WO2021179693A1

WO2021179693A1 - Medical text translation method and device, and storage medium

Info

Publication number: WO2021179693A1
Application number: PCT/CN2020/132476
Authority: WO
Inventors: 李春宇; 朱威; 张开明
Original assignee: 平安科技（深圳）有限公司
Priority date: 2020-10-19
Filing date: 2020-11-27
Publication date: 2021-09-16
Also published as: CN111950303B; CN111950303A

Abstract

A medical text translation method and device, and a storage medium, relating to the field of medical technology. Said method comprises: a medical text translation device acquiring medical text to be translated (101); the medical text translation device performing semantic feature extraction on said medical text to obtain a first feature vector (102); the medical text translation device acquiring a target feature vector corresponding to said medical text, the target feature vector being used to represent a medical knowledge graph corresponding to said medical text (103); the medical text translation device splicing the first feature vector with the target feature vector to obtain a second feature vector (104); and the medical text translation device translating said medical text according to the second feature vector (105). Said method facilitates improving the accuracy of medical text translation.

Description

Medical text translation method, device and storage medium

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on October 19, 2020, the application number is 202011115345.3, and the invention title is "medical text translation method, device and storage medium", the entire content of which is incorporated herein by reference Applying.

Technical field

This application relates to the technical field of text recognition, in particular to a medical text translation method, device and storage medium.

Background technique

Machine translation has gone through a long period of time. It has made great progress from statistical language models to deep learning models. The inventor realized that the current progress in translation is mainly reflected in the general translation field, such as the translation of everyday words. However, progress in the translation of medical texts has been slow. The main reason is that there are a large number of proper nouns and medical terms in the medical field. As a result, the translation of medical documents and the translation of sentences related to medical documents still has great defects, and translation errors often occur. For this situation, it is necessary Manual adjustment.

Therefore, the existing translation accuracy of medical texts is low, and the user experience is poor.

Summary of the invention

The embodiments of the present application provide a medical text translation method, device and storage medium. By combining medical knowledge graphs, the accuracy of medical text translation is improved.

In the first aspect, an embodiment of the present application provides a medical text translation method, including:

Obtain the medical text to be translated;

Performing semantic feature extraction on the medical text to be translated to obtain a first feature vector;

Acquiring a target feature vector corresponding to the medical text to be translated, where the target feature vector is used to represent a medical knowledge graph corresponding to the medical text to be translated;

Splicing the first feature vector with the target feature vector to obtain a second feature vector;

According to the second feature vector, the medical text to be translated is translated.

In the second aspect, an embodiment of the present application provides a medical text translation device, including:

The acquiring unit is used to acquire the medical text to be translated;

A processing unit, configured to perform semantic feature extraction on the medical text to be translated to obtain a first feature vector;

The acquiring unit is further configured to acquire a target feature vector corresponding to the medical text to be translated, and the target feature vector is used to represent a medical knowledge graph corresponding to the medical text to be translated;

The processing unit is further configured to splice the first feature vector and the target feature vector to obtain a second feature vector;

The processing unit is further configured to translate the medical text to be translated according to the second feature vector.

In a third aspect, an embodiment of the present application provides a medical text translation device, including a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and are The configuration is performed by the processor to implement the following methods:

Obtain the medical text to be translated;

In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, and the computer program causes a computer to execute the following method:

Obtain the medical text to be translated;

In a fifth aspect, embodiments of the present application provide a computer program product, the computer program product includes a non-transitory computer-readable storage medium storing a computer program, and the computer is operable to cause the computer to execute the computer program as described in the first aspect Methods.

In the implementation of the embodiments of this application, in the process of translating the medical text to be translated, the medical knowledge graph corresponding to the medical text to be translated is fused, so that the second feature vector can be fused with prior knowledge corresponding to the text to be translated. Then the accuracy of translation is improved, especially the accuracy of translation of medical terminology or medical terminology.

Description of the drawings

In order to more clearly describe the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed in the description of the embodiments. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative work, other drawings can be obtained from these drawings.

FIG. 1 is a schematic flowchart of a medical text translation method provided by an embodiment of the application;

FIG. 2 is a schematic diagram of a neural network provided by an embodiment of this application;

FIG. 3 is a schematic diagram of a self-attention mechanism provided by an embodiment of the application;

FIG. 4 is a schematic flowchart of a neural network training method provided by an embodiment of this application;

FIG. 5 is a schematic structural diagram of a medical text translation device provided by an embodiment of the application;

Fig. 6 is a block diagram of functional units of a medical text translation device provided by an embodiment of the application.

Detailed ways

The technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of them. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this application.

The terms "first", "second", "third" and "fourth" in the specification and claims of this application and the drawings are used to distinguish different objects, not to describe a specific order . In addition, the terms "including" and "having" and any variations of them are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes unlisted steps or units, or optionally also includes Other steps or units inherent to these processes, methods, products or equipment.

Reference to "embodiments" herein means that specific features, results or characteristics described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art clearly and implicitly understand that the embodiments described herein can be combined with other embodiments.

The technical solution of this application can be applied to the fields of artificial intelligence, smart city, digital medical, blockchain and/or big data technology to realize text translation, especially text translation in the medical field. Optionally, the data involved in this application, such as translated text, vectors, and/or tags, can be stored in a database, or can be stored in a blockchain, such as distributed storage through a blockchain, which is not limited by this application .

In order to facilitate the understanding of the technical solutions of the present application, the relevant terms involved in the present application are explained.

Medical knowledge graph: It is composed of a medical entity, a description corresponding to the medical entity (that is, an explanation of the medical entity), and a medical plan corresponding to the medical entity. For example, the gastric cancer medical knowledge map includes the medical entity "gastric cancer" of gastric cancer medicine, and its corresponding description is "gastric cancer is a malignant tumor that originates from the epithelium of the gastric mucosa", and its corresponding medical plans include: differences in gastric cancer, gastric cancer symptoms, and gastric cancer Diffusion and transfer pathways, and so on.

Refer to FIG. 1, which is a schematic flowchart of a medical text translation method provided by an embodiment of the application. This method is applied to medical text translation devices. The method includes the following steps:

101: The medical text translation device obtains the medical text to be translated.

Optionally, the medical text to be translated may be input by the user in the information input field of the medical text translation device.

102: The medical text translation device performs semantic feature extraction on the medical text to be translated to obtain a first feature vector.

Exemplarily, embedding is performed on each word in each text to be translated to obtain a word vector corresponding to each word. Among them, the word mentioned in this application is a complete word in Chinese and a complete word in English. The following words are similar to this and will not be described again.

The word embedding process for each word can be realized by one-hot encoding. For example, encoding can be performed according to the position of each word in the medical text to be translated. For example, the text to be translated is "I am a student", and one-hot encoding of each word can get the word vector corresponding to the word "I" as (1,0,0,0), and the word corresponding to the word "am" The vector is (0,1,0,0), the word vector corresponding to the word "a" is (0,0,1,0), and the word vector corresponding to the word "student" is (0,0,0,1).

Then, semantic feature extraction is performed according to the word vector corresponding to each word to obtain the first feature vector. Among them, the semantic feature extraction can be achieved through a semantic feature extraction network, which is pre-trained. The training process of the semantic feature extraction network is described later, and it will not be described here too much.

In an embodiment of the present application, the number of the semantic feature extraction network may be one or more. In the case that the number of the semantic feature extraction network is multiple, the output result of the previous semantic feature extraction network needs to be taken as The next semantic feature is extracted from the input data of the network. Exemplarily, each semantic feature extraction network may be a long and short-term memory network or a cyclic neural network, and so on.

In this application, the number of semantic feature extraction networks is taken as an example for illustration.

As shown in FIG. 2, the word vector corresponding to each word is input to the semantic feature extraction network, and the semantic feature extraction of the text to be translated is performed to obtain the first feature vector.

In an embodiment of the present application, the semantic feature extraction network also includes an attention module. Therefore, the word vector corresponding to each word is weighted by the attention module to obtain the target word vector corresponding to each word.

Exemplarily, as shown in FIG. 3, the word vector corresponding to word A is encoded to obtain the key value vector, query vector, and value vector corresponding to the word A, and the word A is the word A in the medical text to be translated Any word; then, determine the similarity between the query vector corresponding to the word A and the key value vector corresponding to each word, and use the similarity as the weight between the word A and each word; according to the word A and each word The weight between each word is weighted to the value vector corresponding to each word to obtain the target word vector corresponding to word A.

Exemplarily, the query vector corresponding to each word can be expressed by formula (1):

α _j ＝W _q ·φ _j (1)

Among them, 1≤j≤n, n is the number of words in the text to be translated, W _{q is} the first network parameter of the neural network, α _j is the query vector corresponding to the jth word in the n words, and φ _j is The word vector corresponding to the jth word, where n is an integer greater than or equal to 1.

Exemplarily, the key value vector corresponding to each word can be expressed by formula (2):

β _j ＝W _k ·φ _j (2)

Among them, W _{k is} the second network parameter of the neural network, and β _j is the key value vector corresponding to the j-th word.

Exemplarily, the value vector corresponding to each word can be expressed by formula (3):

λ _j =W _v ·φ _j (3)

Among them, W _{v is} the third network parameter of the neural network, and λ _j is the key value vector corresponding to the j-th word.

Then, determine the similarity between the query vector of the word A and the key value vector corresponding to each word, and obtain the weight between the word A and each word, for example, the weight between the word A and each word It can be expressed by formula (4):

Among them, ξ _j is the similarity between word A and the key value vector corresponding to the j-th word in the n words, that is, the weight between word A and the j-th word, and α _A is the corresponding to the A-th word Query vector, dist is the distance operation.

Finally, according to the weight between the word A and each word, the value vector corresponding to each word is weighted to obtain the fourth feature vector corresponding to the word A.

Exemplarily, the fourth feature vector corresponding to word A can be expressed by formula (5):

Among them, τ _A is the target word vector of word A, and λ _j is the value vector corresponding to the jth word.

It can be seen that according to the self-attention mechanism, the influence of the preceding and following words on the current word can be merged into the target word vector corresponding to the current word, instead of identifying each word in isolation, that is, fusing the context information of the current word. , Which can improve the accuracy of translation.

103: The medical text translation device obtains a target feature vector corresponding to the medical text to be translated, where the target feature vector is used to represent a medical knowledge graph corresponding to the medical text to be translated.

Exemplarily, all the medical knowledge graphs in the medical field may be vectorized first to obtain the third feature vector corresponding to each medical knowledge graph. Because the medical knowledge graph is essentially a relationship composed of multiple medical texts. Therefore, it is also possible to separately vectorize each medical text contained in the medical knowledge map through a method similar to word embedding, and then concatenate multiple word vectors corresponding to multiple medical texts to obtain the corresponding medical map The third eigenvector.

Further, the first entity word corresponding to each medical knowledge graph is determined, and the third feature vector corresponding to each medical knowledge graph is labeled according to the first entity word. For example, if the first entity word is gastric cancer, it is the The third feature vector is labeled "gastric cancer"; then, the second entity word in the text to be translated is determined, and the second entity tag is determined according to the second entity word; finally, the second entity tag is combined with each third entity The first entity tag corresponding to the feature vector is compared one by one to obtain the first entity tag matching the second entity tag, and the third feature vector corresponding to the matched first entity tag is used as the corresponding to the medical text to be translated Target feature vector.

Exemplarily, it is also possible to add a first entity tag to all medical knowledge maps in the medical field according to the first entity word in each medical knowledge map, that is, to identify the first entity word of each medical knowledge map, and according to the first entity word An entity word adds the first entity tag to each medical knowledge graph; then, the second entity tag corresponding to the text to be translated is determined, that is, the second entity word in the text to be translated is identified, and the second entity word is determined The second entity tag corresponding to the text to be translated; finally, the first entity tag matching the second entity tag is determined, and the medical knowledge graph corresponding to the matched first entity tag is used as the target medical knowledge graph; The medical knowledge graph is vectorized to obtain the target feature vector corresponding to the medical text to be translated.

In this application, the target medical knowledge graph is determined first, and then the target medical knowledge graph is vectorized as an example for description.

Exemplarily, as shown in Figure 2, the medical knowledge graph can be vectorized through the graph conversion network to obtain the target feature vector, where the graph conversion network can be a deepwalk network or a transE network, and so on. This application does not limit the type of the graph conversion network.

It should be understood that the entity word recognition on the medical knowledge graph or the text to be translated can be performed through a neural network or through dictionary matching. The present application does not limit the recognition method of the entity word. Among them, the neural network may be a convolutional neural network, a cyclic neural network, a long and short-term memory network, a bert model, and so on.

104: The medical text translation device splices the first feature vector and the target feature vector to obtain a second feature vector.

Exemplarily, the first feature vector and the target feature vector are horizontally spliced to obtain the second feature vector. For example, if the first feature vector is (0,0,0,...,1) and the target feature vector is (1,0,0...,1), then the first feature vector and the second feature vector are spliced together, The third feature vector is obtained as (0,0,0,...,1,1,0,0...,1).

105: The medical text translation device translates the medical text to be translated according to the second feature vector.

Exemplarily, as shown in FIG. 2, the third feature vector may be input to the decoding network for decoding, and the translation result corresponding to the text to be translated is obtained.

Among them, the use of feature vectors for translation can be achieved through an existing decoding network (Decoder).

Specifically, the decoding network includes multiple stack layers. The third feature vector is first input to the first stack layer of the multiple stack layers to obtain the probability that the third feature vector falls into each word in the dictionary library, and the first feature vector is determined according to the probability of falling into each word. The translation result of a stack layer, that is, the word corresponding to the highest probability is used as the translation result of the first stack layer; then, the translation result of the first stack layer and the third feature vector are input to the second stack layer to continue Translate, translate the first word and the second word; and so on, until the last stack layer outputs the translation result corresponding to the text to be translated.

Exemplarily, as shown in Fig. 2, the first word "I" can be translated through the first stack layer; then, the first word "I" and the second word "affected" can be translated through the second stack layer; By analogy, until the last stack layer translates "I have three types of terminal gastric cancer."

It can be seen that in the embodiment of this application, in the process of translating the medical text to be translated, the medical knowledge map corresponding to the medical text to be translated is fused, so that the second feature vector can be fused with the corresponding to the text to be translated. In order to improve the accuracy of translation, especially the accuracy of translation of medical terminology or medical terminology.

In some possible implementation manners, the medical text to be translated includes Chinese medical text or English medical text, and when the medical text to be translated is a Chinese medical text, the medical knowledge graph is a Chinese medical knowledge graph, In the case that the medical text to be translated is an English medical text, the medical knowledge graph is an English medical knowledge graph.

It should be understood that the language type of the medical text to be translated above should not constitute a limitation to this application. In practical applications, the medical text to be translated may be a medical text in any language, and the medical knowledge graph is a medical knowledge graph corresponding to the language type.

In some possible implementation manners, before performing semantic feature extraction on the medical text to be translated to obtain the first feature vector, the method further includes:

Acquiring the vertical keywords in the medical text to be translated and the third entity words corresponding to the vertical keywords;

Standardize the third entity word according to the vertical keywords to obtain a fourth entity word;

The fourth entity word is used to replace the third entity word in the text to be translated to obtain a new medical text to be translated, and the new medical text to be translated is used for translation.

Exemplarily, word embedding can be performed on each word in the vertical keyword to obtain the word vector corresponding to each word in the vertical keyword; then, according to each word in the vertical keyword Perform semantic feature extraction on the corresponding word vector to obtain the third feature vector used to characterize the semantic feature of the vertical keyword; perform word embedding processing on the third entity word to obtain the correspondence of each word in the third entity word Then, according to the self-attention mechanism, the third feature vector and the word vector corresponding to each word in the third entity word are processed to obtain the target word corresponding to each word in the third entity word Vector, that is, calculate the similarity between the third feature vector and the word vector corresponding to each word in the third entity word, and use the similarity as the weight between the third feature vector and each word , And then, the weight corresponding to each word and the word vector corresponding to the word are subjected to a dot product operation to obtain the target word vector corresponding to each word; according to the target word vector corresponding to each word in the third entity word Semantic feature extraction is used to obtain the fourth feature vector used to characterize the third entity word; finally, according to the fourth feature vector, the probability of falling into each standardized entity word is determined, and the standardized keyword corresponding to the highest probability is used as the fourth feature vector. Entity word.

Among them, the standardized keywords are keywords obtained by pre-standardizing the entity words corresponding to various diseases in the medical field. The relationship between the standardized keywords and the disease is unmistakable, and there is a one-to-one correspondence.

It is understandable that if the vertical keyword or the third entity word is an English word, then the word embedding process is performed on the vertical keyword or the third entity word is each character in the English word. The word embedding process is performed to obtain the character vector corresponding to each character.

It can be seen that, in this embodiment, the entity words are standardized first, even if the entity words in the text to be translated input by the user are wrong, they can be converted into corresponding standardized keywords, because the standardized keywords are clear and correct. Yes, avoiding translation errors caused by user input errors. Moreover, in the process of standardization, a self-attention mechanism is added to consider the matching degree between the third entity word and the vertical keyword, which can amplify the role of the word belonging to the medical field in the third entity word and weaken it. The role of words that do not belong to the medical field in the third entity word can improve the accuracy of standardization.

In one embodiment of the present application, the medical text translation method of the present application can also be applied to the field of smart medicine. For example, doctors can quickly and accurately obtain the translation results through the medical text translation method, so that the translation results can be used for data query or medical history query, which can effectively assist the doctor's diagnosis process and promote the development of medical technology.

Refer to FIG. 4, which is a schematic flowchart of a neural network training method provided by an embodiment of the application. The method includes the following steps:

401: Obtain training text.

Wherein, the training text is the training text of the actual translation result that has been marked, that is, the training text includes the training label.

402: Input the training text to the neural network to obtain a translation result of the training text.

Exemplarily, semantic feature extraction can be performed on the training text through the neural network to obtain the feature vector corresponding to the training text; similarly, the medical knowledge graph corresponding to the training text can be vectorized to obtain the target corresponding to the training sample Feature vector: The target feature vector and the feature vector are spliced together, and the spliced vector is used for translation.

403: Adjust the network parameters of the neural network according to the translation result of the training text and the training label, so as to train the neural network.

That is, the first loss is determined according to the difference between the translation result and the training label; the network parameters of the neural network are updated according to the first loss and the gradient descent method.

Exemplarily, the first loss can be expressed by formula (6):

Wherein, Loss ₁ for the first loss, N for the number of words of training labels, σ _i for the i-th training vector word corresponding to a word label, σ _'i that corresponds to the translation of the i-th word Word vector, dist is the distance operation.

Referring to FIG. 5, FIG. 5 is a schematic structural diagram of a medical text translation device provided by an embodiment of the application. As shown in FIG. 5, a medical text translation device 500 includes a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the processor. The above program includes instructions for performing the following steps:

Obtain the medical text to be translated;

In some possible implementation manners, in terms of obtaining the target feature vector corresponding to the medical text to be translated, the above program is specifically used to execute the instructions of the following steps:

All medical knowledge maps in the medical field are vectorized to obtain the third feature vector corresponding to each medical knowledge map, and according to the first entity word in each medical knowledge map, it is the third feature corresponding to each medical knowledge map The vector adds the first entity label;

Determining a second entity tag corresponding to the text to be translated according to the second entity word in the text to be translated;

Determine a first entity tag that matches the second entity tag, and use a third feature vector corresponding to the matched first entity tag as a target feature vector corresponding to the medical text to be translated.

According to the first entity word in each medical knowledge graph, add the first entity tag to all medical knowledge graphs in the medical field;

Determine a first entity tag that matches the second entity tag, and use a medical knowledge graph corresponding to the matched first entity tag as a target medical knowledge graph;

The target medical knowledge graph is vectorized to obtain a target feature vector corresponding to the medical text to be translated.

In some possible implementation manners, in terms of performing semantic feature extraction on the medical text to be translated to obtain the first feature vector, the above procedure is specifically used to execute the instructions of the following steps:

Performing word embedding processing on each word in the medical text to be translated to obtain a word vector corresponding to each word;

Perform semantic feature extraction according to the word vector corresponding to each word to obtain the first feature vector.

In some possible implementation manners, before the semantic feature extraction is performed according to the word vector corresponding to each word to obtain the first feature vector, the above program is further used to execute the instructions of the following steps:

Determine the target word vector corresponding to each word according to the self-attention mechanism and the word vector corresponding to each word;

In terms of performing semantic feature extraction according to the word vector corresponding to each word to obtain the first feature vector, the above program is specifically used to execute the instructions of the following steps: performing semantic feature extraction according to the target word vector corresponding to each word to obtain the first feature vector The first feature vector.

In some possible implementation manners, in terms of determining the target word vector corresponding to each word according to the self-attention mechanism and the word vector corresponding to each word, the above program is specifically used to execute the instructions of the following steps:

Encoding a word vector corresponding to word A to obtain a key value vector, a query vector, and a value vector corresponding to the word A, where the word A is any word in the medical text to be translated;

Determine the similarity between the query vector corresponding to the word A and the key value vector corresponding to each word, and use the similarity as the weight between the word A and each word;

According to the weight between the word A and each word, the value vector corresponding to each word is weighted to obtain the target word vector corresponding to the word A.

Refer to FIG. 6, which is a block diagram of the functional unit composition of a medical text translation device provided by an embodiment of the present application. The medical text translation device 600 includes: an acquisition unit 601 and a processing unit 602, wherein:

The obtaining unit 601 is used to obtain the medical text to be translated;

The processing unit 602 is configured to perform semantic feature extraction on the medical text to be translated to obtain a first feature vector;

The obtaining unit 601 is further configured to obtain a target feature vector corresponding to the medical text to be translated, and the target feature vector is used to represent a medical knowledge graph corresponding to the medical text to be translated;

The processing unit 602 is further configured to splice the first feature vector and the target feature vector to obtain a second feature vector;

The processing unit 602 is further configured to translate the medical text to be translated according to the second feature vector.

In some possible implementation manners, in terms of acquiring the target feature vector corresponding to the medical text to be translated, the acquiring unit 601 is specifically configured to:

In some possible implementation manners, in terms of performing semantic feature extraction on the medical text to be translated to obtain the first feature vector, the processing unit 602 is specifically configured to:

In some possible implementation manners, before performing semantic feature extraction according to the word vector corresponding to each word to obtain the first feature vector, the processing unit 602 is further configured to: according to the self-attention mechanism and the corresponding word Word vector, determine the target word vector corresponding to each word;

In terms of performing semantic feature extraction according to the word vector corresponding to each word to obtain the first feature vector, the processing unit 602 is specifically configured to: perform semantic feature extraction according to the target word vector corresponding to each word to obtain the first feature vector Feature vector.

In some possible implementation manners, in terms of determining the target word vector corresponding to each word according to the self-attention mechanism and the word vector corresponding to each word, the processing unit 602 is specifically configured to:

Encoding the word vector corresponding to word A to obtain the key value vector, query vector, and value vector corresponding to the word A, where the word A is any word in the medical text to be translated;

The embodiments of the present application also provide a computer (readable) storage medium, the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to realize any medical treatment as described in the above method embodiments. Part or all of the steps of the text translation method.

Optionally, the storage medium involved in this application, such as a computer-readable storage medium, may be non-volatile or volatile.

The embodiments of the present application also provide a computer program product. The computer program product includes a non-transitory computer-readable storage medium storing a computer program. The computer program is operable to cause a computer to execute the method described in the foregoing method embodiment. Part or all of the steps of any medical text translation method.

It should be understood that the medical text translation device in this application may include smart phones (such as Android phones, iOS phones, Windows Phone phones, etc.), tablet computers, handheld computers, notebook computers, mobile Internet Devices (Mobile Internet Devices, abbreviated as: MID) ) Or wearable devices, etc. The above-mentioned medical text translation device is only an example, not an exhaustive list, including but not limited to the above-mentioned medical text translation device. In practical applications, the above-mentioned medical text translation device may also include: smart vehicle-mounted terminals, computer equipment, and so on.

It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations, but those skilled in the art should know that this application is not limited by the described sequence of actions. Because according to this application, some steps can be performed in other order or at the same time. Secondly, those skilled in the art should also know that the embodiments described in the specification are all optional embodiments, and the involved actions and modules are not necessarily required by this application.

In the above-mentioned embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in an embodiment, reference may be made to related descriptions of other embodiments.

In the several embodiments provided in this application, it should be understood that the disclosed device may be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other divisions in actual implementation, for example, multiple units or components may be combined or may be Integrate into another system, or some features can be ignored or not implemented. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may be in electrical or other forms.

The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.

In addition, the functional units in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software program modules.

If the integrated unit is implemented in the form of a software program module and sold or used as an independent product, it can be stored in a computer readable memory. Based on this understanding, the technical solution of the present application essentially or the part that contributes to the existing technology or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a memory, A number of instructions are included to enable a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned memory includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above-mentioned embodiments can be completed by a program instructing relevant hardware. The program can be stored in a computer-readable memory, and the memory can include: a flash disk , Read-only memory (English: Read-Only Memory, abbreviation: ROM), random access device (English: Random Access Memory, abbreviation: RAM), magnetic disk or optical disc, etc.

The embodiments of the application are described in detail above, and specific examples are used in this article to illustrate the principles and implementation of the application. The descriptions of the above embodiments are only used to help understand the methods and core ideas of the application; at the same time, for Those of ordinary skill in the art, based on the idea of the application, will have changes in the specific implementation and the scope of application. In summary, the content of this specification should not be construed as a limitation to the application.

Claims

A medical text translation method, including:

Obtain the medical text to be translated;

Performing semantic feature extraction on the medical text to be translated to obtain a first feature vector;

Acquiring a target feature vector corresponding to the medical text to be translated, where the target feature vector is used to represent a medical knowledge graph corresponding to the medical text to be translated;

Splicing the first feature vector with the target feature vector to obtain a second feature vector;

According to the second feature vector, the medical text to be translated is translated.
The method according to claim 1, wherein said acquiring a target feature vector corresponding to said medical text to be translated comprises:

Vectorize all medical knowledge graphs in the medical field to obtain the third feature vector corresponding to each medical knowledge graph, and according to the first entity word in each medical knowledge graph, set the third feature vector corresponding to each medical knowledge graph. Add the first entity label to the feature vector;

Determining a second entity tag corresponding to the text to be translated according to the second entity word in the text to be translated;

Determine a first entity tag that matches the second entity tag, and use a third feature vector corresponding to the matched first entity tag as a target feature vector corresponding to the medical text to be translated.
The method according to claim 1, wherein said acquiring a target feature vector corresponding to said medical text to be translated comprises:

According to the first entity word in each medical knowledge graph, add the first entity tag to all medical knowledge graphs in the medical field;

Determining a second entity tag corresponding to the text to be translated according to the second entity word in the text to be translated;

Determine a first entity tag that matches the second entity tag, and use a medical knowledge graph corresponding to the matched first entity tag as a target medical knowledge graph;

The target medical knowledge graph is vectorized to obtain a target feature vector corresponding to the medical text to be translated.
The method according to any one of claims 1 to 3, wherein the extracting semantic features of the medical text to be translated to obtain the first feature vector comprises:

Performing word embedding processing on each word in the medical text to be translated to obtain a word vector corresponding to each word;

Perform semantic feature extraction according to the word vector corresponding to each word to obtain the first feature vector.
The method according to claim 4, wherein, before performing semantic feature extraction according to the word vector corresponding to each word to obtain the first feature vector, the method further comprises:

Determine the target word vector corresponding to each word according to the self-attention mechanism and the word vector corresponding to each word;

The performing semantic feature extraction according to the word vector corresponding to each word to obtain the first feature vector includes:

Perform semantic feature extraction according to the target word vector corresponding to each word to obtain the first feature vector.
The method according to claim 5, wherein the determining the target word vector corresponding to each word according to the self-attention mechanism and the word vector corresponding to each word comprises:

Encoding a word vector corresponding to word A to obtain a key value vector, a query vector, and a value vector corresponding to the word A, where the word A is any word in the medical text to be translated;

Determine the similarity between the query vector corresponding to the word A and the key value vector corresponding to each word, and use the similarity as the weight between the word A and each word;

According to the weight between the word A and each word, the value vector corresponding to each word is weighted to obtain the target word vector corresponding to the word A.
The method according to any one of claims 2 or 3, wherein:

The medical text to be translated includes Chinese medical text or English medical text, and when the medical text to be translated is a Chinese medical text, the medical knowledge graph is a Chinese medical knowledge graph, and the medical text to be translated is In the case of an English medical text, the medical knowledge graph is an English medical knowledge graph.
A medical text translation device, including:

The acquiring unit is used to acquire the medical text to be translated;

A processing unit, configured to perform semantic feature extraction on the medical text to be translated to obtain a first feature vector;

The acquiring unit is further configured to acquire a target feature vector corresponding to the medical text to be translated, and the target feature vector is used to represent a medical knowledge graph corresponding to the medical text to be translated;

The processing unit is further configured to splice the first feature vector and the target feature vector to obtain a second feature vector;

The processing unit is further configured to translate the medical text to be translated according to the second feature vector.
A medical text translation device includes a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor to Implement the following methods:

Obtain the medical text to be translated;

Performing semantic feature extraction on the medical text to be translated to obtain a first feature vector;

Acquiring a target feature vector corresponding to the medical text to be translated, where the target feature vector is used to represent a medical knowledge graph corresponding to the medical text to be translated;

Splicing the first feature vector with the target feature vector to obtain a second feature vector;

According to the second feature vector, the medical text to be translated is translated.
The device according to claim 9, wherein when said obtaining the target feature vector corresponding to the medical text to be translated, it specifically implements:

Vectorize all medical knowledge graphs in the medical field to obtain the third feature vector corresponding to each medical knowledge graph, and according to the first entity word in each medical knowledge graph, set the third feature vector corresponding to each medical knowledge graph. Add the first entity label to the feature vector;

Determining a second entity tag corresponding to the text to be translated according to the second entity word in the text to be translated;

Determine a first entity tag that matches the second entity tag, and use a third feature vector corresponding to the matched first entity tag as a target feature vector corresponding to the medical text to be translated.
The device according to claim 9, wherein when said obtaining the target feature vector corresponding to the medical text to be translated, it specifically implements:

According to the first entity word in each medical knowledge graph, add the first entity tag to all medical knowledge graphs in the medical field;

Determining a second entity tag corresponding to the text to be translated according to the second entity word in the text to be translated;

Determine a first entity tag that matches the second entity tag, and use a medical knowledge graph corresponding to the matched first entity tag as a target medical knowledge graph;

The target medical knowledge graph is vectorized to obtain a target feature vector corresponding to the medical text to be translated.
The device according to any one of claims 9-11, wherein when the semantic feature extraction is performed on the medical text to be translated to obtain the first feature vector, it is specifically implemented:

Performing word embedding processing on each word in the medical text to be translated to obtain a word vector corresponding to each word;

Perform semantic feature extraction according to the word vector corresponding to each word to obtain the first feature vector.
The device according to claim 12, wherein, before performing semantic feature extraction according to the word vector corresponding to each word to obtain the first feature vector, the processor is further configured to execute:

Determine the target word vector corresponding to each word according to the self-attention mechanism and the word vector corresponding to each word;

When the semantic feature extraction is performed according to the word vector corresponding to each word, and the first feature vector is obtained, the specific implementation is implemented:

Perform semantic feature extraction according to the target word vector corresponding to each word to obtain the first feature vector.
The device according to claim 13, wherein when the target word vector corresponding to each word is determined according to the self-attention mechanism and the word vector corresponding to each word, the specific realization is achieved:

Encoding a word vector corresponding to word A to obtain a key value vector, a query vector, and a value vector corresponding to the word A, where the word A is any word in the medical text to be translated;

Determine the similarity between the query vector corresponding to the word A and the key value vector corresponding to each word, and use the similarity as the weight between the word A and each word;

According to the weight between the word A and each word, the value vector corresponding to each word is weighted to obtain the target word vector corresponding to the word A.
A computer-readable storage medium in which a computer program is stored, and the computer program is executed by a processor to implement the following method:

Obtain the medical text to be translated;

Performing semantic feature extraction on the medical text to be translated to obtain a first feature vector;

Acquiring a target feature vector corresponding to the medical text to be translated, where the target feature vector is used to represent a medical knowledge graph corresponding to the medical text to be translated;

Splicing the first feature vector with the target feature vector to obtain a second feature vector;

According to the second feature vector, the medical text to be translated is translated.
15. The computer-readable storage medium according to claim 15, wherein said obtaining the target feature vector corresponding to the medical text to be translated is specifically implemented:

Vectorize all medical knowledge graphs in the medical field to obtain the third feature vector corresponding to each medical knowledge graph, and according to the first entity word in each medical knowledge graph, set the third feature vector corresponding to each medical knowledge graph. Add the first entity label to the feature vector;

Determining a second entity tag corresponding to the text to be translated according to the second entity word in the text to be translated;

Determine a first entity tag that matches the second entity tag, and use a third feature vector corresponding to the matched first entity tag as a target feature vector corresponding to the medical text to be translated.
15. The computer-readable storage medium according to claim 15, wherein said obtaining the target feature vector corresponding to the medical text to be translated is specifically implemented:

According to the first entity word in each medical knowledge graph, add the first entity tag to all medical knowledge graphs in the medical field;

Determining a second entity tag corresponding to the text to be translated according to the second entity word in the text to be translated;

Determine a first entity tag that matches the second entity tag, and use a medical knowledge graph corresponding to the matched first entity tag as a target medical knowledge graph;

The target medical knowledge graph is vectorized to obtain a target feature vector corresponding to the medical text to be translated.
18. The computer-readable storage medium according to any one of claims 15-17, wherein when the semantic feature extraction of the medical text to be translated is performed to obtain the first feature vector, the specific implementation is implemented:

Performing word embedding processing on each word in the medical text to be translated to obtain a word vector corresponding to each word;

Perform semantic feature extraction according to the word vector corresponding to each word to obtain the first feature vector.
The computer-readable storage medium according to claim 18, wherein, before the semantic feature extraction is performed according to the word vector corresponding to each word to obtain the first feature vector, the computer program is also used when the computer program is executed by the processor. accomplish:

Determine the target word vector corresponding to each word according to the self-attention mechanism and the word vector corresponding to each word;

When the semantic feature extraction is performed according to the word vector corresponding to each word, and the first feature vector is obtained, the specific implementation is implemented:

Perform semantic feature extraction according to the target word vector corresponding to each word to obtain the first feature vector.
The computer-readable storage medium according to claim 19, wherein when the target word vector corresponding to each word is determined according to the self-attention mechanism and the word vector corresponding to each word, the specific realization is implemented:

Encoding a word vector corresponding to word A to obtain a key value vector, a query vector, and a value vector corresponding to the word A, where the word A is any word in the medical text to be translated;

Determine the similarity between the query vector corresponding to the word A and the key value vector corresponding to each word, and use the similarity as the weight between the word A and each word;

According to the weight between the word A and each word, the value vector corresponding to each word is weighted to obtain the target word vector corresponding to the word A.