Large model application method, device and system based on semantic space conversion

Abstract

The invention provides a large model application method, device and system based on semantic space conversion, belongs to the technical field of large models, and is used for solving the problem that in the related technology, the cost of obtaining user data assets by a large model operator is low. In the training process or the using process, interaction between the user and the large model can be achieved through the semantic conversion model, the user works in the first semantic space, the large model works in the second semantic space, and the first semantic space and the second semantic space correspond to each other, so that the large model has good performance when facing a specific field and a specific task; according to the method and the system, the isolation between the user original data and the large model is realized, the user can work in a familiar semantic space, the threshold of obtaining the private data assets of the user by a large model operator is improved, and the security of the private data assets of the user is guaranteed.

Description

Technical Field

[0001] The present application relates to the technical field of large models, and in particular to a large model application method, device and system based on semantic space conversion. Background Art

[0002] With the rise of deep learning technology, especially the innovation of neural network architecture and the continuous improvement of computing power, large model technology has made great progress and has been widely used. The deployment, training, operation and maintenance of large models and prompt word engineering are extremely costly tasks. Generally speaking, only a few companies and personnel are capable of doing this, and most companies and personnel use large models trained by others.

[0003] When the pre-trained general-purpose big model cannot meet the user's needs for specific scenarios or specific tasks, the general-purpose big model is generally trained again using the professional data required for the specific scenario or specific task, so that the trained big model can perform well in the specific scenario or specific task. However, when using professional data to train the general-purpose big model, the user's professional data and the user's consideration of the prompt word engineering based on it are equivalent to being fully disclosed to the operator of the big model. Although there are relevant rules and regulations of the operator and relevant laws and regulations of the country to protect, there is no technical threshold, resulting in the big model operator's acquisition of user private professional data and corresponding prompt word engineering and other data assets at a very low cost. Summary of the invention

[0004] The present application provides a large model application method, device and system based on semantic space conversion, which is conducive to improving the cost of the operator of the large model to obtain user data assets.

[0005] In a first aspect, the present application provides a large model application method based on semantic space conversion. The method comprises: When users train a general large model, Obtain original professional data used by users to train general large models; Using a preconfigured semantic conversion model to convert the original professional data into processed professional data; Using the processing professional data to train the general large model to obtain a professional large model; When the user uses the professional large model, Acquire first interaction data sent by the user to the large model; Converting the first interaction data into second interaction data using the semantic conversion model; Sending the second interaction data to the professional big model so that the professional big model returns second feedback data; Using the semantic conversion model, converting the second feedback data into first feedback data, wherein the first feedback data is used to be presented to a user; The semantic conversion model is used to realize the semantic conversion of data between the first semantic space and the second semantic space. The interpretation of the data before and after the semantic conversion is basically the same but the expression is different. The original professional data, the first interaction data and the first feedback data are all in the first semantic space, and the processed professional data, the second interaction data and the second feedback data are all in the second semantic space.

[0006] By adopting the above technical solution and utilizing the semantic conversion model, the user's original professional data and the first interactive data generated during the interaction are not directly disclosed to the big model operator. The big model operator can directly obtain the original training data and the first interactive data after the data is converted from the first semantic space to the second semantic space. This technically increases the cost threshold for the big model operator to obtain private data assets such as the user's original professional data, which is beneficial to ensuring the security of user data assets.

[0007] Furthermore, the first semantic space and the second semantic space are both one of natural language expression, audio information expression, image information expression, coding information expression, programming language expression, and mathematical language expression.

[0008] Furthermore, when the first semantic space is a natural language expression, If the second semantic space is expressed in a natural language, the semantic conversion model is used to realize the conversion of data between one natural language and another natural language; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between natural language and audio information; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between natural language text and an image containing natural language text; If the second semantic space is expressed as coded information, the semantic conversion model is used to realize the conversion of data between natural language text and coded information describing the natural language text; If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between natural language text and the programming language describing the natural language text; If the second semantic space is expressed in mathematical language, the semantic conversion model is used to realize the conversion of data between natural language text and mathematical language describing the natural language text.

[0009] Furthermore, when the first semantic space is an audio information expression, If the second semantic space is a natural language expression, the semantic conversion model is used to realize the conversion of data between audio information and natural language text describing the audio information; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between one audio information and another audio information; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between audio information and image information describing the audio information; If the second semantic space is expressed as coded information, the semantic conversion model is used to realize the conversion of data between audio information and coded information describing the audio information; If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between audio information and the programming language describing the audio information; If the second semantic space is expressed in a mathematical language, the semantic conversion model is used to realize the conversion of data between audio information and the mathematical language describing the audio information.

[0010] Furthermore, when the first semantic space is an image information expression, If the second semantic space is a natural language expression, the semantic conversion model is used to realize the conversion of data between image information and natural language text describing the image information; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between image information and audio information describing the image information; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between one image information and another image information; If the second semantic space is expressed as coded information, the semantic conversion model is used to realize the conversion of data between image information and coded information describing the image information; If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between image information and the programming language describing the image information; If the second semantic space is expressed in a mathematical language, the semantic conversion model is used to realize the conversion of data between image information and the mathematical language describing the image information.

[0011] Furthermore, when the first semantic space is an encoding information expression, If the second semantic space is expressed in natural language, the semantic conversion model is used to realize the conversion of data between the encoded information and the natural language describing the encoded information; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between the coded information and the audio information describing the coded information; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between the encoding information and the image information describing the encoding information; If the second semantic space is expressed as coded information, the semantic conversion model is used to realize the conversion of data between one coded information and another coded information; If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between the encoding information and the programming language describing the encoding information; If the second semantic space is expressed in a mathematical language, the semantic conversion model is used to realize the conversion of data between the encoded information and the mathematical language describing the encoded information.

[0012] Furthermore, when the first semantic space is expressed in a programming language, If the second semantic space is expressed in natural language, the semantic conversion model is used to realize the conversion of data between the programming language and the natural language describing the programming language; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between the programming language and the audio information describing the programming language; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between the program language and the image information describing the program language; If the second semantic space is a coded information expression, the semantic conversion model is used to realize the conversion of data between program language features and coding features; If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between one programming language and another programming language; If the second semantic space is expressed in a mathematical language, the semantic conversion model is used to realize the conversion of data between the programming language and the mathematical language describing the programming language.

[0013] Furthermore, when the first semantic space is expressed in mathematical language, If the second semantic space is expressed in natural language, the semantic conversion model is used to realize the conversion of data between mathematical language and natural language describing mathematical language; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between mathematical language and audio information describing the mathematical language; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between mathematical language and image information describing the mathematical language; If the second semantic space is expressed as coded information, the semantic conversion model is used to realize the conversion of data between mathematical language and coded information describing the mathematical language; If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between a mathematical language and a programming language describing the mathematical language; If the second semantic space is expressed in a mathematical language, the semantic conversion model is used to realize the conversion of data between one mathematical language and another mathematical language.

[0014] Furthermore, when both the first semantic space and the second semantic space are expressed in natural language, the semantic conversion model realizes conversion between one natural language and another natural language based on natural language mutual translation technology.

[0015] Furthermore, when the first semantic space is expressed in natural language and the second semantic space is expressed in audio information, the semantic conversion model realizes conversion between natural language and audio information based on speech-to-text conversion technology.

[0016] Furthermore, when the first semantic space is expressed in natural language and the second semantic space is expressed in image information, the semantic conversion model realizes the conversion between natural language and image information based on natural language text to image technology and image text recognition technology.

[0017] Furthermore, when the first semantic space is expressed in natural language and the second semantic space is expressed in coded information, the semantic conversion model realizes conversion between natural language and coded information based on language coding conversion technology.

[0018] Furthermore, when the first semantic space is expressed in natural language and the second semantic space is expressed in programming language, the semantic conversion model realizes conversion between natural language and programming language based on language-program conversion technology.

[0019] Furthermore, when the first semantic space is expressed in natural language and the second semantic space is expressed in mathematical language, the semantic conversion model realizes conversion between natural language and mathematical language based on language character mapping technology.

[0020] Furthermore, when the first semantic space is an audio information expression and the second semantic space is a natural language expression, the semantic conversion model realizes the conversion between the audio information and the natural language based on the speech-to-text conversion technology.

[0021] Furthermore, when both the first semantic space and the second semantic space are audio information expressions, the semantic conversion model realizes conversion between one audio information and another audio information based on audio conversion technology, and the converted audio information carries the content or waveform of the audio information.

[0022] Furthermore, when the first semantic space is used to express audio information and the second semantic space is used to express image information, the semantic conversion model realizes conversion between audio information and image information based on audio-image conversion technology, and the converted image information contains the content or waveform of the audio information.

[0023] Furthermore, when the first semantic space is for audio information expression and the second semantic space is for coded information expression, the semantic conversion model includes a mapping relationship between audio features and coded characters.

[0024] Furthermore, when the first semantic space is an audio information expression and the second semantic space is a programming language expression, the semantic conversion model includes a mapping relationship between audio features and programming languages.

[0025] Furthermore, when the first semantic space is an audio information expression and the second semantic space is a mathematical language expression, the semantic conversion model includes a mapping relationship between audio features and mathematical characters.

[0026] Furthermore, when the first semantic space is an image information expression and the second semantic space is a natural language expression, the semantic conversion model realizes the conversion between image information and natural language based on image text recognition technology and text-to-image technology.

[0027] Furthermore, when the first semantic space is for image information expression and the second semantic space is for audio information expression, the semantic conversion model includes a mapping relationship between image features and audio features.

[0028] Furthermore, when both the first semantic space and the second semantic space are expressions of image information, the semantic conversion model realizes conversion between image information based on image conversion technology, and the converted image information contains image features of the image information before conversion.

[0029] Furthermore, when the first semantic space is used to express image information and the second semantic space is used to express coding information, the semantic conversion model realizes conversion between image information and coding information based on image coding conversion technology, and the semantic conversion model includes a mapping relationship between image features and coding features.

[0030] Furthermore, when the first semantic space is expressed by image information and the second semantic space is expressed by a programming language, the semantic conversion model includes a mapping relationship between image features and programming language features.

[0031] Furthermore, when the first semantic space is an image information expression and the second semantic space is a mathematical language expression, the semantic conversion model includes a mapping relationship between image features and mathematical characters.

[0032] Furthermore, when the first semantic space is a coded information expression and the second semantic space is a natural language expression, the semantic conversion model includes a mapping relationship between coded features and natural language.

[0033] Furthermore, when the first semantic space is for coding information expression and the second semantic space is for audio information expression, the semantic conversion model includes a mapping relationship between coding features and audio features.

[0034] Furthermore, when the first semantic space is for coding information expression and the second semantic space is for image information expression, the semantic conversion model includes a mapping relationship between coding features and image features.

[0035] Furthermore, when both the first semantic space and the second semantic space are expressed by coded information, the semantic conversion model is implemented based on a coding conversion technology, and the converted coded information includes coding features of the coded information before the conversion.

[0036] Furthermore, when the first semantic space is a coding information expression and the second semantic space is a programming language expression, the semantic conversion model includes a mapping relationship between coding features and programming language features.

[0037] Furthermore, when the first semantic space is a coded information expression and the second semantic space is a mathematical language expression, the semantic conversion model includes a mapping relationship between coded features and mathematical characters.

[0038] Furthermore, when the first semantic space is expressed in a programming language and the second semantic space is expressed in a natural language, the semantic conversion model includes a mapping relationship between programming language features and natural language features.

[0039] Furthermore, when the first semantic space is expressed in a programming language and the second semantic space is expressed in audio information, the semantic conversion model includes a mapping relationship between programming language features and audio features.

[0040] Furthermore, when the first semantic space is expressed in a programming language and the second semantic space is expressed in image information, the semantic conversion model includes a mapping relationship between programming language features and image features.

[0041] Furthermore, when the first semantic space is expressed in a programming language and the second semantic space is expressed in coding information, the semantic conversion model includes a mapping relationship between programming language features and coding features.

[0042] Furthermore, when both the first semantic space and the second semantic space are expressed in a programming language, the semantic conversion model is implemented based on a programming language conversion technology, and the converted programming language includes the features of the programming language before the conversion.

[0043] Furthermore, when the first semantic space is expressed in a programming language and the second semantic space is expressed in a mathematical language, the semantic conversion model includes a mapping relationship between programming language features and mathematical characters.

[0044] Furthermore, when the first semantic space is expressed in mathematical language and the second semantic space is expressed in natural language, the semantic conversion model includes a mapping relationship between mathematical characters and natural language features.

[0045] Furthermore, when the first semantic space is expressed in mathematical language and the second semantic space is expressed in audio information, the semantic conversion model includes a mapping relationship between mathematical characters and audio features.

[0046] Furthermore, when the first semantic space is expressed in mathematical language and the second semantic space is expressed in image information, the semantic conversion model includes a mapping relationship between mathematical characters and image features.

[0047] Furthermore, when the first semantic space is expressed in mathematical language and the second semantic space is expressed in coded information, the semantic conversion model includes a mapping relationship between mathematical characters and coded features.

[0048] Furthermore, when the first semantic space is expressed in a mathematical language and the second semantic space is expressed in a programming language, the semantic conversion model includes a mapping relationship between mathematical characters and programming language features.

[0049] Furthermore, when both the first semantic space and the second semantic space are expressed in mathematical language, the semantic conversion model is implemented based on mathematical language conversion technology, and the converted mathematical language includes mathematical language features before the conversion.

[0050] Furthermore, the semantic conversion model is also used to provide semantic conversion annotations during semantic conversion, so that the data feature distortion rate before the semantic conversion is lower than the distortion rate threshold.

[0051] In a second aspect, the present application provides a large model application device based on semantic space conversion, and the device includes an expression conversion model as described in any one of the first aspects above.

[0052] In a third aspect, the present application provides a large model application system based on semantic space conversion, and the system applies any one of the methods described in the first aspect above.

[0053] In summary, this application at least has the following beneficial effects: A big model application method, device and system based on semantic space conversion are provided. By setting a semantic conversion model on the user side that is necessary for the user to interact with the big model, the direct disclosure of the user's original data assets to the big model is avoided, thereby technically raising the threshold for the big model operator to obtain the user's private data assets, which is conducive to protecting the user's private data assets.

[0054] It should be understood that the contents described in the Summary of the Invention are not intended to limit the key or important features of the embodiments of the present application, nor are they intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The above and other features, advantages and aspects of the embodiments of the present application will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, wherein: Figure 1 A schematic diagram showing an exemplary operating environment in which embodiments of the present application can be implemented; Figure 2 A flowchart of a method for applying a large model based on semantic space conversion when a user trains a general large model in an embodiment of the present application is shown; Figure 3 A flowchart of a large model application method based on semantic space conversion when a user uses a professional large model in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0057] In addition, the term "and / or" in this article is only a description of the association relationship between the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0058] The present application provides a big model application method, device and system based on semantic space conversion, which avoids the direct disclosure of user original data assets to the big model by setting a semantic conversion model on the user side that is necessary for the user to interact with the big model, thereby raising the threshold for the big model operator to obtain user private data assets from a technical perspective, which is conducive to protecting user private data assets.

[0059] Figure 1 A schematic diagram of an exemplary operating environment in which embodiments of the present application can be implemented is shown.

[0060] Reference Figure 1 The operating environment includes a user terminal 110 and a server 120. There are generally multiple user terminals 110, and only one is shown in the figure. The user terminal 110 is for users to use, and the server 120 is deployed with a pre-trained general large model. The server 120 is used by the large model operator. The user terminal 110 and the server 120 are connected in communication to facilitate the interaction between the user and the large model.

[0061] The user terminal 110 may be a mobile terminal device, a computer, etc., and generally identifies the user through a unique identity such as an account, rather than using a device hardware identity to identify the user. The server 120 may be deployed in the cloud or locally, and may be a centralized server or a distributed server. In order to ensure the reliability of the server 120, a backup may also be deployed for the server 120.

[0062] In a first aspect, the present application provides a large model application method based on semantic space transformation.

[0063] Figure 2 A flowchart of a large model application method based on semantic space conversion when a user trains a general large model in an embodiment of the present application is shown.

[0064] Reference Figure 2 , when the user trains a general large model, the method specifically includes the following steps: S210: Obtain original professional data used by the user to train the general large model.

[0065] S220: Convert the original professional data into processed professional data using a preconfigured semantic conversion model.

[0066] S230: Using the processing professional data to train the general large model to obtain a professional large model.

[0067] Original professional data is the experience data generated by users in the long-term work process facing specific scenarios or specific tasks. Its form and format may be arbitrary, but it all reflects the user's experience in a specific field. It is itself a data asset that reflects the results of the user's labor and even contains the user's potential intellectual property considerations. Therefore, if the original professional data is directly leaked, it itself is a manifestation of being unfavorable to the protection of the user's intangible assets.

[0068] Since the professional big model is trained with the processed professional data provided by the user, the professional big model can perform well in the specific scenarios or specific tasks for the processed professional data, and the original professional data is the same as the specific scenarios or specific tasks for the processed professional data. Therefore, the professional big model can basically reach the level of ability to perform in specific scenarios or specific tasks after being trained with the original professional data, and the data disclosed to the big model operator is the processed professional data rather than the original professional data. Even if the big model operator can obtain the original professional data by parsing, extracting or translating the processed professional data, the big model operator cannot directly obtain the user's original professional data by copying or exporting, thereby raising the threshold for the big model operator to obtain the user's private data assets from a technical perspective.

[0069] Taking into account the consistency between training and use, and the fact that the user's private data in specific scenarios or tasks will be disclosed during use, the semantic conversion model is also needed when using the aforementioned trained professional large model. Figure 3 A flowchart of a large model application method based on semantic space conversion when a user uses a professional large model in an embodiment of the present application is shown.

[0070] Reference Figure 3 When the user uses the professional large model, the method specifically includes the following steps: S310: Acquire first interaction data sent by the user to the large model.

[0071] S320: Convert the first interaction data into second interaction data using the semantic conversion model.

[0072] S330: Send the second interaction data to the professional big model, so that the professional big model returns second feedback data.

[0073] S340: Using the semantic conversion model, convert the second feedback data into first feedback data, where the first feedback data is used to be presented to the user.

[0074] In this way, the big model always works in the second semantic space, which can ensure the consistency of big model training and use, and ensure the application effect of the big model for specific scenarios or specific tasks. Since the interaction data received by the big model and the feedback data returned are always in the second semantic space, it can avoid users disclosing private data assets to the big model operator during long-term use, and users only need to submit interaction data and receive feedback from the big model in the first semantic space they are familiar with, which is conducive to users using professional big models more conveniently and quickly.

[0075] Based on the description of a large model application method based on semantic space conversion in the embodiment of the present application, the key improvement of the present application lies in the semantic conversion model, and the semantic conversion model is specifically introduced below.

[0076] The meaning of the semantic conversion model defined in the embodiments of the present application is to realize the semantic conversion of data between a first semantic space and a second semantic space, where the interpretation of the data before and after the conversion is the same or basically the same, but the expression is different, that is, the information carried by the data before and after the conversion is the same or basically the same, but the format, expression, etc. of the data before and after the conversion are distinguished.

[0077] Specifically, the first semantic space and the second semantic space are both natural language expressions, audio information expressions, image information expressions, coded information expressions, program language expressions, and mathematical language expressions. Of course, the first semantic space and the second semantic space can also be other forms or formats such as video information expressions, which cannot be exhaustively listed here.

[0078] Some specific examples of the semantic transformation model are introduced below.

[0079] In one example, when the first semantic space is a natural language expression, If the second semantic space is expressed in natural language, the semantic conversion model is used to realize the conversion of data between one natural language and another natural language; in this case, the semantic conversion model realizes the conversion between one natural language and another natural language based on the natural language mutual translation technology; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between natural language and audio information; in this case, the semantic conversion model realizes the conversion between natural language and audio information based on speech-to-text conversion technology; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between natural language text and images containing natural language text; in this case, the semantic conversion model realizes the conversion between natural language and image information based on natural language text to image technology and image text recognition technology; If the second semantic space is expressed as coded information, the semantic conversion model is used to realize the conversion of data between natural language text and coded information describing the natural language text; in this case, the semantic conversion model realizes the conversion between natural language and coded information based on language coding conversion technology; If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between natural language text and the programming language describing the natural language text; in this case, the semantic conversion model realizes the conversion between natural language and programming language based on language-program conversion technology; If the second semantic space is expressed in mathematical language, the semantic conversion model is used to realize the conversion of data between natural language text and mathematical language describing the natural language text. At this time, the semantic conversion model realizes the conversion between natural language and mathematical language based on language character mapping technology.

[0080] In one example, when the first semantic space is an audio information expression, If the second semantic space is a natural language expression, the semantic conversion model is used to realize the conversion of data between audio information and natural language text describing the audio information. In this case, the semantic conversion model realizes the conversion between audio information and natural language based on speech-to-text conversion technology; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between one audio information and another audio information. At this time, the semantic conversion model realizes the conversion between one audio information and another audio information based on the audio conversion technology, and the converted audio information carries the content or waveform of the audio information; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between audio information and image information describing the audio information. In this case, the semantic conversion model realizes the conversion between audio information and image information based on audio-image conversion technology, and the converted image information contains the content or waveform of the audio information; If the second semantic space is expressed by coded information, the semantic conversion model is used to realize the conversion of data between audio information and coded information describing the audio information. In this case, the semantic conversion model includes a mapping relationship between audio features and coded characters. If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between audio information and a programming language describing the audio information. In this case, the semantic conversion model includes a mapping relationship between audio features and programming languages; If the second semantic space is expressed in mathematical language, the semantic conversion model is used to realize the conversion of data between audio information and the mathematical language describing the audio information. In this case, the semantic conversion model includes a mapping relationship between audio features and mathematical characters.

[0081] In one example, when the first semantic space is an image information expression, If the second semantic space is a natural language expression, the semantic conversion model is used to realize the conversion of data between image information and natural language text describing the image information. In this case, the semantic conversion model realizes the conversion between image information and natural language based on image text recognition technology and text-to-image technology; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between image information and audio information describing the image information. In this case, the semantic conversion model includes a mapping relationship between image features and audio features. If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between one image information and another image information. In this case, the semantic conversion model realizes the conversion between image information based on image conversion technology, and the converted image information contains the image features of the image information before the conversion. If the second semantic space is expressed as coded information, the semantic conversion model is used to realize the conversion of data between image information and coded information describing the image information. In this case, the semantic conversion model realizes the conversion between image information and coded information based on image coding conversion technology, and the semantic conversion model includes a mapping relationship between image features and coded features; If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between image information and a programming language describing the image information. In this case, the semantic conversion model includes a mapping relationship between image features and programming language features; If the second semantic space is expressed in mathematical language, the semantic conversion model is used to realize the conversion of data between image information and the mathematical language describing the image information. In this case, the semantic conversion model includes the mapping relationship between image features and mathematical characters.

[0082] In one example, when the first semantic space is an encoded information expression, If the second semantic space is expressed in natural language, the semantic conversion model is used to realize the conversion of data between the encoded information and the natural language describing the encoded information. In this case, the semantic conversion model includes a mapping relationship between the encoded features and the natural language; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between the coding information and the audio information describing the coding information. In this case, the semantic conversion model includes a mapping relationship between the coding features and the audio features; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between the coding information and the image information describing the coding information. In this case, the semantic conversion model includes a mapping relationship between coding features and image features; If the second semantic space is expressed as coded information, the semantic conversion model is used to realize the conversion of data between one coded information and another coded information. In this case, the semantic conversion model is implemented based on the coding conversion technology, and the converted coded information includes the coding features of the coded information before the conversion. If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between the encoding information and the programming language describing the encoding information. In this case, the semantic conversion model includes a mapping relationship between encoding features and programming language features; If the second semantic space is expressed in mathematical language, the semantic conversion model is used to realize the conversion of data between the encoding information and the mathematical language describing the encoding information. In this case, the semantic conversion model includes the mapping relationship between the encoding features and the mathematical characters.

[0083] In one example, when the first semantic space is expressed in a programming language, If the second semantic space is expressed in natural language, the semantic conversion model is used to realize the conversion of data between the programming language and the natural language describing the programming language. In this case, the semantic conversion model includes a mapping relationship between programming language features and natural language features; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between the programming language and the audio information describing the programming language. In this case, the semantic conversion model includes a mapping relationship between programming language features and audio features; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between the program language and the image information describing the program language. In this case, the semantic conversion model includes a mapping relationship between program language features and image features; If the second semantic space is a coded information expression, the semantic conversion model is used to realize the conversion of data between one programming language and another programming language. In this case, the semantic conversion model includes a mapping relationship between programming language features and coding features; If the second semantic space is a coded information expression, the semantic conversion model is used to realize the conversion of data between program language features and coding features. In this case, the semantic conversion model includes a mapping relationship between program language features and coding features. If the second semantic space is expressed in a mathematical language, the semantic conversion model is used to realize the conversion of data between the programming language and the mathematical language describing the programming language. At this time, the semantic conversion model is implemented based on the programming language conversion technology, and the converted programming language includes the programming language features before the conversion.

[0084] In one example, when the first semantic space is expressed in mathematical language, If the second semantic space is expressed in natural language, the semantic conversion model is used to realize the conversion of data between mathematical language and natural language describing mathematical language. In this case, the semantic conversion model includes a mapping relationship between mathematical characters and natural language features; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between mathematical language and audio information describing the mathematical language. In this case, the semantic conversion model includes a mapping relationship between mathematical characters and audio features; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between mathematical language and image information describing the mathematical language. In this case, the semantic conversion model includes a mapping relationship between mathematical characters and image features; If the second semantic space is expressed by coded information, the semantic conversion model is used to realize the conversion of data between mathematical language and coded information describing the mathematical language. In this case, the semantic conversion model includes a mapping relationship between mathematical characters and coded features; If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between a mathematical language and a programming language describing the mathematical language. In this case, the semantic conversion model includes a mapping relationship between mathematical characters and programming language features; If the second semantic space is expressed in mathematical language, the semantic conversion model is used to realize the conversion of data between one mathematical language and another mathematical language. At this time, the semantic conversion model is implemented based on mathematical language conversion technology, and the converted mathematical language includes the mathematical language features before the conversion.

[0085] In the above content, in order to ensure that there is no distortion before and after the semantic conversion, semantic conversion annotations can also be configured for the semantic conversion model. For example, in the specific example of converting between two natural language texts, if there are multiple translation methods for translating from one natural language to another natural language, the natural language annotations configured in the semantic conversion model can be used to determine the only translation method to ensure that there is no distortion before and after the conversion; in addition, in order to ensure that the big model can fully understand the data originally in the first semantic space through the data converted to the second semantic space, all rules such as the semantic conversion model and the corresponding semantic conversion annotations can be synchronized to the big model. For example, when converting natural language text to a coded language, in order to avoid the big model having difficulty in understanding the converted coded language, the conversion rules in the semantic conversion model can be sent to the big model together, which is equivalent to the user and the big model agreeing on a "code word". In this way, the big model can fully understand what the user wants to express, but the user's content will not be directly included in the database of the big model and can be directly and conveniently obtained by the operation and maintenance personnel.

[0086] A semantic conversion model is configured on the user side. Whether in training or use, the user's interaction with the big model will go through the semantic conversion model. The user works in the first semantic space, and the big model works in the second semantic space. The first semantic space and the second semantic space complement each other, thereby achieving good performance of the big model when facing specific fields and specific tasks, and isolating the user's original data from the big model. It also enables the user to work in the semantic space he is familiar with, which is conducive to raising the threshold for the big model operator to obtain the user's private data assets, and thus is conducive to ensuring the security of the user's private data assets.

[0087] Regarding the scope division of specific semantic space conversion schemes, in addition to the method of dividing according to the semantic space before and after the conversion, which was first reflected in the application documents, it can also be classified according to whether the conversion model tools required for the specific semantic space conversion are available, combined with the distortion rate control of the information before and after the conversion, in order to achieve the purpose of low-cost semantic space conversion; based on the above classification ideas, the semantic space conversion schemes can be specifically divided into the following three categories: the first category is the existing conversion model tools, and the conversion distortion rate of the existing conversion model tools can be within the required preset distortion rate. This type of conversion model tool can be directly applied as a semantic space conversion model, such as existing translation software, existing Chinese character mapping software, etc.; the second category is also an existing conversion model tool, but the conversion distortion rate of the existing conversion model tool exceeds the required preset distortion rate. At this time, considering the information before and after the conversion, the conversion distortion rate of the existing conversion model tool exceeds the required preset distortion rate. In order to ensure the substantial consistency of the content, some additional rules can be configured for the existing conversion model tools for low cost considerations to reduce the distortion rate of the existing conversion model tools. For example, the existing translation software may have translations between different languages, and some language content may have multiple translation methods. At this time, a unique translation method can be agreed upon to avoid the "one-to-many" situation that may exist in the existing conversion model tools. The impact on the substantial content before and after the information conversion is avoided, and the distortion rate of the information during the conversion process is controlled; the third category is customized conversion model tools. If there is no existing conversion model tool that can be applied to the specific scenario of semantic space conversion, a customized conversion model tool is required. At this time, the consideration of customizing the conversion model tool is completely based on the "feature mapping" method, such as Chinese characters and character mapping, etc. In this way, the conversion of information between different semantic spaces can be achieved at a low cost, and the information before and after the conversion is basically not distorted.

[0088] Whether fully utilizing the existing conversion model tool, utilizing the conversion model tool adjusted from the existing conversion model tool, or utilizing a customized conversion model tool as the semantic space conversion model, the total performance of the cost and distortion rate of the semantic space conversion model may be comprehensively considered.

[0089] Here, if there are multiple implementation schemes for the semantic space conversion model, and the distortion rate is required not to be lower than the distortion rate threshold, the distortion rate of each implementation scheme can be verified through interactive experiments, and it can be determined whether the distortion rate of each implementation scheme is lower than the distortion rate threshold, and the implementation scheme with a distortion rate lower than the distortion rate threshold is determined as an optional implementation scheme, and then the cost of each implementation scheme is determined according to the workload of each implementation scheme, and the result of subtracting the distortion rate from the distortion rate threshold is calculated as the threshold gap, and the preset multiple of the ratio of the threshold gap to the cost is calculated as the preference of the optional implementation scheme, and the implementation scheme with the highest preference is selected as the preferred implementation scheme, and the preferred implementation scheme can be applied to the semantic space conversion model.

[0090] In the above content, the cost of the implementation plan can be determined by utilizing the collected descriptive information of multiple dimensions of the implementation plan, such as the number of additional rules, the content volume of the additional rules, the number of mapping rules, etc. A cost component is preset for each type of descriptive information in each dimension, and then all cost components of the implementation plan can be determined. The result of calculating all cost components is the cost of the implementation plan.

[0091] It should be understood that the key to the solution sought to be protected in this application is to raise the technical threshold for large-scale model operation and maintenance personnel to obtain user private data assets, that is, it is impossible to directly obtain user original data by copying or exporting. In some cases, it may have a certain encryption effect, but the original data does not necessarily need to be encrypted. On the contrary, if the large-scale model operator spends a certain amount of time and cost, it can always obtain the user's original data. Of course, since the amount of user private data is generally large, the time and cost for the large-scale model operator to obtain this part of the original data is correspondingly high. In this way, the threshold for the large-scale model operator to obtain user private data is raised from a technical level, which is conducive to ensuring the security of user private data.

[0092] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the embodiments of the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present application.

[0093] In a second aspect, the present application provides a large model application device based on semantic space conversion, wherein the device comprises the semantic conversion model described in the first aspect above.

[0094] In a third aspect, the present application provides a large model application system based on semantic space conversion, and the system applies the large model application method based on semantic space conversion as described in the first aspect above.

[0095] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described device and system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0096] In summary, this application at least has the following beneficial effects: A big model application method, device and system based on semantic space conversion are provided. By setting a semantic conversion model on the user side that is necessary for the user to interact with the big model, the direct disclosure of the user's original data assets to the big model is avoided, thereby technically raising the threshold for the big model operator to obtain the user's private data assets, which is conducive to protecting the user's private data assets.

[0097] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the aforementioned disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.

Claims

1. A large model application method based on semantic space conversion, characterized in that: include: When users train a general large model, Obtain original professional data used by users to train general large models; Using a preconfigured semantic conversion model to convert the original professional data into processed professional data; Using the processing professional data to train the general large model to obtain a professional large model; When the user uses the professional large model, Acquire first interaction data sent by the user to the large model; Converting the first interaction data into second interaction data using the semantic conversion model; Sending the second interaction data to the professional big model so that the professional big model returns second feedback data; Using the semantic conversion model, converting the second feedback data into first feedback data, wherein the first feedback data is used to be presented to a user; The semantic conversion model is used to realize the semantic conversion of data between the first semantic space and the second semantic space. The interpretation of the data before and after the semantic conversion is basically the same but the expression is different. The original professional data, the first interaction data and the first feedback data are all in the first semantic space, and the processed professional data, the second interaction data and the second feedback data are all in the second semantic space.

2. The large model application method based on semantic space conversion according to claim 1 is characterized in that: The first semantic space and the second semantic space are both one of natural language expression, audio information expression, image information expression, coding information expression, programming language expression, and mathematical language expression.

3. The large model application method based on semantic space conversion according to claim 2 is characterized in that: When the first semantic space is a natural language expression, If the second semantic space is expressed in a natural language, the semantic conversion model is used to realize the conversion of data between one natural language and another natural language; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between natural language and audio information; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between natural language text and an image containing natural language text; If the second semantic space is expressed as coded information, the semantic conversion model is used to realize the conversion of data between natural language text and coded information describing the natural language text; If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between natural language text and the programming language describing the natural language text; If the second semantic space is expressed in mathematical language, the semantic conversion model is used to realize the conversion of data between natural language text and mathematical language describing the natural language text.

4. The large model application method based on semantic space conversion according to claim 2 is characterized in that: When the first semantic space is an audio information expression, If the second semantic space is a natural language expression, the semantic conversion model is used to realize the conversion of data between audio information and natural language text describing the audio information; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between one audio information and another audio information; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between audio information and image information describing the audio information; If the second semantic space is expressed as coded information, the semantic conversion model is used to realize the conversion of data between audio information and coded information describing the audio information; If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between audio information and the programming language describing the audio information; If the second semantic space is expressed in a mathematical language, the semantic conversion model is used to realize the conversion of data between audio information and the mathematical language describing the audio information.

5. The large model application method based on semantic space conversion according to claim 2 is characterized in that: When the first semantic space is an image information expression, If the second semantic space is a natural language expression, the semantic conversion model is used to realize the conversion of data between image information and natural language text describing the image information; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between image information and audio information describing the image information; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between one image information and another image information; If the second semantic space is expressed as coded information, the semantic conversion model is used to realize the conversion of data between image information and coded information describing the image information; If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between image information and the programming language describing the image information; If the second semantic space is expressed in a mathematical language, the semantic conversion model is used to realize the conversion of data between image information and the mathematical language describing the image information.

6. The large model application method based on semantic space conversion according to claim 2 is characterized in that: When the first semantic space is used to express the coded information, If the second semantic space is expressed in natural language, the semantic conversion model is used to realize the conversion of data between the encoded information and the natural language describing the encoded information; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between the coded information and the audio information describing the coded information; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between the encoding information and the image information describing the encoding information; If the second semantic space is expressed as coded information, the semantic conversion model is used to realize the conversion of data between one coded information and another coded information; If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between the encoding information and the programming language describing the encoding information; If the second semantic space is expressed in a mathematical language, the semantic conversion model is used to realize the conversion of data between the encoded information and the mathematical language describing the encoded information.

7. The large model application method based on semantic space conversion according to claim 2 is characterized in that: When the first semantic space is expressed in a programming language, If the second semantic space is expressed in natural language, the semantic conversion model is used to realize the conversion of data between the programming language and the natural language describing the programming language; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between the program language and the audio information describing the program language; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between the program language and the image information describing the program language; If the second semantic space is a coded information expression, the semantic conversion model is used to realize the conversion of data between program language features and coding features; If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between one programming language and another programming language; If the second semantic space is expressed in a mathematical language, the semantic conversion model is used to realize the conversion of data between the programming language and the mathematical language describing the programming language.

8. The large model application method based on semantic space conversion according to claim 2 is characterized in that: When the first semantic space is expressed in mathematical language, If the second semantic space is expressed in natural language, the semantic conversion model is used to realize the conversion of data between mathematical language and natural language describing mathematical language; If the second semantic space is an audio information expression, the semantic conversion model is used to realize the conversion of data between mathematical language and audio information describing the mathematical language; If the second semantic space is an image information expression, the semantic conversion model is used to realize the conversion of data between mathematical language and image information describing the mathematical language; If the second semantic space is expressed as coded information, the semantic conversion model is used to realize the conversion of data between mathematical language and coded information describing the mathematical language; If the second semantic space is expressed in a programming language, the semantic conversion model is used to realize the conversion of data between a mathematical language and a programming language describing the mathematical language; If the second semantic space is expressed in a mathematical language, the semantic conversion model is used to realize the conversion of data between one mathematical language and another mathematical language.

9. The large model application method based on semantic space conversion according to claim 3 is characterized in that: When both the first semantic space and the second semantic space are expressed in natural language, the semantic conversion model realizes conversion between one natural language and another natural language based on natural language mutual translation technology.

10. The large model application method based on semantic space conversion according to claim 3 is characterized in that: When the first semantic space is expressed in natural language and the second semantic space is expressed in audio information, the semantic conversion model realizes conversion between natural language and audio information based on speech-to-text conversion technology.

11. The large model application method based on semantic space conversion according to claim 3 is characterized in that: When the first semantic space is expressed in natural language and the second semantic space is expressed in image information, the semantic conversion model realizes conversion between natural language and image information based on natural language text to image technology and image text recognition technology.

12. The large model application method based on semantic space conversion according to claim 3 is characterized in that: When the first semantic space is expressed in natural language and the second semantic space is expressed in coded information, the semantic conversion model realizes conversion between natural language and coded information based on language coding conversion technology.

13. The large model application method based on semantic space conversion according to claim 3 is characterized in that: When the first semantic space is expressed in natural language and the second semantic space is expressed in programming language, the semantic conversion model realizes conversion between natural language and programming language based on language-program conversion technology.

14. The large model application method based on semantic space conversion according to claim 3 is characterized in that: When the first semantic space is expressed in natural language and the second semantic space is expressed in mathematical language, the semantic conversion model realizes conversion between natural language and mathematical language based on language character mapping technology.

15. The large model application method based on semantic space conversion according to claim 4 is characterized in that: When the first semantic space is an audio information expression and the second semantic space is a natural language expression, the semantic conversion model realizes the conversion between the audio information and the natural language based on the speech-to-text conversion technology.

16. The large model application method based on semantic space conversion according to claim 4 is characterized in that: When both the first semantic space and the second semantic space are audio information expressions, the semantic conversion model realizes conversion between one audio information and another audio information based on audio conversion technology, and the converted audio information carries the content or waveform of the audio information.

17. The large model application method based on semantic space conversion according to claim 4 is characterized in that: When the first semantic space is for audio information expression and the second semantic space is for image information expression, the semantic conversion model realizes conversion between audio information and image information based on audio-image conversion technology, and the converted image information contains the content or waveform of the audio information.

18. The large model application method based on semantic space conversion according to claim 4 is characterized in that: When the first semantic space is for audio information expression and the second semantic space is for coded information expression, the semantic conversion model includes a mapping relationship between audio features and coded characters.

19. The large model application method based on semantic space conversion according to claim 4 is characterized in that: When the first semantic space is an audio information expression and the second semantic space is a programming language expression, the semantic conversion model includes a mapping relationship between audio features and programming languages.

20. The large model application method based on semantic space conversion according to claim 4 is characterized in that: When the first semantic space is an audio information expression and the second semantic space is a mathematical language expression, the semantic conversion model includes a mapping relationship between audio features and mathematical characters.

21. The large model application method based on semantic space conversion according to claim 5 is characterized in that: When the first semantic space is an image information expression and the second semantic space is a natural language expression, the semantic conversion model realizes the conversion between image information and natural language based on image text recognition technology and text-to-image technology.

22. The large model application method based on semantic space conversion according to claim 5 is characterized in that: When the first semantic space is for image information expression and the second semantic space is for audio information expression, the semantic conversion model includes a mapping relationship between image features and audio features.

23. The large model application method based on semantic space conversion according to claim 5 is characterized in that: When both the first semantic space and the second semantic space are image information expressions, the semantic conversion model realizes conversion between image information based on image conversion technology, and the converted image information contains image features of the image information before conversion.

24. The large model application method based on semantic space conversion according to claim 5 is characterized in that: When the first semantic space is for image information expression and the second semantic space is for coding information expression, the semantic conversion model realizes conversion between image information and coding information based on image coding conversion technology, and the semantic conversion model includes a mapping relationship between image features and coding features.

25. The large model application method based on semantic space conversion according to claim 5 is characterized in that: When the first semantic space is an image information expression and the second semantic space is a program language expression, the semantic conversion model includes a mapping relationship between image features and program language features.

26. The large model application method based on semantic space conversion according to claim 5 is characterized in that: When the first semantic space is an image information expression and the second semantic space is a mathematical language expression, the semantic conversion model includes a mapping relationship between image features and mathematical characters.

27. The large model application method based on semantic space conversion according to claim 6 is characterized in that: When the first semantic space is a coded information expression and the second semantic space is a natural language expression, the semantic conversion model includes a mapping relationship between coded features and natural language.

28. The large model application method based on semantic space conversion according to claim 6 is characterized in that: When the first semantic space is for coding information expression and the second semantic space is for audio information expression, the semantic conversion model includes a mapping relationship between coding features and audio features.

29. The large model application method based on semantic space conversion according to claim 6 is characterized in that: When the first semantic space is for coding information expression and the second semantic space is for image information expression, the semantic conversion model includes a mapping relationship between coding features and image features.

30. The large model application method based on semantic space conversion according to claim 6 is characterized in that: When both the first semantic space and the second semantic space are expressed in coded information, the semantic conversion model is implemented based on a coding conversion technology, and the converted coded information includes coding features of the coded information before the conversion.

31. The large model application method based on semantic space conversion according to claim 6 is characterized in that: When the first semantic space is a coding information expression and the second semantic space is a programming language expression, the semantic conversion model includes a mapping relationship between coding features and programming language features.

32. The large model application method based on semantic space conversion according to claim 6 is characterized in that: When the first semantic space is a coded information expression and the second semantic space is a mathematical language expression, the semantic conversion model includes a mapping relationship between coded features and mathematical characters.

33. The large model application method based on semantic space conversion according to claim 7 is characterized in that: When the first semantic space is expressed in a programming language and the second semantic space is expressed in a natural language, the semantic conversion model includes a mapping relationship between programming language features and natural language features.

34. The large model application method based on semantic space conversion according to claim 7 is characterized in that: When the first semantic space is expressed in a programming language and the second semantic space is expressed in audio information, the semantic conversion model includes a mapping relationship between programming language features and audio features.

35. The large model application method based on semantic space conversion according to claim 7 is characterized in that: When the first semantic space is expressed in a programming language and the second semantic space is expressed in image information, the semantic conversion model includes a mapping relationship between programming language features and image features.

36. The large model application method based on semantic space conversion according to claim 7 is characterized in that: When the first semantic space is expressed in a programming language and the second semantic space is expressed in coding information, the semantic conversion model includes a mapping relationship between programming language features and coding features.

37. The large model application method based on semantic space conversion according to claim 7 is characterized in that: When both the first semantic space and the second semantic space are expressed in programming language, the semantic conversion model is implemented based on programming language conversion technology, and the converted programming language includes the features of the programming language before conversion.

38. The large model application method based on semantic space conversion according to claim 7 is characterized in that: When the first semantic space is expressed in a programming language and the second semantic space is expressed in a mathematical language, the semantic conversion model includes a mapping relationship between programming language features and mathematical characters.

39. The large model application method based on semantic space conversion according to claim 8 is characterized in that: When the first semantic space is expressed in mathematical language and the second semantic space is expressed in natural language, the semantic conversion model includes a mapping relationship between mathematical characters and natural language features.

40. The large model application method based on semantic space conversion according to claim 7 is characterized in that: When the first semantic space is expressed in mathematical language and the second semantic space is expressed in audio information, the semantic conversion model includes a mapping relationship between mathematical characters and audio features.

41. The large model application method based on semantic space conversion according to claim 7 is characterized in that: When the first semantic space is expressed in mathematical language and the second semantic space is expressed in image information, the semantic conversion model includes a mapping relationship between mathematical characters and image features.

42. The large model application method based on semantic space conversion according to claim 7 is characterized in that: When the first semantic space is expressed in mathematical language and the second semantic space is expressed in coded information, the semantic conversion model includes a mapping relationship between mathematical characters and coded features.

43. The large model application method based on semantic space conversion according to claim 7 is characterized in that: When the first semantic space is expressed in a mathematical language and the second semantic space is expressed in a programming language, the semantic conversion model includes a mapping relationship between mathematical characters and programming language features.

44. The large model application method based on semantic space conversion according to claim 7 is characterized in that: When both the first semantic space and the second semantic space are expressed in mathematical language, the semantic conversion model is implemented based on mathematical language conversion technology, and the converted mathematical language includes the mathematical language features before the conversion.

45. The large model application method based on semantic space conversion according to any one of claims 2 to 44, characterized in that: The semantic conversion model is also used to provide semantic conversion annotations during semantic conversion, so that the data feature distortion rate before semantic conversion is lower than the distortion rate threshold.

46. ​​A large model application device based on semantic space conversion, characterized in that: The device includes a semantic conversion model as described in any one of claims 1-45.

47. A large model application system based on semantic space conversion, characterized in that: The system applies the method described in any one of claims 1-45.

Images