting each text into a high-dimensional semantic vector can better capture more semantic details, capture the subtle differences and complex semantics in the text, and improve the accuracy of similarity calculation between texts.

[0044] Step S103, calculating the similarity between each first semantic vector and each second semantic vector, and obtaining a target semantic vector from each second semantic vector according to the similarity, where the second semantic vector is a semantic vector of the technical text of the second object in the target technical field.

[0045] Specifically, the second semantic vector is a semantic vector of a technical text of a second object in the target technical field, the second object is different from the first object, and the second semantic vector is a semantic vector of a technical text of a non-first object in the target technical field.

[0046] After obtaining the first semantic vector, in step S103, the similarity between each first semantic vector and each second semantic vector is calculated, and the target semantic vector is obtained from each second semantic vector according to the similarity between each first semantic vector and the second semantic vector. The similarity between semantic vectors measures the proximity of two vectors in the semantic space, and the similarity metric can be determined according to actual needs, including but not limited to cosine similarity, Pearson Correlation Coefficient, etc.

[0047] It is understandable that, considering that the goal of calculating the semantic vector similarity in the embodiment of the present disclosure is to identify the technical competitors of the first object, that is, the target requirement is to identify the technical text with a high similarity to the first text of the first object.

[0048] Based on this, in the step of obtaining the target semantic vector corresponding to each second semantic vector according to the similarity, a preset screening rule can be set to determine the target semantic vector with higher similarity from each second semantic vector, and the screening rule can be determined according to actual needs.

[0049] For example, based on the multilingual technical texts in all target technical fields in the total text set, the semantic vector corresponding to each technical text is obtained according to the preset first model, and the semantic vector of the technical text that is not the first object is used as the second semantic vector. For each first semantic vector of the first object, the similarity between the first semantic vector and each second semantic vector is calculated.

[0050] When obtaining the target semantic vector, a filtering rule can be preset. For each first semantic vector, A second semantic vectors with the highest similarity to the first semantic vector are filtered out as the target semantic vector, or all similarities are sorted and B similarities with the highest similarities are filtered out, and the second semantic vectors corresponding to each filtered similarity are used as the target semantic vector.

[0051] It should be noted that the above example is only used as a specific example to assist in explaining the various method steps of the embodiment of the present disclosure, and is not intended to limit the embodiment of the present disclosure.

[0052] Step S104: for each second object, the competition intensity between the second object and the first object is obtained according to the similarity corresponding to each target semantic vector corresponding to the second object, where the competition intensity represents the degree of technical similarity between the first object and the second object.

[0053] Specifically, considering that there may be multiple technical texts belonging to the same second object that are similar to the first text of the first object, in order to determine whether the second object is a technical competitor of the first object, and to accurately measure the intensity of technical competition between the second object and the first object, the embodiment of the present disclosure sets competition intensity as a measurement indicator to characterize the similarity of the technology between the first object and the second object, and is used to measure the intensity of technical competition between the first object and the second object.

[0054] For each target semantic vector in step S103, there is a corresponding second object. The second objects corresponding to each target semantic vector may be the same or different. It is necessary to use the second object as a processing perspective and uniformly process the target semantic vectors belonging to the same second object.

[0055] It is understandable that there may be a situation where a certain technical text is shared by two or more objects. When processing, the technical text shared by multiple objects can be recorded as belonging to the object with the first signature order, or it can be recorded as shared by multiple objects. The processing method can be determined according to actual needs.

[0056] In step S104, for each determined second object, at least one target semantic vector corresponding to the second object is obtained, and the competition intensity between the second object and the first object is obtained according to the similarity corresponding to each target semantic vector corresponding to the second object.

[0057] It can be understood that the competition intensity represents the degree of similarity of the technologies between the first object and the second object, and the competition intensity calculation formula can be determined according to actual needs.

[0058] For example, for each second object, the competition intensity may be the average value of the similarities corresponding to the target semantic vectors corresponding to the second object. It may also be considered to add corresponding weights to the similarities of different target semantic vectors. The competition intensity is determined by weighted fusion of the similarities corresponding to the target semantic vectors corresponding to the second object. It is understandable that when determining the weight, the referenced information may include but is not limited to the time when the technical text corresponding to the target semantic vector was disclosed, the market share of the product related to the technical text, whether the technical text is shared by multiple objects, and the number of shared objects.

[0059] Step S105 , determining the technical competitors of the first object from each second object according to the competition intensity of the second object.

[0060] Specifically, the technical competitors of the first object can be understood as second objects that are highly similar to the first object from a technical perspective. Therefore, it is necessary to screen the second objects according to their competitive intensity and select the technical competitors of the first object from among the second objects.

[0061] In step S105, the second objects are screened according to the competition intensity of each second object determined in step S104, and the technical competitors of the first object are determined from the second objects.

[0062] It can be understood that the purpose of screening the second objects is to screen out second objects with higher competition intensity from various second objects as technical competitors. The specific screening method can be determined according to actual needs.

[0063] For example, all competition intensities may be ranked, and the C with the highest competition intensities or the top D% of the competition intensities may be screened out, and the second objects corresponding to the screened out competition intensities may be used as technical competitors.

[0064] The technical solution provided by the embodiments of the present disclosure can construct a total text set based on a large number of multilingual technical texts in the target technical field, and retrain the multilingual first embedding model through a comparative learning mechanism to obtain a first model, which can fine-tune and optimize the first embedding model's ability to express multilingual text vectors in the target technical field. By using the trained first model, for a determined first object, each first text in the text set of the first object can be converted into a first semantic vector without relying on any human intervention, and a target semantic vector can be determined based on the similarity between the first semantic vector and each second semantic vector in the target technical field, and the second object can be determined by the target semantic vector and the competitive intensity of the second object can be calculated, and the technical competitors of the first object can be determined from each second object by the competitive intensity, so that the identification of technical competitors can transcend the restrictions of language and country, thereby greatly saving the human resource cost for judging technical competitors, and providing huge support for analysis scenarios such as industry analysis and technology layout analysis on a global scale.

[0065] In the application scenario of multilingual text processing, the existing technology usually uses a multilingual machine translation (dictionary-based translation or statistics-based translation) engine to introduce multilingual technical texts into the machine translation engine and translate them into monolingual technical texts (for example, translating non-English technical texts such as English, German, Japanese, Korean, Russian, etc. into English technical texts), and converting the obtained monolingual technical texts into semantic vectors using a monolingual embedding model. This method relies on direct text conversion by the machine translation engine and may not be able to accurately capture complex semantic relationships and deep similarities between texts.

[0066] The embodiment of the present disclosure provides a first model obtained by retraining the first embedding model of multiple languages ​​through comparative learning, so that the first model can learn deeper and more fine-grained language features and semantic information, and better understand and represent text content between different languages. Compared with the above existing technologies, the method of directly converting multilingual technical texts into semantic vectors with the first model does not need to rely on machine translation engines with varying translation quality, avoids possible translation errors of machine translation engines, maintains the accuracy and professionalism of the original text during the conversion process, avoids terminology inconsistencies caused by translation between different languages, and simplifies the processing flow of obtaining semantic vectors from technical texts during application, which reduces the cost of translation and processing to a certain extent and improves processing efficiency.

[0067] In a possible implementation, the first model is generated in the following manner: Constructing a first sample set of multiple batches according to the total text set; According to the first sample sets of each batch, unsupervised training is performed on the first embedding model in batches, and the parameters of the first embedding model are fine-tuned with minimizing the contrast loss function as the optimization goal to obtain the first model; The first sample set of any batch is constructed in the following way: Determine a first technical text corresponding to the batch from the total text set, process the first technical text with a data enhancement technology, and obtain a plurality of second technical texts; Acquire several technical texts different from the first technical text from the total text collection as the third technical text; According to the first embedding model, obtaining semantic vectors corresponding to the first technical text, the second technical text, and the third technical text respectively; The semantic vector of the first technical text is used as an original sample, the semantic vector of the second technical text is used as a positive sample, and the semantic vector of the third technical text is used as a negative sample to obtain a first sample set.

[0068] Specifically, when training the first model, the contrast mechanism used is the unsupervised SimCSE (Simple Contrastive Learning of Sentence Embeddings) method, which uses a data enhancement strategy and batch training to optimize the model using a contrast loss function.

[0069] Specifically, before training, it is necessary to determine the training samples used in each batch of model training. The first sample set of any batch is constructed in the following way: For any first technical text in the total text set, the first technical text is processed using data enhancement technology, different sentence representations are generated based on the first technical text, and several second technical texts corresponding to the first text are obtained. The number of second technical texts can be determined based on the number of positive samples required for each batch.

[0070] A number of technical texts different from the first technical text are obtained from the total text set as the third technical text. The number of the third technical texts can be determined according to the number of negative samples required for each batch.

[0071] According to the first embedding model, each determined technical text is converted into a semantic vector, and the semantic vectors corresponding to the first technical text, the second technical text and the third technical text are obtained.

[0072] The semantic vector corresponding to the first technical text is obtained as an original sample, the semantic vector of the second technical text is used as a positive sample, and the semantic vector of the third technical text is used as a negative sample to obtain a first sample set.

[0073] In the above manner, for the total text set, multiple batches of first sample sets can be constructed, and the first embedding model can be unsupervisedly trained in batches according to the first sample sets of each batch.

[0074] In each batch training, the original samples of the first sample set and the positive samples form positive sample pairs, and the original samples and the negative samples form negative sample pairs. The optimization goal is to minimize the contrast loss function, backpropagate the contrast loss and fine-tune the parameters of the first embedding model (update the model parameters), and save the updated model until the fine-tuning is completed to obtain the first model.

[0075] It can be understood that the contrast loss function is a loss function used to measure the similarity between two input samples in the embedding space. Minimizing the contrast loss function is the optimization goal, which means that the first embedding model learns to maximize the similarity between positive sample pairs and minimize the similarity between negative sample pairs in the embedding space to learn more effective feature representation.

[0076] The contrast loss function used may include but is not limited to Information Noise-Contrastive Estimation loss (InfoNCE for short), the normalized temperature-scaled cross entropy loss (NT-Xent for short), and the like.

[0077] For example, the first model obtained after the BCE model is fine-tuned by unsupervised SimCSE is denoted as the BCE-UF-SimCSE model, where "U" stands for unsupervised and "F" stands for fine-tune.

[0078] Figure 2 A flowchart of an unsupervised SimCSE training method provided in an embodiment of the present disclosure is shown in FIG. Figure 2 As shown, for each training batch, the total text set is used as the basis for the technical text corresponding to the corresponding original sample, and the positive samples are obtained by the data enhancement strategy. The negative samples are obtained from the technical text corresponding to the non-original samples in the total text set to obtain the first sample set. The first embedding model is fine-tuned by unsupervised SimCSE using multiple first sample sets to obtain the BCE-UF-SimCSE model, wherein the method of obtaining the positive and negative samples is not described in detail here.

[0079] The contrast loss function formula is as follows:

[0080] In the formula, represents the original sample, represents a positive sample, represents negative samples, is the temperature coefficient, is the cosine similarity.

[0081] The technical solution provided by the embodiment of the present disclosure uses a data enhancement strategy to obtain a large number of positive samples without the need for complex sample annotation, which greatly reduces the cost of the training sample preparation stage. Unsupervised learning in the form of batch training can improve the generalization ability of the model, and the parameters of the first embedding model are fine-tuned using the contrast loss function to obtain the first model, which can enhance the first model's ability to represent texts in different languages, making similar texts in different languages ​​closer in the embedding space, which helps to achieve alignment between different languages ​​when processing cross-language texts and improve the performance of the first model.

[0082] In a possible implementation, the first model is generated in the following manner: Constructing multiple batches of second sample sets based on the total text set; According to the second sample sets of each batch, supervised training is performed on the first embedding model in batches, and the parameters of the first embedding model are fine-tuned with minimizing the contrast loss function as the optimization goal to obtain the first model; The second sample set of any batch is constructed in the following way: Determine from the total text set a fourth technical text in the first language that is preset and corresponds to the batch, and a fifth technical text that has a reference relationship with the fourth technical text; For the fourth technical text and each fifth technical text, obtaining a translation of the fourth technical text and a translation of the fifth text according to a preset second language; According to the first embedding model, obtaining semantic vectors corresponding to the fourth technical text, the fifth technical text, the translation of the fourth technical text, and the translation of the fifth technical text respectively; Calculate the similarity between the semantic vector of the fourth technical text and the semantic vectors of each sixth technical text in the target technical field, and determine the semantic vector of the target sixth technical text that meets a preset similarity threshold from each sixth technical text, the sixth technical text corresponds to the first language, and the semantic vector of the sixth technical text is obtained by processing the sixth technical text with the second model, and the second model is a monolingual embedding model that processes the target language; Calculate the similarity between the semantic vector of the translation of the fourth technical text and the semantic vectors of each seventh technical text in the target technical field, and determine the semantic vector of the target seventh technical text that meets a preset similarity threshold from each seventh technical text, where the seventh technical text corresponds to the second language, and the semantic vector of the seventh technical text is obtained by processing the seventh technical text with a third model, where the third model is a monolingual embedding model that processes the second language; According to the first embedding model, obtaining semantic vectors corresponding to the target sixth technical text and the target seventh technical text respectively; The semantic vector of the fourth technical text is taken as the original sample, the semantic vectors corresponding to the fourth technical text translation, the fifth technical text and the fifth technical text translation are taken as positive samples, and the semantic vectors corresponding to the target sixth technical text and the target seventh technical text are taken as negative samples to obtain the second sample set.

[0083] Specifically, in order to enable the first model to further improve the performance of processing technical texts and more accurately capture the language features and semantic relationships in the technical texts, the comparison mechanism adopted by the embodiment of the present disclosure when training the first model is a supervised SimCSE algorithm, which annotates positive samples based on the reference relationship between technical texts and annotates negative samples based on the similarity between technical texts.

[0084] Specifically, before training, it is necessary to determine the training samples used in each batch of model training. The embodiment of the present disclosure adopts the idea of ​​semantic alignment to determine the training samples. The second sample set of any batch is constructed in the following way: The first language and the second language are pre-set as the basis for semantic alignment. The first language and the second language are both language types that the first model targets to process. A second model that processes the first language and a third model that processes the second language are determined. Both the second model and the third model are monolingual embedding models.

[0085] Based on the language type corresponding to each technical text in the total text set, any fourth technical text is determined from each technical text in the first language in the total text set, and the fourth text is used as the technical text corresponding to the original samples of the batch.

[0086] For the fourth technical text, according to the reference relationship between technical texts, the fifth technical text having a reference relationship with the fourth technical text is determined. It can be understood that the reference relationship includes citing and being cited, which can be determined according to actual needs.

[0087] For the obtained fourth technical text and each fifth technical text, a translation of the fourth technical text and a translation of the fifth technical text are obtained according to the preset first language and the preset second language.

[0088] It can be understood that the fourth technical text is in the first language, and its corresponding translation is in the second language. The fifth technical texts may involve multiple types of languages. When obtaining the translation, if the fifth technical text is in the first language, the corresponding translation in the second language is obtained. If the fifth technical text is in the second language, the corresponding translation in the first language is obtained. If the fifth technical text is in a language other than the first and second languages, the corresponding translations in the first and second languages ​​of the fifth technical text are obtained at the same time. By obtaining the translation, multiple pairs of semantically identical aligned corpora corresponding to the first and second languages ​​can be obtained.

[0089] For example, when the language types processed by the first model target include Chinese, English and German, the first language and the second language are set to be Chinese and English respectively, the fourth technical text is a technical text in Chinese, the fifth technical text that has a reference relationship with the fourth technical text is obtained, the language types corresponding to each fifth technical text include Chinese, English and German, the Chinese translation corresponding to the fifth technical text in English is obtained, the English translation corresponding to the fifth technical text in Chinese is obtained, and the Chinese and English translations corresponding to the fifth technical text in German are obtained. It can be understood that according to the demand for the number of samples, the order of the first language and the second language can also be replaced, and for the case where the fourth technical text is a technical text in English, multiple pairs of semantically identical aligned corpora are generated in the same way.

[0090] According to the first embedding model, each determined technical text is converted into a semantic vector, and the semantic vectors corresponding to the fourth technical text, the fourth technical text translation, and the fifth technical text translation are obtained.

[0091] In the total text set, except for the fourth technical text, other technical texts corresponding to the first language are recorded as sixth technical texts, and based on the second model (i.e., the monolingual embedding model processing the first language), the semantic vector of the sixth technical text corresponding to each sixth technical text is obtained.

[0092] In the total text set, except for the fourth technical text, other technical texts corresponding to the second language are recorded as seventh technical texts, and based on the third model (i.e., the monolingual embedding model for processing the second language), the semantic vector of the seventh technical text corresponding to each seventh technical text is obtained.

[0093] The similarity between the semantic vector of the fourth technical text and the semantic vectors of each sixth technical text is calculated, a target sixth technical text satisfying a preset similarity threshold is determined from each sixth technical text, and the semantic vector of the target sixth technical text is determined.

[0094] The similarity between the semantic vector of the fourth technical text translation and the semantic vectors of each seventh technical text is calculated, a target seventh technical text satisfying a preset similarity threshold is determined from each seventh technical text, and the semantic vector of the target seventh technical text is determined.

[0095] It should be noted that when determining the target sixth semantic vector and the seventh semantic vector, in order to improve processing efficiency, a vector database corresponding to each technical text in the first language and the second language can be constructed based on the total text set. For example, the vector database in the first language may include the semantic vectors of the first language technical texts corresponding to all technical texts in the total text set (that is, for all technical texts in the total text set that are not in the first language, their corresponding translations in the first language are obtained, and the technical texts in the first language are not processed). The vector database in the second language can be obtained in the same way.

[0096] Based on the vector databases corresponding to the first language and the second language, the target sixth technical text corresponding to the fourth technical text and the target seventh technical text corresponding to the translation of the fourth technical text can be determined directly from the two vector databases according to the retrieval algorithm.

[0097] It can be understood that the similarity threshold is used to screen out technical texts with semantic differences from the fourth technical text from the sixth technical texts and the seventh technical texts based on the similarities, that is, the similarity threshold indicates the tail interval of the similarity value, and the similarity thresholds used when screening the sixth technical text or the seventh technical text can be the same or different. The specific value of the similarity threshold can be a direct similarity value or other values ​​that reflect the size of the similarity (such as a normal distribution value interval of the similarity, a sorting value interval), etc., which can be determined according to actual needs.

[0098] After obtaining the target sixth technical text and the target seventh technical text, the semantic vectors corresponding to the technical texts are obtained according to the first embedding model.

[0099] The acquired semantic vector of the fourth technical text is taken as the original sample, the semantic vectors corresponding to the fourth technical text translation, the fifth technical text and the fifth technical text translation are taken as positive samples, and the semantic vectors of the target sixth technical text and the target seventh technical text are taken as negative samples to obtain a second sample set.

[0100] It is understandable that in machine learning and natural language processing (NLP) tasks, negative samples refer to data samples that do not match the target or correct answer, and are usually used to train the model to learn to distinguish between correct and incorrect situations. In frameworks such as contrastive learning (such as SimCSE), negative samples can help the model better identify dissimilar samples to improve the performance of the model, which is particularly important.

[0101] For example, the first model obtained after the BCE model is fine-tuned by supervised SimCSE is denoted as the BCE-SF-SimCSE model, where "S" stands for supervised and "F" stands for fine-tuning.

[0102] Figure 3 A flow chart of a supervised SimCSE training method provided in an embodiment of the present disclosure is shown as follows: Figure 3 As shown, for each training batch, the total text set is used as the basis, and the technical text corresponding to the corresponding original sample is used as the basis. The positive sample is determined based on the reference relationship between the technical texts, and the negative sample is determined by combining the semantic vector generated by the monolingual embedding model and combining the similarity calculation and screening method to obtain the second sample set. The first embedding model is fine-tuned by supervised SimCSE using multiple second sample sets to obtain the BCE-SF-SimCSE model. The method of obtaining the positive and negative samples is not described in detail here.

[0103] The contrast loss function formula is as follows:

[0104] In the formula, represents the original sample, represents a positive sample, represents negative samples, is the temperature coefficient, is the cosine similarity.

[0105] The disclosed embodiment is based on the second model and the third model for processing a single language, obtains multiple groups of aligned corpora with the same semantics, constructs positive samples based on the reference relationship between technical texts, and enables the first embedding model to learn aligned semantics on the basis of semantic similarity of the positive samples. In combination with the screening of similarity between semantic vectors, the semantic vectors of technical texts with different semantics from the corresponding technical texts of the original samples are screened out from the technical texts in the first language and the second language as negative samples, which can make the acquired negative samples more accurate in monolingual expression, enable the multilingual embedding model to better capture language features and semantic relationships during training, and improve the performance of the model when processing technical texts.

[0106] In the above manner, for the total text set, multiple batches of second sample sets can be constructed, and the first embedding model can be supervised trained in batches according to the second sample sets of each batch.

[0107] During each batch training, the original samples of the second sample set and the positive samples form positive sample pairs, and the original samples and the negative samples form negative sample pairs. The optimization goal is to minimize the contrast loss function, backpropagate the contrast loss and fine-tune the parameters of the first embedding model (update the model parameters), and save the updated model until the fine-tuning is completed to obtain the first model.

[0108] The technical solution provided by the embodiments of the present disclosure adopts a supervised contrastive learning mechanism, takes into account the correlation between texts, and annotates positive samples based on the reference relationship between technical texts, so that the positive samples can help the model capture complex technical concepts. By combining the semantic vectors generated by the embedding model of a single language and combining similarity calculation and screening to determine the negative samples, it is possible to more accurately capture the language features and semantic relationships between different languages ​​while ensuring that the negative samples are semantically sufficiently different from the original samples, which helps the model learn more sophisticated differentiation capabilities, and can effectively improve the performance of the multi-first model in processing technical texts, meeting the needs of understanding complex technical content and cross-language semantic relationships.

[0109] In a possible implementation, the second model is generated in the following manner: Determine a sub-text set corresponding to the first language from the total text set in the target technical field; Retrain the pre-trained monolingual embedding model through the subtext set, and repeat the following first operation until the training stop condition is met to obtain a second model; The first operation includes: Performing a masking operation on the target segment in the eighth technical text in the sub-text set to obtain a masked text; Input the masked text into the monolingual embedding model to obtain the semantic vector of the masked text; By decoding the semantic vector of the mask text, an output result is obtained, and the output result is used to indicate the prediction result of the target segment; According to the loss between the target segment and the predicted result, the parameters of the monolingual embedding model of this iteration are updated to obtain the monolingual embedding model of the next iteration.

[0110] Specifically, before applying the second model, a dual structure of encoder and decoder can be introduced to perform secondary pre-training on the embedding model to improve the text representation capability. The architectures used include but are not limited to Masked Autoencoders and RetroMAE (retrieval-oriented pre-training paradigm based on MAE).

[0111] From the total text set in the target technical field, determine the technical texts corresponding to the target language, and construct a sub-text set. The sub-text set includes multiple technical texts in the first language, which are recorded as the eighth technical text. The eighth technical text can be obtained by screening or translating the technical texts in the total text set. The pre-trained monolingual embedding model is retrained through the sub-text set, and the following first operation is repeatedly performed: The pre-trained monolingual embedding model is retrained through the sub-text set. In each training process, the corresponding eighth technical text is determined. It can be understood that the number of eighth technical texts corresponding to one training can be determined according to actual needs.

[0112] For each eighth technical text, the eighth technical text in the sub-text set is encoded by the encoder, and the target segment in the eighth technical text is masked to obtain a masked text. The masked text is input into the monolingual embedding model of the corresponding processing target language text to obtain a semantic vector of the masked text.

[0113] Through the decoder, the semantic vector of the masked text is decoded, the masked target segment in the masked text is predicted, and an output result is obtained. The output result is used to indicate the prediction result of the target segment.

[0114] For the loss between the target segment and the prediction result corresponding to each eighth technical text, the total loss of this round of iterative training can be determined, and the parameters of the encoder and decoder are updated based on the total loss. Based on the updated parameters of the encoder and decoder, the parameters of the monolingual embedding model of this round of iteration are updated to obtain the monolingual embedding model of the next round of iteration.

[0115] Repeat the above first operation to iteratively train the monolingual embedding model until the training stop condition is met to obtain a second model. The training stop condition can be determined according to actual needs.

[0116] RetroMAE (Retrospective Masked Autoencoder) performs a second pre-training on the embedding model. The purpose of the second pre-training is to further improve the performance of the model on technical texts in the target language of the target domain, so that it can more accurately capture the language features and semantic relationships in technical texts.

[0117] It is understandable that the secondary pre-training method of the third model is the same as that of the second model, which will not be elaborated here.

[0118] The technical solution provided by the embodiment of the present disclosure introduces a dual structure of an encoder and a decoder to perform secondary pre-training on a monolingual embedding model, so as to further improve the performance of the second model when processing technical texts in the target language in the target technical field, so that it can more accurately capture the language features and semantic relationships in the technical texts in the target technical field. In this way, the quality of negative samples in supervised training of the first embedding model can be further improved.

[0119] In a possible implementation, calculating the similarity between each first semantic vector and each second semantic vector, and obtaining a target semantic vector from each second semantic vector according to the similarity, includes: For each first semantic vector, a search algorithm corresponding to the index structure of the pre-constructed vector database is used to determine a preset number of second semantic vectors with the greatest similarity from the second semantic vectors stored in the vector database as target semantic vectors; Among them, the vector database stores the semantic vector of each technical text in the total text set.

[0120] Specifically, after converting each technical text into a semantic vector, directly calculating the similarity matrix between the semantic vectors will consume huge storage space and computing resources, and the increase in vector dimension will lead to an exponential increase in computational complexity. The disclosed embodiment combines the index structure of the vector database to achieve efficient data management, and through the corresponding search algorithm, improves the efficiency of obtaining the target semantic vector from each second semantic vector based on the similarity between the semantic vectors, thereby shortening the retrieval time.

[0121] For the semantic vectors of each technical text in the total text set, each semantic vector is stored in a vector database, and an index of the vector database is constructed. The type of vector database can be determined according to actual needs, including but not limited to Faiss, Elasticsearch, and Milvus.

[0122] When determining the target semantic vector corresponding to the first semantic vector, for each first semantic vector, the semantic vector that comes from an object different from the first semantic vector is recorded as the second semantic vector. That is, the semantic vector belonging to the first object in the vector database is the first semantic vector, and the semantic vector belonging to the non-first object (each second object) is the second semantic vector.

[0123] Using the index structure, a search algorithm corresponding to the index structure of the pre-constructed vector database is executed to search for a preset number of second semantic vectors with the greatest similarity to the first semantic vector from the second semantic vectors stored in the vector database, and the second semantic vectors determined by the search are used as target semantic vectors.

[0124] It can be understood that the index structure and the corresponding search algorithm can be determined according to actual needs, wherein the types of index structures can include but are not limited to inverted file product quantization index (Inverted File ProductQuantization Index, referred to as IVF_PQ), inverted file flat index (Inverted File Flat Index, referred to as IVF_FLAT) and hierarchical navigation small world (Hierarchical Navigable Small World, referred to as HNSW), etc. The search algorithm can be determined according to the corresponding index structure, which is not limited in this embodiment.

[0125] It should be noted that in addition to the step of determining the target semantic vector, the vector database can also be applied in the related step of determining negative samples based on the similarity between semantic vectors. The specific application method is the same as the principle described above and will not be repeated here.

[0126] The technical solution provided by the embodiment of the present disclosure uses a vector database combined with an index structure and a search algorithm to perform similarity retrieval, which can efficiently find similar semantic vectors in a large-scale vector space, improve the efficiency of vector similarity calculation, shorten the retrieval time, and save computing resources.

[0127] In one possible implementation, the index of the vector database is constructed in the following way: According to the first model, the semantic vector of each technical text in the total text set is obtained; Construct the graph index of the vector database according to the hierarchical navigable small-world graph index structure; Among them, the graph index includes a multi-layer navigable small-world graph arranged from top to bottom and connected in a directed manner. Each layer of the graph includes multiple nodes, and each node in the bottom-level graph is used to represent the semantic vector of each technical text in the total text set; for any non-bottom-level graph, each node in the graph is a representative node of at least one node in the next-level graph; in each layer of the graph, the edge between any pair of nodes indicates the similarity between the two semantic vectors represented by the node pair.

[0128] Specifically, the embodiment of the present disclosure applies a hierarchical navigable small-world graph as an index structure. After obtaining the semantic vector of each technical text in the total text set according to the first model, a graph index of the vector database (i.e., a hierarchical navigable small-world graph index) is constructed according to the hierarchical navigable small-world graph index structure.

[0129] The graph index consists of multiple layers of navigable small-world graphs, each of which is arranged from top to bottom and has a directed connection. In each layer of the graph, there is at least one node, and the edge between any pair of nodes indicates the similarity between the two semantic vectors represented by the node pair.

[0130] In different graphs, the nodes reflect different information granularities. In the bottom-level graph, each node directly corresponds to the semantic vector of a corresponding technical text in the total text set. For any non-bottom-level graph, each node in the graph is a representative node of at least one node in the next-level graph, that is, a node in the previous-level graph can represent a node group composed of multiple nodes in the next-level graph that have similarities in the semantic space.

[0131] During the search phase, the sparse connection characteristics of the hierarchical navigation small-world graph are utilized to significantly reduce memory usage, making it particularly suitable for processing large-scale data sets. The hierarchical navigation small-world graph can effectively balance query speed and accuracy through a multi-level graph structure.

[0132] For example, after constructing an index structure corresponding to a vector database, the first semantic vector can be input into a hierarchical navigable small-world graph index through a graph-based approximate nearest neighbor search algorithm, and searched in each layer of the navigable small-world graph in order from top to bottom until multiple neighboring nodes of the first semantic vector are determined from multiple nodes included in the bottom-level navigable small-world graph, nodes corresponding to the semantic vector that belongs to the first object with the first semantic vector are removed from the multiple neighboring nodes, and a preset number of neighboring nodes with the greatest similarity are determined from each neighboring node based on the size of the similarity, and the corresponding target semantic vector is determined based on the acquired neighboring nodes.

[0133] The technical solution provided by the embodiment of the present disclosure uses a layered navigation small-world graph as an index structure. When performing a similarity search, the number of node connections in each layer of the graph is limited to a smaller range, which greatly reduces the search path length and computational complexity, significantly improves the search efficiency, and can efficiently find similar vectors in a large-scale vector space, significantly improving the retrieval speed and accuracy of similar texts.

[0134] In a possible implementation, for each second object, obtaining the competition intensity between the second object and the first object according to the similarities corresponding to each target semantic vector corresponding to the second object includes: For each second object, weighted fusion is performed on the similarities corresponding to each target semantic vector corresponding to the second object to obtain a total similarity; The total similarity of each second object is normalized, and the normalized total similarity is used as the competition intensity between the second object and the first object.

[0135] Specifically, for each determined second object, at least one target semantic vector corresponding to the second object is obtained, and the similarities corresponding to the target semantic vectors corresponding to the second object are weighted fused to obtain a total similarity, which is the total similarity between the second object and the first object, wherein the weights applied during weighted fusion can be set according to actual needs.

[0136] After obtaining the total similarity of each second object, the total similarity of each second object is normalized based on the maximum total similarity value, and the normalized total similarity is used as the competition intensity between the second object and the first object.

[0137] The technical solution provided by the embodiment of the present disclosure calculates the total similarity between the second object and the first object, and through normalization processing, converts the total similarity into a quantifiable value with consistent measurement standards as the competition intensity. This can more objectively evaluate the degree of proximity between different second objects and the first object at the technical level, making the competition intensity between different second objects comparable, so as to accurately screen out the most likely technical competitors.

[0138] Considering that patent data is a direct reflection of technological innovation and R&D activities, it contains a large amount of information about technological development trends, R&D results, and technological layout. The following is a specific application example, taking patent data as the source of technical texts, and multiple languages ​​such as Chinese and English as an example, to illustrate the method for identifying technology competitors provided by the embodiment of the present disclosure: The disclosed embodiments evaluate existing multilingual embedding models (including OpenAI, BGE-Large, BCE, and GTE-Large models), and construct bilingual and cross-lingual evaluation datasets covering multiple fields such as computer science, physics, biology, economics, mathematics, and quantitative finance. The evaluation uses two key indicators: hit rate and average reciprocal ranking. The hit rate reflects the ability of the model to find the correct answer in the first few searches, while the average reciprocal ranking reflects the average accuracy of the model in locating relevant documents in all queries. The larger the two indicators, the better.

[0139] Based on the evaluation results, the present embodiment selects the BCE model as the first embedding model by comprehensively considering the hit rate and the average reciprocal ranking. The BCE model fully utilizes the advantages of the Youdao translation engine, has excellent bilingual and cross-language capabilities, eliminates the differences between Chinese and English in semantic retrieval, and thus achieves powerful bilingual and cross-language semantic representation capabilities.

[0140] The disclosed embodiment is based on the Chinese and English patent text data of patents in the field of new materials, and selects to merge the title and abstract of each patent document to obtain the "title + abstract" text as the technical text to construct a total text set. It can be understood that in the total text set, for each patent document, the Chinese version and the English version of the technical text can be obtained based on the translation of the technical text, which effectively ensures the consistency and alignment of the data in different languages, so as to ensure the content consistency of the cited patents in a multilingual environment, and help to achieve accurate matching in subsequent cross-language similarity calculations.

[0141] Based on the total text set, for each technical text, the corresponding cited patent document is retrieved from the patent database according to the citation relationship of its corresponding patent document, and a positive sample is constructed based on the technical text corresponding to the cited patent document.

[0142] It is understandable that in actual operation, there may be missing data in the cited patent field. In addition, there may be differences in the integrity of Chinese and foreign patent data. In addition, the number of patents cited by each target patent document may be different, which leads to the fact that the number of cited patent texts generated by each target patent document is inconsistent. Therefore, in the actual data processing process, the impact of inconsistent numbers on model training can be considered to ensure the accuracy and reliability of the mining results.

[0143] For the annotation of negative samples, Chinese-BERT-WWM-EXT is selected as the Chinese embedding model, and BERT-for-patents is selected as the English embedding model. Each technical text is represented by a vector and stored in a vector database. A hierarchical navigation small-world graph is constructed as the index structure, and cosine similarity is used as the similarity metric. For each technical text, six negative samples are selected from the pre-set similarity threshold range according to the similarity between the semantic vectors.

[0144] Based on the negative sample and positive sample labeling method described above, a second sample set corresponding to each batch is constructed.

[0145] For example, when the original sample comes from Chinese technical text, the positive sample comes from the English translation text of the Chinese technical text + 3 Chinese-English technical text pairs of cited patents, and the negative sample is 3 Chinese technical texts mined by the chinese-bert-wwm-ext model based on RetroMAE secondary pre-training + 3 English technical texts mined by the bert-for-patents model based on RetroMAE secondary pre-training.

[0146] Correspondingly, when the original samples come from English technical texts, the positive samples come from the Chinese translation of the English technical texts + 3 Chinese-English technical text pairs of cited patents, and the negative samples are 3 Chinese technical texts mined by the chinese-bert-wwm-ext model based on RetroMAE secondary pre-training + 3 English technical texts mined by the bert-for-patents model based on RetroMAE secondary pre-training.

[0147] After obtaining the corresponding technical text, the technical text is processed by the BCE model to obtain the corresponding semantic vector, and the original text, negative samples and positive samples contained in each second sample set are obtained.

[0148] Based on the obtained multiple second sample sets, the BCE model is fine-tuned by using a supervised SimCSE fine-tuning method to obtain a BCE-SF-SimCSE model.

[0149] For example, Figure 4 A flowchart of a method for identifying a technology competitor of a patentee provided in an embodiment of the present disclosure is shown in FIG. Figure 4 As shown, the BCE-SF-SimCSE model is used to obtain the semantic vector corresponding to each technical text in the total text set, and after being indexed and stored in the vector database, patent similarity calculation is performed, and based on this, technical competitors are identified.

[0150] For patentee A (first object), a patent list M of patentee A is obtained, and the patent list M includes multiple patents m.

[0151] For each patent m, N similar patents similar to patent m are searched from the vector database to obtain a similar patent set Nm, a similarity set Sm and a patent owner (second object) set Pm. It can be understood that there is a mapping relationship between the data in the above three sets.

[0152] For each patentee in the patentee set Pm, the similarity data related thereto is determined from the similarity set Sm, and a weighted sum is performed to calculate the total similarity corresponding to each patentee.

[0153] The total similarity corresponding to each patent owner is normalized, and the competition intensity corresponding to each patent owner is calculated.

[0154] Output the patent owner set Pm corresponding to patent owner A, mark the corresponding competition intensity, and select the top 5 patent owners in terms of competition intensity as technical competitors.

[0155] Using patent data to identify technology competitors has the following advantages: Comprehensiveness and systematicity: Patent data covers technological innovation information worldwide, and patent data is systematic and standardized. By analyzing patent data, we can fully understand the competition situation in a specific technology field and comprehensively identify technological competitors.

[0156] Timeliness and dynamism: Patent data is constantly updated and can reflect the latest technological developments and competitive dynamics. By analyzing the latest patent data, it is possible to promptly identify emerging technological competitors, understand technological trends, and help companies maintain their leading position in technological competition.

[0157] Technical details and innovations: Patent data records the details and core innovations of technological innovations. By analyzing patent data, we can gain an in-depth understanding of competitors’ technical strength and innovation capabilities. This information is of great reference value for evaluating competitors’ technical levels and formulating technical research and development strategies.

[0158] Legal protection and competitive barriers: Patents have a legal protection function. By analyzing patent data, we can understand competitors’ protection strategies and competitive barriers in the technology field. This helps companies avoid infringement risks in technology research and development and market promotion, and build their own technology protection system through patent applications.

[0159] Technology cooperation and M&A opportunities: Patent data can also reveal potential technology partners or M&A targets. By analyzing competitors’ patent layout, opportunities for technology complementarity or cooperation can be discovered, thereby promoting technology cooperation and resource integration and enhancing the competitiveness of enterprises.

[0160] In summary, using patent data to identify technological competitors can not only provide a comprehensive understanding of competitors' technological layout and innovation capabilities, but also provide important basis for a company's technological research and development, market expansion and strategic planning.

[0161] The disclosed embodiment uses a massive multilingual patent data set as the basis of technical texts, and constructs a first model through the design of an artificial intelligence algorithm for converting technical texts into semantic vectors. This allows the identification of technology competitors to transcend language and country restrictions without relying on any human intervention, thereby greatly saving the human resource costs for judging technology competitors and providing tremendous support for analysis scenarios such as industry analysis and technology layout analysis on a global scale.

[0162] It should be understood that the embodiments of the present disclosure are only used as a specific example to explain the present invention in detail, and the described embodiments are only intended to facilitate the understanding of the present invention and do not have any limiting effect on the present invention.

[0163] Figure 5 A schematic diagram of a device for identifying a technology competitor provided by an embodiment of the present disclosure is shown in FIG. Figure 5 As shown, the device 500 for identifying a technology competitor comprises: A technical text acquisition module 501 is used to acquire a text set, where the text set includes a plurality of first texts, where the first text is a technical text of a first object in the target technical field; A semantic vector acquisition module 502 is used to obtain a first semantic vector of each first text according to a pre-trained first model, the first model being obtained by retraining a pre-trained multilingual first embedding model according to a total text set using a contrastive learning mechanism, wherein the total text set includes multilingual technical texts in the target technical field; A similarity comparison module 503 is used to calculate the similarity between each first semantic vector and each second semantic vector, and obtain a target semantic vector from each second semantic vector according to the similarity, where the second semantic vector is a semantic vector of a technical text of a second object in a target technical field; A competition intensity acquisition module 504 is used to obtain, for each second object, the competition intensity between the second object and the first object according to the similarity corresponding to each target semantic vector corresponding to the second object, where the competition intensity represents the degree of technical similarity between the first object and the second object; The competitor determination module 505 determines the technical competitors of the first object from each second object according to the competition intensity of the second object.

[0164] In a possible implementation, the device for identifying a technology competitor further includes: a model generation module; the model generation module is used to generate a first model; The first model was generated by: Constructing a first sample set of multiple batches according to the total text set; According to the first sample sets of each batch, unsupervised training is performed on the first embedding model in batches, and the parameters of the first embedding model are fine-tuned with minimizing the contrast loss function as the optimization goal to obtain the first model; The first sample set of any batch is constructed in the following way: Determine a first technical text corresponding to the batch from the total text set, process the first technical text with a data enhancement technology, and obtain a plurality of second technical texts; Acquire several technical texts different from the first technical text from the total text collection as the third technical text; According to the first embedding model, obtaining semantic vectors corresponding to the first technical text, the second technical text, and the third technical text respectively; The semantic vector of the first technical text is used as an original sample, the semantic vector of the second technical text is used as a positive sample, and the semantic vector of the third technical text is used as a negative sample to obtain a first sample set.

[0165] In a possible implementation, the device for identifying a technology competitor further includes: a model generation module; the model generation module is used to generate a first model; The first model was generated by: Determine from the total text set a fourth technical text in the first language that is preset and corresponds to the batch, and a fifth technical text that has a reference relationship with the fourth technical text; For the fourth technical text and each fifth technical text, obtaining a translation of the fourth technical text and a translation of the fifth text according to a preset second language; According to the first embedding model, obtaining semantic vectors corresponding to the fourth technical text, the fifth technical text, the translation of the fourth technical text, and the translation of the fifth technical text respectively; Calculate the similarity between the semantic vector of the fourth technical text and the semantic vectors of each sixth technical text in the target technical field, and determine the semantic vector of the target sixth technical text that meets a preset similarity threshold from each sixth technical text, the sixth technical text corresponds to the first language, and the semantic vector of the sixth technical text is obtained by processing the sixth technical text with the second model, and the second model is a monolingual embedding model that processes the target language; Calculate the similarity between the semantic vector of the translation of the fourth technical text and the semantic vectors of each seventh technical text in the target technical field, and determine the semantic vector of the target seventh technical text that meets a preset similarity threshold from each seventh technical text, where the seventh technical text corresponds to the second language, and the semantic vector of the seventh technical text is obtained by processing the seventh technical text with a third model, where the third model is a monolingual embedding model that processes the second language; According to the first embedding model, obtaining semantic vectors corresponding to the target sixth technical text and the target seventh technical text respectively; The semantic vector of the fourth technical text is taken as the original sample, the semantic vectors corresponding to the fourth technical text translation, the fifth technical text and the fifth technical text translation are taken as positive samples, and the semantic vectors corresponding to the target sixth technical text and the target seventh technical text are taken as negative samples to obtain the second sample set.

[0166] In a possible implementation, the model generation module is used to generate the second model; The second model was generated by: Determine a sub-text set corresponding to the first language from the total text set in the target technical field; Retrain the pre-trained monolingual embedding model through the subtext set, and repeat the following first operation until the training stop condition is met to obtain a second model; The first operation includes: Performing a masking operation on the target segment in the eighth technical text in the sub-text set to obtain a masked text; Input the masked text into the monolingual embedding model to obtain the semantic vector of the masked text; By decoding the semantic vector of the mask text, an output result is obtained, and the output result is used to indicate the prediction result of the target segment; According to the loss between the target segment and the predicted result, the parameters of the monolingual embedding model of this iteration are updated to obtain the monolingual embedding model of the next iteration.

[0167] In a possible implementation, the similarity comparison module is used to: For each first semantic vector, a search algorithm corresponding to the index structure of the pre-constructed vector database is used to determine a preset number of second semantic vectors with the greatest similarity from the second semantic vectors stored in the vector database as target semantic vectors; Among them, the vector database stores the semantic vector of each technical text in the total text set.

[0168] In a possible implementation, the device for identifying a technology competitor further includes: a database processing module; the database processing module is used to construct an index of the vector database; The index of the vector database is constructed in the following way: According to the first model, the semantic vector of each technical text in the total text set is obtained; Construct the graph index of the vector database according to the hierarchical navigable small-world graph index structure; Among them, the graph index includes a multi-layer navigable small-world graph arranged from top to bottom and connected in a directed manner. Each layer of the graph includes multiple nodes, and each node in the bottom-level graph is used to represent the semantic vector of each technical text in the total text set; for any non-bottom-level graph, each node in the graph is a representative node of at least one node in the next-level graph; in each layer of the graph, the edge between any pair of nodes indicates the similarity between the two semantic vectors represented by the node pair.

[0169] In a possible implementation, the competition intensity acquisition module is used to: For each second object, weighted fusion is performed on the similarities corresponding to each target semantic vector corresponding to the second object to obtain a total similarity; The total similarity of each second object is normalized, and the normalized total similarity is used as the competition intensity between the second object and the first object.

[0170] The device of the embodiments of the present disclosure can execute the method provided by the embodiments of the present disclosure, and the implementation principles are similar. The actions performed by each module in the device of each embodiment of the present disclosure correspond to the steps in the method of each embodiment of the present disclosure. For the detailed functional description of each module of the device, please refer to the description in the corresponding method shown in the previous text, which will not be repeated here.

[0171] In addition, in the embodiments of the present disclosure, the term "module" or "unit" refers to a computer program or a part of a computer program with a predetermined function, and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories), or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0172] An embodiment of the present disclosure provides an electronic device (computer device / equipment / system), including a memory, a processor, and a computer program stored in the memory, and the processor executes the above-mentioned computer program to implement the steps of the method provided in any optional embodiment of the present disclosure and achieve corresponding technical effects.

[0173] In an optional embodiment, an electronic device is provided. Figure 6 A schematic diagram of the structure of an electronic device provided by an embodiment of the present disclosure is shown in FIG. Figure 6As shown, the electronic device 600 includes: a processor 601 and a memory 603. The processor 601 and the memory 603 are connected, such as through a bus 602. Optionally, the electronic device 600 may also include a transceiver 604, which may be used for data interaction between the electronic device and other electronic devices, such as data transmission and / or data reception. It should be noted that in actual applications, the transceiver 604 is not limited to one, and the structure of the electronic device 600 does not constitute a limitation on the embodiments of the present disclosure.

[0174] Processor 601 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of the present invention. Processor 601 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0175] The bus 602 may include a path to transmit information between the above components. The bus 602 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus 602 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0176] The memory 603 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), magnetic disk storage medium, other magnetic storage devices, or any other medium that can be used to carry or store computer programs and can be read by a computer, without limitation herein.

[0177] The memory 603 is used to store the computer program for executing the embodiment of the present disclosure, and the execution is controlled by the processor 601. The processor 601 is used to execute the computer program stored in the memory 603 to implement the steps shown in the above method embodiment.

[0178] The electronic devices in the embodiments of the present disclosure may include, but are not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), wearable devices, etc., and fixed terminals such as digital TVs, desktop computers, etc.

[0179] An embodiment of the present disclosure provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps and corresponding contents of the aforementioned method embodiment can be implemented.

[0180] The embodiments of the present disclosure also provide a computer program product, including a computer program, which can implement the steps and corresponding contents of the aforementioned method embodiments when executed by a processor.

[0181] It should be noted that the computer-readable storage medium disclosed above may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0182] In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, apparatus, or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0183] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including, but not limited to, object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0184] The terms "first", "second", "third", "fourth", "1", "2", etc. (if any) in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than that shown or described in the text.

[0185] It should be understood that, although the flowchart of the embodiment of the present disclosure indicates each operation step by arrows, the implementation order of these steps is not limited to the order indicated by the arrows. Unless clearly stated herein, in some implementation scenarios of the embodiment of the present disclosure, the implementation steps in each flowchart can be executed in other orders according to demand. In addition, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages based on the actual implementation scenario. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage in these sub-steps or stages can also be executed at different times. In scenarios with different execution times, the execution order of these sub-steps or stages can be flexibly configured according to demand, and the embodiment of the present disclosure does not limit this.

[0186] The above is only an optional implementation method for some implementation scenarios of the present disclosure. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the scheme of the present disclosure, other similar implementation methods based on the technical ideas of the present disclosure are also within the protection scope of the embodiments of the present disclosure.

Claims

1. A method for identifying a technology competitor, characterized in that: include: Acquire a text set, the text set including a plurality of first texts, the first texts being technical texts of first objects in the target technical field; Obtaining a first semantic vector of each first text according to a pre-trained first model, wherein the first model is obtained by retraining a pre-trained multilingual first embedding model according to a total text set using a contrastive learning mechanism, wherein the total text set includes multilingual technical texts in the target technical field; Calculating the similarity between each first semantic vector and each second semantic vector, and obtaining a target semantic vector from each second semantic vector according to the similarity, wherein the second semantic vector is a semantic vector of a technical text of a second object in the target technical field; For each second object, according to the similarities corresponding to each target semantic vector corresponding to the second object, obtain the competition intensity between the second object and the first object, where the competition intensity represents the degree of technical similarity between the first object and the second object; According to the competition intensity of the second object, technical competitors of the first object are determined from each second object.

2. The method for identifying technology competitors according to claim 1, characterized in that: The first model is generated by: Constructing a plurality of batches of first sample sets according to the total text set; According to the first sample sets of each batch, unsupervised training is performed on the first embedding model in batches, and the parameters of the first embedding model are fine-tuned with minimizing the contrast loss function as an optimization goal to obtain the first model; The first sample set of any batch is constructed in the following way: Determine a first technical text corresponding to the batch from the total text set, process the first technical text with a data enhancement technology, and obtain a plurality of second technical texts; Acquire a number of technical texts different from the first technical text from the total text set as third technical texts; According to the first embedding model, obtaining semantic vectors corresponding to the first technical text, the second technical text, and the third technical text respectively; The semantic vector of the first technical text is used as an original sample, the semantic vector of the second technical text is used as a positive sample, and the semantic vector of the third technical text is used as a negative sample to obtain the first sample set.

3. The method for identifying technology competitors according to claim 1, characterized in that: The first model is generated by: Constructing a plurality of batches of second sample sets according to the total text set; According to the second sample sets of each batch, supervised training is performed on the first embedding model in batches, minimizing the contrast loss function as an optimization goal, and the parameters of the first embedding model are fine-tuned to obtain the first model; The second sample set of any batch is constructed in the following way: Determining from the total text set a fourth technical text in the first language that is preset and corresponds to the batch, and a fifth technical text that has a reference relationship with the fourth technical text; For the fourth technical text and each fifth technical text, obtaining a translation of the fourth technical text and a translation of the fifth text according to a preset second language; According to the first embedding model, obtaining semantic vectors corresponding to the fourth technical text, the fifth technical text, the translation of the fourth technical text, and the translation of the fifth technical text respectively; Calculating the similarity between the semantic vector of the fourth technical text and the semantic vectors of each sixth technical text in the target technical field, and determining the semantic vector of the target sixth technical text that meets a preset similarity threshold from each sixth technical text, wherein the sixth technical text corresponds to the first language, and the semantic vector of the sixth technical text is obtained by processing the sixth technical text with a second model, and the second model is a monolingual embedding model that processes the target language; Calculating the similarity between the semantic vector of the translation of the fourth technical text and the semantic vectors of each seventh technical text in the target technical field, and determining the semantic vector of the target seventh technical text that meets a preset similarity threshold from each seventh technical text, wherein the seventh technical text corresponds to the second language, and the semantic vector of the seventh technical text is obtained by processing the seventh technical text with a third model, and the third model is a monolingual embedding model that processes the second language; According to the first embedding model, obtaining semantic vectors corresponding to the target sixth technical text and the target seventh technical text respectively; The second sample set is obtained by taking the semantic vector of the fourth technical text as the original sample, the semantic vectors corresponding to the fourth technical text translation, the fifth technical text and the fifth technical text translation as positive samples, and the semantic vectors corresponding to the target sixth technical text and the target seventh technical text as negative samples.

4. The method for identifying technology competitors according to claim 3, characterized in that: The second model is generated by: Determining a sub-text set corresponding to the first language from the total text set in the target technical field; Retraining the pre-trained monolingual embedding model through the sub-text set, and repeatedly performing the following first operation until a training stop condition is met, thereby obtaining the second model; The first operation includes: Performing a masking operation on the target segment in the eighth technical text in the sub-text set to obtain a masked text; Inputting the masked text into the monolingual embedding model to obtain a semantic vector of the masked text; Obtaining an output result by decoding the semantic vector of the mask text, wherein the output result is used to indicate a prediction result of the target segment; According to the loss between the target segment and the prediction result, the parameters of the monolingual embedding model of the current iteration are updated to obtain the monolingual embedding model of the next iteration.

5. The method for identifying a technology competitor according to any one of claims 1 to 4, characterized in that: The calculating the similarity between each first semantic vector and each second semantic vector, and obtaining the target semantic vector from each second semantic vector according to the similarity, includes: For each first semantic vector, using a search algorithm corresponding to the index structure of the pre-constructed vector database, a preset number of second semantic vectors with the greatest similarity are determined as target semantic vectors from the second semantic vectors stored in the vector database; The vector database stores the semantic vector of each technical text in the total text set.

6. The method for identifying technology competitors according to claim 5, characterized in that: The index of the vector database is constructed in the following way: According to the first model, obtaining the semantic vector of each technical text in the total text set; Constructing a graph index of the vector database according to a hierarchical navigable small-world graph index structure; The graph index includes a multi-layer navigable small-world graph arranged from top to bottom and connected in a directed manner, each layer of the graph includes multiple nodes, and each node in the bottom layer graph is used to represent the semantic vector of each technical text in the total text set; for any non-bottom layer graph, each node in the graph is a representative node of at least one node in the next layer graph; in each layer of the graph, the edge between any pair of nodes indicates the similarity between the two semantic vectors represented by the node pair.

7. The method for identifying a technology competitor according to any one of claims 1 to 4, characterized in that: For each second object, obtaining the competition intensity between the second object and the first object according to the similarities corresponding to each target semantic vector corresponding to the second object, including: For each second object, weighted fusion is performed on the similarities corresponding to each target semantic vector corresponding to the second object to obtain a total similarity; The total similarity of each second object is normalized, and the normalized total similarity is used as the competition intensity between the second object and the first object.

8. A device for identifying a technology competitor, characterized in that: include: A technical text acquisition module, used to acquire a text set, wherein the text set includes a plurality of first texts, wherein the first texts are technical texts of first objects in the target technical field; A semantic vector acquisition module, configured to obtain a first semantic vector of each first text according to a pre-trained first model, wherein the first model is obtained by retraining a pre-trained multilingual first embedding model according to a total text set using a contrastive learning mechanism, wherein the total text set includes multilingual technical texts in the target technical field; A similarity comparison module, used for calculating the similarity between each first semantic vector and each second semantic vector, and obtaining a target semantic vector from each second semantic vector according to the similarity, wherein the second semantic vector is a semantic vector of a technical text of a second object in the target technical field; A competition intensity acquisition module, configured to obtain, for each second object, the competition intensity between the second object and the first object according to the similarity corresponding to each target semantic vector corresponding to the second object, wherein the competition intensity represents the degree of technical similarity between the first object and the second object; A competitor determination module determines a technical competitor of the first object from each second object according to the competition intensity of the second object.

9. An electronic device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. 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 7 are implemented.

11. 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 7 are implemented.

Images