Medical document translation method, device and electronic device

Abstract

The present application discloses a medical document translation method, device and electronic device, which, after translating the medical document content through a medical machine translation model, introduces a semantic understanding algorithm based on artificial intelligence and natural language processing at the back end to perform segment-level semantic understanding of the medical document translation content, and guides and optimizes the expression of the semantics of the medical document translation content based on the full-text content topic tags between these different segment contents, thereby achieving polishing and optimization of the medical document translation content. This helps to enable the semantics of the translated medical document content to accurately grasp the intention and background theme of the original text, and improve the intelligence level and quality of medical document translation.

Description

Technical Field

[0001] The present application relates to the field of intelligent translation, and more specifically, to a medical document translation method, device and electronic device. Background Art

[0002] With the accelerated development of globalization, cross-border medical cooperation and communication are becoming more frequent, and the demand for translation of medical documents is also increasing. These documents may include clinical trial reports, drug instructions, research papers, etc., which require extremely high professionalism and accuracy. However, traditional medical document translation solutions mainly rely on manual translation or basic machine translation, which has many defects. For example: First, manual translation is time-consuming and costly, and the quality of translation is highly dependent on the translator's professional knowledge and experience. Once the translator has an inaccurate understanding of certain professional terms or concepts, it may lead to translation errors. Secondly, although basic machine translation can quickly generate translation results, it performs poorly when dealing with complex professional terms and contexts, and is prone to problems such as inconsistent terminology and deviations in context understanding, which in turn affects the quality of translation and hinders efficient and accurate cross-border medical exchanges and cooperation.

[0003] Therefore, an optimized medical document translation solution is desired. Summary of the invention

[0004] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a medical document translation method, device and electronic device, which, after translating the medical document content through a medical machine translation model, introduces a semantic understanding algorithm based on artificial intelligence and natural language processing at the back end to perform segment-granular semantic understanding of the medical document translation content, and guides the optimization of the expression of the semantics of the medical document translation content based on the full-text content topic tags between these different segments, thereby achieving polishing optimization of the medical document translation content. This helps to enable the semantics of the translated medical document content to accurately grasp the intent and background theme of the original text, and improve the intelligence level and quality of medical document translation.

[0005] According to one aspect of the present application, a medical document translation method is provided, which includes: obtaining a medical document to be translated uploaded by a user; extracting the text content of the medical document to be translated from the medical document to be translated; performing segment granularity decomposition on the text content of the medical document to be translated and obtaining a sequence of medical document segment translation content through a trained medical machine translation model; performing semantic embedding coding on each medical document segment translation content in the sequence of medical document segment translation content to obtain a sequence of medical document segment translation content semantic embedding coding vectors; based on the sequence of medical document segment translation content semantic embedding coding vectors, determining the global segment content theme between each medical document segment translation content to obtain a segment content theme label full-text distribution cascade coding vector; performing segment content theme label full-text distribution cascade coding vector based on segment content theme label query-guided medical document segment translation content semantic enhancement processing on the sequence of medical document segment translation content semantic embedding coding vectors to obtain a sequence of theme-guided medical document segment translation content semantic enhancement coding vectors; performing translation polishing based on the sequence of theme-guided medical document segment translation content semantic enhancement coding vectors to generate a sequence of polished medical document segment translation content.

[0006] According to another aspect of the present application, a medical document translation device is provided, which includes: a medical document acquisition module for acquiring a medical document to be translated uploaded by a user; a text content extraction module for extracting text content of the medical document to be translated from the medical document to be translated; a medical machine translation module for performing segment granularity decomposition on the text content of the medical document to be translated and obtaining a sequence of medical document segment translation content through a trained medical machine translation model; a semantic embedding encoding module for performing semantic embedding encoding on each medical document segment translation content in the sequence of medical document segment translation content to obtain a sequence of semantic embedding encoding vectors of medical document segment translation content; a global segment content theme determination module for determining the subject of the medical document segment translation based on the segment granularity decomposition. A sequence of semantically embedded coding vectors of translated content of medical document segments, determining the global segment content theme between the translated content of each medical document segment to obtain a segment content theme label full-text distribution cascade coding vector; a medical document segment translated content semantic enhancement module, used to perform medical document segment translated content semantic enhancement processing based on segment content theme label query guidance on the sequence of full-text distribution cascade coding vectors of the segment content theme label and the sequence of semantically embedded coding vectors of the translated content of medical document segments to obtain a sequence of theme-guided medical document segment translated content semantic enhancement coding vectors; a translation polishing module, used to perform translation polishing based on the sequence of theme-guided medical document segment translated content semantic enhancement coding vectors to generate a sequence of polished translated content of medical document segments.

[0007] According to another aspect of the present application, an electronic device is provided, comprising: a processor; and a memory, wherein computer program instructions are stored in the memory, and when the computer program instructions are executed by the processor, the processor executes the medical document translation method.

[0008] Compared with the prior art, the medical document translation method, device and electronic device provided by the present application, after translating the medical document content through the medical machine translation model, introduces a semantic understanding algorithm based on artificial intelligence and natural language processing at the back end to perform segment-level semantic understanding of the medical document translation content, and guides and optimizes the expression of the semantics of the medical document translation content based on the full-text content topic tags between these different segments, thereby achieving polishing and optimization of the medical document translation content. This helps the semantics of the translated medical document content to accurately grasp the intention and background theme of the original text, and improves the intelligence level and quality of medical document translation. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 Flowchart of a medical document translation method according to an embodiment of the present application.

[0011] Figure 2 Schematic diagram of data flow of a medical document translation method according to an embodiment of the present application.

[0012] Figure 3 4 is a block diagram of a medical document translation device according to an embodiment of the present application. DETAILED DESCRIPTION

[0013] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0014] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0015] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0016] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0017] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0018] In recent years, with the development of artificial intelligence technology, especially the application of deep learning in the field of natural language processing, machine translation technology has made significant progress. Especially in specific fields, such as machine translation in the medical field, the accuracy and efficiency of translation can be greatly improved by using specially trained models. However, medical document translation is not just a simple text conversion. The same medical term may have different meanings in different contexts. Therefore, when translating medical documents, it is also necessary to consider the consistency of terminology, the correct understanding of the context, and the accurate communication of professional knowledge.

[0019] In the technical solution of the present application, a medical document translation method is proposed. Figure 1 Flowchart of a medical document translation method according to an embodiment of the present application. Figure 2 FIG. 1 is a data flow diagram of a medical document translation method according to an embodiment of the present application. Figure 1 and Figure 2As shown, according to the embodiment of the present application, the medical document translation method includes the following steps: S1, obtaining a medical document to be translated uploaded by a user; S2, extracting the text content of the medical document to be translated from the medical document to be translated; S3, performing segment granularity decomposition on the text content of the medical document to be translated and obtaining a sequence of medical document segment translation contents through a trained medical machine translation model; S4, performing semantic embedding coding on each medical document segment translation content in the sequence of medical document segment translation contents to obtain a sequence of semantic embedding coding vectors of medical document segment translation contents; S5, performing semantic embedding coding on the translation content of the medical document segment based on the semantic embedding coding of the translation content of the medical document segment. S6, performing segment content topic label full-text distribution cascade coding vector on the segment content topic label full-text distribution cascade coding vector and the sequence of medical document segment translation content semantic embedding coding vectors based on segment content topic label query guidance to obtain a sequence of topic-guided medical document segment translation content semantic enhancement coding vectors; S7, performing translation polishing based on the sequence of topic-guided medical document segment translation content semantic enhancement coding vectors to generate a sequence of polished medical document segment translation content.

[0020] In particular, S1 obtains the medical document to be translated uploaded by the user. In order to facilitate users to upload medical documents to be translated, it is necessary to first design a user-friendly front-end page. This page can be a web page or mobile application interface that provides a file upload function. Specifically, the front-end page should include a file selector button, and the user can select a local file to upload after clicking it. The file selector should support multiple file formats, such as PDF, DOCX, TXT, etc., to ensure that users can upload different types of medical documents. In addition, there should be an obvious upload button on the page, and the user can upload the file by clicking the button after selecting the file.

[0021] To improve the user experience, the front-end page should also display a progress bar for file uploads to let users know the status of the file upload. After the file is successfully uploaded, a success prompt is displayed to the user, and guidance for the next step is provided, such as a start translation button. If the file upload fails, an error message is displayed to the user, and an option to re-upload is provided. Common error messages include incorrect file format, file size exceeding the limit, network connection problems, etc.

[0022] The backend needs to design a file upload interface to receive files uploaded by users. This interface is usually an HTTP POST request that carries file data. The specific implementation steps are as follows: First, define a file upload interface, such as / api / upload, to accept POST requests. After receiving the file, temporarily store it on the server, or directly save it to a cloud storage service such as Alibaba Cloud OSS. Verify the received file to ensure that the file format is correct, the size is within the allowed range, and there are no viruses or malicious code. Common verification methods include checking file extensions, file size limits, virus scanning, etc.

[0023] After the file is successfully uploaded, it is necessary to parse the file and extract the content for subsequent translation processing. The specific steps are as follows: Use the corresponding file parsing library to read the file content and extract the text information. Commonly used file parsing libraries include Python's PyPDF2, docx, etc. For PDF files, use the PyPDF2 library to read the PDF file content and extract the text of each page. For DOCX files, use the docx library to read the Word document content and extract the text of each paragraph. For TXT files, read the TXT file content directly. Extract the text content to be translated from the parsed file. For PDF files, you may need to process multiple pages of content; for Word documents, you may need to process non-text content such as tables and pictures. Clean the extracted text content to remove irrelevant characters, blank lines, etc. to ensure the accuracy of subsequent processing.

[0024] In order to improve the user experience, timely status feedback needs to be provided on the user interface. Display a progress bar for file upload on the user interface to let users know the status of the file upload. After the file is successfully uploaded, a success prompt is displayed to the user, and guidance for the next step is provided, such as a start translation button. If the file upload fails, an error message is displayed to the user, and an option to re-upload is provided. Common error messages include incorrect file format, file size exceeding the limit, network connection problems, etc.

[0025] In particular, S2 extracts the text content of the medical document to be translated from the medical document to be translated. In one example, the process of extracting text content from the medical document to be translated includes two main steps: file parsing and content extraction. The file parsing stage uses appropriate tools and libraries to read file contents of different formats to ensure that the text can be accurately extracted. The content extraction stage further processes the extracted text, including removing irrelevant characters, segmenting, processing special formats and content cleaning, etc., to ensure that the text content is clean and tidy and suitable for subsequent translation processing. Through these steps, high-quality text content can be effectively extracted from the medical document to be translated, laying a solid foundation for subsequent translation work.

[0026] In particular, the S3, after the segment granularity decomposition of the medical document text content to be translated, is used to obtain a sequence of medical document segment translation content through a trained medical machine translation model. That is, in a specific example of the present application, first, the medical document text content to be translated is segmented to obtain a sequence of medical document segment content to be translated; considering that when the medical document text content to be translated is translated and polished and optimized, if the long-length document text content is input into the translation model at one time, the translation accuracy may decrease due to context information overload. Therefore, the medical document text content to be translated is further segmented to obtain a sequence of medical document segment content to be translated. The segmentation process can ensure that each medical document paragraph to be translated can obtain sufficient context information during translation and semantic understanding, thereby improving the accuracy and coherence of the translation. Then, each medical document segment content to be translated in the sequence of medical document segment content to be translated is input into the trained medical machine translation model to obtain a sequence of medical document segment translation content. Each of the medical document segments to be translated in the sequence of medical document segments to be translated is input into the trained medical machine translation model to obtain a sequence of medical document segment translation contents. Through the processing of the medical machine translation model, each of the medical document segments to be translated can be translated to generate a medical document segment translation content corresponding to each of the medical document segments to be translated.

[0027] In one example, a medical machine translation model refers to a translation system designed specifically for medical literature, clinical reports, drug instructions, and other medical-related texts. These models usually need to handle highly specialized terminology and complex sentence structures, so they have some specific requirements and features in their design. The following are examples of several common medical machine translation models:

[0028] 1. Statistical Machine Translation (SMT): The SMT model learns translation patterns by analyzing a large amount of bilingual text. For medical translation, SMT can be trained using a large amount of existing medical literature to improve the translation quality in specific fields.

[0029] 2. Neural Machine Translation (NMT): The NMT model is one of the most popular machine translation methods. It uses deep learning technology to build an end-to-end translation model. NMT can better understand the context and performs particularly well in processing long and complex sentences.

[0030] 3. Hybrid model: This model combines rule-based methods with statistical or neural network methods. For example, a specific medical term can be processed first with a rule-based method, and then the NMT model can be used to process the rest, which can ensure the accuracy of the term translation and maintain a natural and fluent translation effect.

[0031] 4. Multimodal medical translation model: The multimodal medical translation model not only considers text information, but also combines other forms of information such as images and videos for translation. For example, when translating documents about surgical procedures, surgical videos can be referenced at the same time to make the translation more accurate.

[0032] 5. Domain-adaptive translation model: Train specialized translation models based on the characteristics of different medical subfields (such as cardiology, oncology, etc.). This model can more accurately understand and translate the professional terms and technical details of a specific field.

[0033] 6. Interactive medical translation system: allows users to participate in the translation process, provide feedback or correct the translation results, so as to gradually improve the accuracy and applicability of the translation. This system is particularly suitable for handling medical literature translation tasks that require high precision.

[0034] Each model has its advantages and limitations. In practical applications, we may choose a suitable model or combine multiple models according to specific translation needs.

[0035] In particular, the S4 performs semantic embedding encoding on each medical document segment translation content in the sequence of the medical document segment translation content to obtain a sequence of semantic embedding encoding vectors of the medical document segment translation content. It should be understood that the semantic embedding encoding method can convert the text data of each medical document segment translation content into a vector form, which is convenient for subsequent processing and analysis. At the same time, the semantic embedding encoding feature information in each medical document segment translation content can be captured through semantic embedding, including the relationship between words, the semantic dependency feature information of the context, etc., so that the model can better understand the content and structure of each paragraph, so as to make more reasonable adjustments in the subsequent translation polishing process.

[0036] In particular, the S5, based on the sequence of the semantic embedding coding vectors of the translated content of the medical document segments, determines the global segment content theme between the translated content of each medical document segment to obtain the segment content theme label full-text distribution cascade coding vector. That is, in a specific example of the present application, first, the semantic embedding coding vectors of the translated content of each medical document segment in the sequence of the semantic embedding coding vectors of the translated content of the medical document segment are input into the segment content theme identifier based on the classifier to obtain the sequence of segment content theme label coding vectors; it should be understood that by semantically embedding the translated content of each medical document segment granularity, the deep semantic information of the paragraph can be captured. Moreover, the obtained semantic embedding coding vectors of the translated content of these medical document segments contain rich context and semantic features, so that the model can more accurately identify the theme of each paragraph, which is more accurate and reliable than simple keyword matching or rule matching. Based on this, in the technical solution of the present application, the semantic embedding coding vectors of the translated content of each medical document segment in the sequence of the semantic embedding coding vectors of the translated content of the medical document segment are further input into the segment content theme identifier based on the classifier to obtain the sequence of segment content theme label coding vectors. Next, the sequence of the segment content topic label encoding vectors is cascaded to obtain the segment content topic label full-text distribution cascade encoding vector. Here, by cascading the topic label encoding features of each segment content, the coherence and consistency of the entire document in terms of subject can be ensured, and a more comprehensive understanding of the global topic distribution information of the medical document text can be obtained, which will help to make more reasonable polishing and adjustments to the subsequent medical document translation content. This global perspective also helps to avoid conflicts or inconsistencies in the subject of different paragraphs, and ensure the overall logic and coherence of the document.

[0037] In particular, the S6 performs segment content topic label full-text distribution cascade coding vector and the sequence of medical document segment translation content semantic embedding coding vector based on segment content topic label query-guided medical document segment translation content semantic enhancement processing to obtain a sequence of topic-guided medical document segment translation content semantic enhancement coding vector. It should be understood that the segment content topic label full-text distribution cascade coding vector and the medical document segment translation content semantic embedding coding vector represent the subject information and semantic information of the medical document text content respectively. Both belong to different modes of medical document text content, but both have an important impact on translation quality and polishing optimization tasks. Specifically, the segment content topic label full-text distribution information in the segment content topic label full-text distribution cascade coding vector can be used as a label for segment content semantic understanding and translation polishing optimization, so as to guide the accurate implementation of subsequent tasks, which is very important for understanding the background and context of the medical document segment translation content. Therefore, in the technical solution of the present application, the sequence of the segment content topic label full-text distribution cascade coding vector and the medical document segment translation content semantic embedding coding vector is further processed based on the segment content topic label query-guided medical document segment translation content semantic enhancement processing to obtain a sequence of topic-guided medical document segment translation content semantic enhancement coding vectors, so as to dynamically guide and enhance the expression of the semantic features of the medical document segment translation content according to the full-text distribution of the subject of the medical document segment content, thereby enhancing the ability to understand the semantic information of the medical document segment content and the generalization ability of the model to cope with different translation scenarios, and providing a basis for subsequent medical document translation content polishing optimization tasks. Furthermore, through the process of semantic reinforcement learning of medical document segment translation content guided by segment content topic label query, the model can better capture the complex relationship between segment content theme and semantics to overcome the problem of cross-modal information isolation, and can still maintain good performance when facing unseen medical document content.

[0038] Specifically, in an embodiment of the present application, the specific steps of performing a segment content topic label full-text distribution cascade coding vector and a sequence of the medical document segment translation content semantic embedding coding vector based on a segment content topic label query-guided medical document segment translation content semantic enhancement processing include: first, using the segment content topic label full-text distribution cascade coding vector as a query vector, and using the medical document segment translation content semantic embedding key matrix obtained by arranging the sequence of the medical document segment translation content semantic embedding coding vector as a key matrix, calculating the query hint semantic feature between the query vector and the key matrix to obtain a segment topic label-segment content semantic query hint semantic coding vector. That is, in the technical solution of the present application, the sequence of the medical document segment translation content semantic embedding coding vector is firstly matrix-arranged to obtain a medical document segment translation content semantic embedding key matrix; then, using the segment content topic label full-text distribution cascade coding vector as a query vector and using the medical document segment translation content semantic embedding key matrix as a key matrix, the query vector and the key matrix are input into a prompt learning network based on a converter structure to obtain the segment topic label-segment content semantic query hint semantic coding vector.

[0039] More specifically, in the above embodiment, the segment content topic label full-text distribution cascade coding vector is used as the query vector, and the medical document segment translation content semantic embedding key matrix obtained by the sequence arrangement of the medical document segment translation content semantic embedding coding vector is used as the key matrix, and the query hint semantic feature between the query vector and the key matrix is ​​calculated to obtain the segment topic label-segment content semantic query hint semantic coding vector, including: using the segment content topic label full-text distribution cascade coding vector as the query vector, and the medical document segment translation content semantic embedding key matrix obtained by the sequence arrangement of the medical document segment translation content semantic embedding coding vector as the key matrix, and calculating the query hint semantic feature between the query vector and the key matrix to obtain the segment topic label-segment content semantic query hint semantic coding vector according to the following query formula; wherein, the query formula is:

[0040] M={v1,v2,...,v j ,...,v m}

[0041]

[0042] Among them, v1, v2, v j and v mare the first, second, jth and mth medical document segment translation content semantic embedding coding vectors in the sequence of the medical document segment translation content semantic embedding coding vectors, M is the medical document segment translation content semantic embedding key matrix composed of the sequence of the medical document segment translation content semantic embedding coding vectors, Transformer(·,·) is the Transformer encoding, v s is the segment content topic label full text distribution cascade encoding vector, v j T Yes j The transposed vector of is vector multiplication, ||·|| is the vector norm, s j v s and v j The semantic association between the segment topic label and the segment content, exp(·) represents the value of the natural exponential function with the natural constant e as the base, s k v s The segment topic label-segment content semantic correlation between the k-th medical document segment translation content semantic embedding encoding vector, e j v j The corresponding segment topic label-segment content semantic weight coefficient, m is the number of vectors in the sequence of the semantic embedding encoding vector of the medical document segment translation content, and v t is the segment topic label - segment content semantic query hint semantic encoding vector.

[0043] Next, the segment topic label-segment content semantic query hint semantic coding vector is used as the query hint semantic feature, and the sequence of the medical document segment translation content semantic embedding coding vector is semantically guided to obtain the sequence of the topic-guided medical document segment translation content semantic enhancement coding vector. That is, in the technical solution of the present application, the segment content topic label full-text distribution cascade coding vector and the segment topic label-segment content semantic query hint semantic coding vector are first input into the cross-modal mask network based on the hint information to obtain the segment topic label-segment content semantic cross-modal feature mask weight matrix based on the hint information; and based on the segment topic label-segment content semantic cross-modal feature mask weight matrix based on the hint information, the medical document segment translation content semantic embedding key matrix is ​​feature-guided and optimized to obtain the topic-guided medical document segment translation content semantic enhancement coding matrix; and the topic-guided medical document segment translation content semantic enhancement coding matrix is ​​vectorized and decomposed to obtain the sequence of the topic-guided medical document segment translation content semantic enhancement coding vectors.

[0044] More specifically, in the above embodiment, the segment topic label-segment content semantic query hint semantic coding vector is used as a query hint semantic feature, and the sequence of the medical document segment translation content semantic embedding coding vectors is semantically guided encoded to obtain the sequence of the topic-guided medical document segment translation content semantic reinforcement coding vectors, including: using the segment topic label-segment content semantic query hint semantic coding vector as a query hint semantic feature, and using the following semantic guidance coding formula to semantically guide the sequence of the medical document segment translation content semantic embedding coding vectors to obtain the sequence of the topic-guided medical document segment translation content semantic reinforcement coding vectors; wherein the semantic guidance coding formula is:

[0045]

[0046] M c =M⊙S t

[0047] Decompose(M c )={v1',v2',...,v m '}

[0048] Among them, v t T v t , L is the scale of the semantic encoding vector of segment topic label-segment content semantic query prompt, softmax(·) is the softmax function, S t is the segment topic label-segment content semantic cross-modal feature mask weight matrix based on the prompt information, ⊙ is the position point multiplication, M c is the semantic enhancement encoding matrix of the topic-guided medical document segment translation content, Decompose is the vectorized decomposition of the matrix, v1', v2', v m ' are respectively the 1st, 2nd and mth topic-guided medical document segment translation content semantic enhancement coding vectors in the sequence of the topic-guided medical document segment translation content semantic enhancement coding vectors.

[0049] In particular, the S7 performs translation polishing based on the sequence of semantically enhanced coding vectors of the subject-guided translation content of the medical document segment to generate a sequence of the polished translation content of the medical document segment. In a specific example of the present application, the sequence of semantically enhanced coding vectors of the subject-guided translation content of the medical document segment is input into a translation polisher based on the AIGC model to obtain the sequence of the polished translation content of the medical document segment. That is, translation polishing is performed using the semantically enhanced features of the medical document translation content guided by the full-text distribution of the subject of the medical document segment content, thereby achieving polishing optimization of the translation content of the medical document. This helps to enable the semantics of the translated medical document content to accurately grasp the intention and background theme of the original text, and improve the intelligence level and quality of medical document translation.

[0050] AIGC (Artificial Intelligence Generated Content) models refer to models that use artificial intelligence technology to generate content. These models are usually based on deep learning frameworks and can understand and generate various types of content such as natural language, images, audio, etc. The core of the AIGC model lies in its powerful generation ability, which can generate high-quality content that meets specific requirements given certain input conditions.

[0051] In the field of natural language processing, AIGC models mainly include the following types: Transformer models, such as BERT, GPT, T5, etc. These models capture long-distance dependencies through the self-attention mechanism (Self-Attention) and can generate coherent and semantically rich text; sequence-to-sequence models (Seq2Seq), which are usually composed of encoders and decoders and are widely used in tasks such as machine translation and text summarization; generative adversarial networks (GANs), which generate realistic text content through adversarial training of generators and discriminators; recurrent neural networks (RNNs), especially long short-term memory networks (LSTMs) and gated recurrent units (GRUs), can process sequence data and are often used in text generation tasks.

[0052] In the medical document translation solution, the AIGC model is mainly used in the translation polishing stage to ensure the accuracy and fluency of the translation results by further optimizing the translated content.

[0053] First, a pre-trained machine translation model (such as BioBERT) is used to perform a preliminary translation of the original medical document to generate preliminary translation results. The main purpose of this step is to quickly generate translated texts to provide a basis for subsequent optimization.

[0054] Next, the initially translated paragraphs are semantically embedded and encoded to generate semantically enhanced encoding vectors. This step encodes the translated text through a pre-trained language model (such as BioBERT) to capture its deep semantic information. Through this encoding, the model can better understand the semantic structure and contextual relationships of the text.

[0055] Then, topic labels are generated based on the semantic embedding encoding vectors. This can be achieved through clustering algorithms (such as K-means) or classification models (such as Logistic Regression). Topic labels help capture the topic relationship between paragraphs and provide a basis for subsequent semantic enhancement encoding. By generating topic labels, the model can be better guided to focus on specific topics and content during the translation polishing process.

[0056] Next, the topic tags and semantic embedding encoding vectors are combined to semantically enhance the translation content. This is achieved by inputting the topic tags into the model as additional features, enhancing the model's understanding of the topic of the paragraph. Semantic enhancement encoding allows the model to more accurately reflect the intent and context of the original text when generating translation results.

[0057] Finally, the translation polisher based on the AIGC model is used to polish the semantically enhanced encoded paragraphs. The specific steps are as follows: First, load a pre-trained model suitable for text generation tasks, such as T5 or GPT-2. Then, combine the preliminary translated text and the semantically enhanced encoded vector into an input tensor. Next, input the input tensor into the pre-trained model for forward propagation to generate the polished translated text. Finally, extract the text generated by the model as the final translation result.

[0058] By using the AIGC model, high-quality translation content can be generated in a short time, significantly shortening the translation time and meeting the needs of large-scale medical document translation. The AIGC model has strong generation and semantic understanding capabilities, and can generate coherent and accurate translation texts, especially when dealing with professional terminology and complex contexts. In addition, through semantic enhancement encoding and topic tag generation, the professional terminology and context of the translated content are ensured to remain consistent, avoiding misunderstandings caused by inconsistent terminology. The solution can adapt to different types of medical documents, including but not limited to clinical trial reports, drug instructions, and research papers. It reduces the reliance on high-level manual translation, reduces translation costs, and improves economic benefits.

[0059] Preferably, inputting the sequence of semantically enhanced coding vectors of the subject-guided medical document segment translation content into a translation polisher based on the AIGC model to obtain a sequence of polished medical document segment translation content includes: performing feature clustering on the feature set of the sequence of semantically enhanced coding vectors of the subject-guided medical document segment translation content to obtain a feature set within the subject-guided medical document segment translation content semantically enhanced coding class and a feature set outside the subject-guided medical document segment translation content semantically enhanced coding class, that is:

[0060]

[0061] Among them, f 1i represents the set of semantically enhanced features encoded within the translation content of the topic-guided medical document segment, f 2j It represents a set of out-of-class features of semantically enhanced encoding of the topic-guided medical document segment translation content.

[0062] The ratio of the number of feature values ​​in the feature set of the subject-guided medical document segment translation content semantic enhancement encoding class to the number of feature values ​​in the feature set of the sequence of subject-guided medical document segment translation content semantic enhancement encoding vectors is calculated, that is:

[0063]

[0064] Among them, k represents the number of feature values ​​in the feature set within the semantic enhancement coding class of the topic-guided medical document segment translation content, n represents the number of feature values ​​in the feature set of the sequence of semantic enhancement coding vectors of the topic-guided medical document segment translation content, and λ represents the ratio of the number of feature values ​​in the feature set within the semantic enhancement coding class of the topic-guided medical document segment translation content to the number of feature values ​​in the feature set of the sequence of semantic enhancement coding vectors of the topic-guided medical document segment translation content.

[0065] The ratio of the λ-th power of the sum of the absolute values ​​of all feature values ​​in the feature set of the subject-guided medical document segment translation content semantic enhancement coding class to the λ-th power of the sum of the absolute values ​​of all feature values ​​in the feature set of the sequence of subject-guided medical document segment translation content semantic enhancement coding vectors is calculated to obtain the subject-guided medical document segment translation content semantic enhancement coding modulation weight, that is:

[0066]

[0067] Among them, f p The feature set representing the sequence of semantic enhancement coding vectors of the topic-guided medical document segment translation content, w1 represents the modulation weight of the semantic enhancement coding of the topic-guided medical document segment translation content.

[0068] The ratio of the λ / 2 power of the sum of squares of all feature values ​​in the feature set of the subject-guided medical document segment translation content semantic enhancement encoding class to the λ / 2 power of the sum of squares of all feature values ​​in the feature set of the sequence of subject-guided medical document segment translation content semantic enhancement encoding vectors is calculated to obtain the subject-guided medical document segment translation content semantic enhancement encoding harmonic weight, that is:

[0069]

[0070] Wherein, w2 represents the semantic enhancement encoding and reconciliation weight of the translation content of the topic-guided medical document segment.

[0071] For each feature value in the feature set of the topic-guided medical document segment translation content semantic enhancement coding class, the product of the feature value and the topic-guided medical document segment translation content semantic enhancement coding harmonic weight is calculated, and then the modulated weight of the topic-guided medical document segment translation content semantic enhancement coding is added to obtain the optimized feature value, that is:

[0072] f' 1i =w2×f 1i +w1

[0073] Among them, f' 1i The optimized feature value of each feature value in the feature set of the semantic enhancement encoding class of the topic-guided medical document segment translation content is represented.

[0074] For each feature value in the out-of-class feature set of the topic-guided medical document segment translation content semantic enhancement encoding, the product of the feature value and the topic-guided medical document segment translation content semantic enhancement encoding tuning weight is calculated to obtain the optimized feature value, that is:

[0075] f 2j '=w1×f 2j

[0076] Among them, f 2j ' represents the optimized feature value of each feature value in the out-of-class feature set of the subject-guided medical document segment translation content semantic enhancement encoding.

[0077] A sequence of optimized topic-guided medical document segment translation content semantic enhancement encoding vectors based on the optimized feature values ​​of the topic-guided medical document segment translation content semantic enhancement encoding in-class feature set and the topic-guided medical document segment translation content semantic enhancement encoding out-class feature set is input into a translation polisher based on the AIGC model to obtain a sequence of polished medical document segment translation content.

[0078] Here, when the sequence of semantic embedded coding vectors of the translated content of the medical document segment and the full-text distribution cascade coding vectors of the segment content topic tags respectively represent the embedded coding semantic features of the translated content of the medical document and its topic tag coding features, when performing cross-modal joint encoding assisted by continuous prompts, it is also expected to improve the cross-modal semantically consistent aggregation expression effect of the sequence of semantically reinforced coding vectors of the topic-guided translated content of the medical document segment.

[0079] Therefore, while performing feature clustering on the sequence of semantically enhanced coding vectors of the topic-guided medical document segment translation content, the interaction of key feature information of the sequence of semantically enhanced coding vectors of the topic-guided medical document segment translation content in the clustering process is described, and a geometric equivariant topology of features is constructed by low-rank harmonic modulation of the clustering features of the sequence of semantically enhanced coding vectors of the topic-guided medical document segment translation content and the equivariance of the features as a whole, so as to obtain the schematic distribution translation and rotational symmetry of the clustering features of the sequence of semantically enhanced coding vectors of the topic-guided medical document segment translation content relative to the features as a whole, thereby introducing geometric message passing into the feature expression of the sequence of semantically enhanced coding vectors of the topic-guided medical document segment translation content, and realizing the cluster mapping symmetry of the sequence of semantically enhanced coding vectors of the topic-guided medical document segment translation content through manipulation of irreducible low-rank order coefficients, thereby improving the consistency of clustering-based feature representation of the sequence of semantically enhanced coding vectors of the topic-guided medical document segment translation content, thereby improving the text quality of the sequence of polished medical document segment translation content obtained by inputting the sequence of semantically enhanced coding vectors of the topic-guided medical document segment translation content into the translation polisher based on the AI ​​GC model.

[0080] In summary, the medical document translation method according to the embodiment of the present application is explained, which, after translating the medical document content through the medical machine translation model, introduces a semantic understanding algorithm based on artificial intelligence and natural language processing at the back end to perform segment-level semantic understanding of the medical document translation content, and guides and optimizes the expression of the semantics of the medical document translation content based on the full-text content topic tags between these different segments, thereby achieving polishing optimization of the medical document translation content. This helps to enable the semantics of the translated medical document content to accurately grasp the intention and background theme of the original text, and improve the intelligence level and quality of medical document translation.

[0081] Furthermore, a medical document translation device is also provided.

[0082] Figure 3 FIG. 1 is a block diagram of a medical document translation device according to an embodiment of the present application. Figure 3As shown, according to the embodiment of the present application, the medical document translation device 300 includes: a medical document acquisition module 310 for acquiring a medical document to be translated uploaded by a user; a text content extraction module 320 for extracting the text content of the medical document to be translated from the medical document to be translated; a medical machine translation module 330 for performing segment granularity decomposition on the text content of the medical document to be translated and obtaining a sequence of medical document segment translation contents through a trained medical machine translation model; a semantic embedding coding module 340 for performing semantic embedding coding on each medical document segment translation content in the sequence of medical document segment translation contents to obtain a sequence of semantic embedding coding vectors of medical document segment translation contents; a global segment content theme determination module 350 for determining the subject of the medical document segment translation based on the segment granularity decomposition. Based on the sequence of the medical document segment translation content semantic embedding coding vectors, the global segment content theme between the translation contents of each medical document segment is determined to obtain the segment content theme label full-text distribution cascade coding vector; the medical document segment translation content semantic enhancement module 360 ​​is used to perform medical document segment translation content semantic enhancement processing based on segment content theme label query guidance on the sequence of the segment content theme label full-text distribution cascade coding vector and the medical document segment translation content semantic embedding coding vector to obtain a sequence of theme-guided medical document segment translation content semantic enhancement coding vectors; the translation polishing module 370 is used to perform translation polishing based on the sequence of theme-guided medical document segment translation content semantic enhancement coding vectors to generate a sequence of polished medical document segment translation content.

[0083] As described above, the medical document translation device 300 according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with a medical document translation algorithm. In a possible implementation, the medical document translation device 300 according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the medical document translation device 300 can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the medical document translation device 300 can also be one of the many hardware modules of the wireless terminal.

[0084] Alternatively, in another example, the medical document translation apparatus 300 and the wireless terminal may also be separate devices, and the medical document translation apparatus 300 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0085] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for translating medical documents, characterized in that: include: Obtain medical documents to be translated uploaded by users; Extracting text content of the medical document to be translated from the medical document to be translated; The medical document text content to be translated is decomposed into segments into granularity and then trained into a medical machine translation model to obtain a sequence of medical document segment translation contents; each medical document segment translation content in the sequence of medical document segment translation contents is semantically embedded and encoded to obtain a sequence of medical document segment translation content semantic embedded encoding vectors; based on the sequence of medical document segment translation content semantic embedded encoding vectors, the global segment content theme between each medical document segment translation content is determined to obtain a segment content theme label full-text distribution cascade encoding vector; the sequence of the segment content theme label full-text distribution cascade encoding vector and the medical document segment translation content semantic embedded encoding vector is subjected to a medical document segment translation content semantic enhancement processing guided by a segment content theme label query to obtain a sequence of theme-guided medical document segment translation content semantic enhancement encoding vectors; Perform translation polishing based on the sequence of semantically enhanced encoding vectors of the subject-guided medical document segment translation content to generate a sequence of polished medical document segment translation content; Among them, the sequence of the segment content topic label full-text distribution cascade coding vector and the medical document segment translation content semantic embedding coding vector is subjected to the segment content topic label query-guided medical document segment translation content semantic enhancement processing to obtain a sequence of topic-guided medical document segment translation content semantic enhancement coding vectors, including: taking the segment content topic label full-text distribution cascade coding vector as the query vector, and taking the medical document segment translation content semantic embedding key matrix obtained by arranging the sequence of the medical document segment translation content semantic embedding coding vectors as the key matrix, calculating the query hint semantic feature between the query vector and the key matrix to obtain the segment topic label-segment content semantic query hint semantic coding vector; taking the segment topic label-segment content semantic query hint semantic coding vector as the query hint semantic feature, performing semantic-guided coding on the sequence of the medical document segment translation content semantic embedding coding vectors to obtain the sequence of topic-guided medical document segment translation content semantic enhancement coding vectors; Among them, the segment topic label-segment content semantic query hint semantic coding vector is used as the query hint semantic feature, and the sequence of the medical document segment translation content semantic embedding coding vectors is semantically guided encoded to obtain the sequence of the topic-guided medical document segment translation content semantic enhancement coding vectors, including: inputting the segment content topic label full-text distribution cascade coding vector and the segment topic label-segment content semantic query hint semantic coding vector into a cross-modal mask network based on hint information to obtain a segment topic label-segment content semantic cross-modal feature mask weight matrix based on hint information; based on the segment topic label-segment content semantic cross-modal feature mask weight matrix based on hint information, the medical document segment translation content semantic embedding key matrix is ​​feature-guided optimized expression to obtain a topic-guided medical document segment translation content semantic enhancement coding matrix; the topic-guided medical document segment translation content semantic enhancement coding matrix is ​​vectorized and decomposed to obtain the sequence of the topic-guided medical document segment translation content semantic enhancement coding vectors.

2. The medical document translation method according to claim 1, characterized in that: After decomposing the text content of the medical document to be translated into segments, a sequence of medical document segment translation contents is obtained through a trained medical machine translation model, including: segmenting the text content of the medical document to be translated to obtain a sequence of medical document segment contents to be translated; and inputting each medical document segment content to be translated in the sequence of medical document segment contents to be translated into the trained medical machine translation model to obtain a sequence of medical document segment translation contents.

3. The medical document translation method according to claim 2, characterized in that: Based on the sequence of the medical document segment translation content semantic embedding coding vectors, the global segment content theme between the translation contents of each medical document segment is determined to obtain a segment content theme label full-text distribution cascade coding vector, including: inputting each medical document segment translation content semantic embedding coding vector in the sequence of the medical document segment translation content semantic embedding coding vector into a classifier-based segment content theme identifier to obtain a sequence of segment content theme label coding vectors; cascading the sequence of segment content theme label coding vectors to obtain the segment content theme label full-text distribution cascade coding vector.

4. The medical document translation method according to claim 3, characterized in that: The full-text distribution cascade coding vector of the segment content topic label is used as the query vector, and the medical document segment translation content semantic embedding key matrix obtained by arranging the sequence of the medical document segment translation content semantic embedding coding vectors is used as the key matrix, and the query prompt semantic features between the query vector and the key matrix are calculated to obtain the segment topic label-segment content semantic query prompt semantic coding vector, including: arranging the sequence of the medical document segment translation content semantic embedding coding vectors in a matrix to obtain the medical document segment translation content semantic embedding key matrix; the full-text distribution cascade coding vector of the segment content topic label is used as the query vector and the medical document segment translation content semantic embedding key matrix is ​​used as the key matrix, and the query vector and the key matrix are input into a prompt learning network based on a converter structure to obtain the segment topic label-segment content semantic query prompt semantic coding vector.

5. The medical document translation method according to claim 4, characterized in that: Based on the sequence of semantically enhanced coding vectors of the subject-guided medical document segment translation content, translation polishing is performed to generate a sequence of polished medical document segment translation content, including: inputting the sequence of semantically enhanced coding vectors of the subject-guided medical document segment translation content into a translation polisher based on the AIGC model to obtain the sequence of polished medical document segment translation content.

6. A medical document translation device, characterized in that: include: A module for obtaining medical documents to be translated, used to obtain medical documents to be translated uploaded by users; A text content extraction module, used to extract the text content of the medical document to be translated from the medical document to be translated; A medical machine translation module, used to decompose the text content of the medical document to be translated into segments and obtain a sequence of translated content of the medical document segments through a trained medical machine translation model; A semantic embedding coding module, used to perform semantic embedding coding on each medical document segment translation content in the sequence of medical document segment translation content to obtain a sequence of medical document segment translation content semantic embedding coding vectors; a global segment content theme determination module, used to determine the global segment content theme between each medical document segment translation content based on the sequence of medical document segment translation content semantic embedding coding vectors to obtain a segment content theme label full-text distribution cascade coding vector; a medical document segment translation content semantic enhancement module, used to perform segment content theme label full-text distribution cascade coding vector-based segment content theme label query-guided medical document segment translation content semantic enhancement processing on the sequence of medical document segment translation content semantic embedding coding vectors to obtain a sequence of theme-guided medical document segment translation content semantic enhancement coding vectors; a translation polishing module, used to perform translation polishing based on the sequence of theme-guided medical document segment translation content semantic enhancement coding vectors to generate a sequence of polished medical document segment translation content; Among them, the sequence of the segment content topic label full-text distribution cascade coding vector and the medical document segment translation content semantic embedding coding vector is subjected to the segment content topic label query-guided medical document segment translation content semantic enhancement processing to obtain a sequence of topic-guided medical document segment translation content semantic enhancement coding vectors, including: taking the segment content topic label full-text distribution cascade coding vector as the query vector, and taking the medical document segment translation content semantic embedding key matrix obtained by arranging the sequence of the medical document segment translation content semantic embedding coding vectors as the key matrix, calculating the query hint semantic feature between the query vector and the key matrix to obtain the segment topic label-segment content semantic query hint semantic coding vector; taking the segment topic label-segment content semantic query hint semantic coding vector as the query hint semantic feature, performing semantic-guided coding on the sequence of the medical document segment translation content semantic embedding coding vectors to obtain the sequence of topic-guided medical document segment translation content semantic enhancement coding vectors; Among them, the segment topic label-segment content semantic query hint semantic coding vector is used as the query hint semantic feature, and the sequence of the medical document segment translation content semantic embedding coding vectors is semantically guided encoded to obtain the sequence of the topic-guided medical document segment translation content semantic enhancement coding vectors, including: inputting the segment content topic label full-text distribution cascade coding vector and the segment topic label-segment content semantic query hint semantic coding vector into a cross-modal mask network based on hint information to obtain a segment topic label-segment content semantic cross-modal feature mask weight matrix based on hint information; based on the segment topic label-segment content semantic cross-modal feature mask weight matrix based on hint information, the medical document segment translation content semantic embedding key matrix is ​​feature-guided optimized expression to obtain a topic-guided medical document segment translation content semantic enhancement coding matrix; the topic-guided medical document segment translation content semantic enhancement coding matrix is ​​vectorized and decomposed to obtain the sequence of the topic-guided medical document segment translation content semantic enhancement coding vectors.

7. An electronic device, characterized in that: include: processor; and a memory, in which computer program instructions are stored, and when the computer program instructions are executed by the processor, the processor executes the medical document translation method according to any one of claims 1 to 5.

Images