Text generation method and system based on large language model

Abstract

The invention provides a text generation method and system based on a large language model, demand information input by a user is responded, the demand information and a cue word template are input into the large language model, a target directory corresponding to the demand information is generated, and the cue word template comprises guide information and an initial directory; text information, corresponding to the target directory, in a knowledge base is obtained, and non-text elements are embedded in the text information; inputting the text information and the target directory into a large language model, and generating a target text consistent with the format of the target directory; and replacing the non-text element in the target text with a non-text object corresponding to the non-text element. The generated target text comprises the non-text elements, multi-modal data query and text content generation are achieved, the generated target text meets the format requirement of the target directory and has high accuracy, and the satisfaction degree of a user is improved.

Description

Technical Field

[0001] The present disclosure relates to the technical fields of artificial intelligence and natural language processing, and in particular to a text generation method and system based on a large language model. Background Art

[0002] Traditional text generation methods often rely on keyword matching. At the same time, the knowledge bases for relevant content retrieval mostly contain single-modal data that only contains text objects. This also limits the ability to generate coherent and in-depth long texts, resulting in the generated text being inaccurate and deviating from user expectations. Summary of the invention

[0003] The technical problem to be solved by the present disclosure is to overcome the defect of low accuracy of text generated in the prior art and to provide a text generation method and system based on a large language model.

[0004] The present invention solves the above technical problems through the following technical solutions:

[0005] A first aspect of the present disclosure provides a text generation method based on a large language model, the text generation method comprising:

[0006] In response to the demand information input by the user, the demand information and the prompt word template are input into the large language model to generate a target directory corresponding to the demand information, wherein the prompt word template includes guide information and an initial directory;

[0007] Acquire text information corresponding to the target directory in a knowledge base, wherein non-text elements are embedded in the text information;

[0008] Inputting the text information and the target directory into a large language model to generate a target text consistent with the format of the target directory;

[0009] The non-text elements in the target text are replaced with non-text objects corresponding to the non-text elements.

[0010] Optionally, the step of acquiring text information corresponding to the target directory in the knowledge base specifically includes:

[0011] Calculating a target similarity between a reference directory and the target directory, wherein the reference directory is any directory in a knowledge base, and the target similarity is determined according to a similarity between a name of the reference directory and a name of the target directory and a similarity between a text content corresponding to the reference directory and the name of the target directory;

[0012] The text information corresponding to the target directory is determined according to all target similarities.

[0013] Optionally, the target similarity is determined by weighting the similarity between the name of the reference directory and the name of the target directory and the similarity between the text content corresponding to the reference directory and the name of the target directory, wherein the weight of the name of the reference directory is greater than the weight of the text content corresponding to the reference directory.

[0014] Optionally, the text generation method also includes constructing a knowledge base, and the step of constructing the knowledge base specifically includes: acquiring text objects and non-text objects in historical data; embedding non-text elements corresponding to the non-text objects into text information of the text objects; and storing the text information in the knowledge base.

[0015] Optionally, the non-text object includes at least one of the following: a picture, a table, and a file.

[0016] Optionally, before the step of inputting the text information and the target directory into the large language model, the method further includes: in response to modification information for the target directory, updating the target directory according to the modification information.

[0017] A second aspect of the present disclosure provides a text generation system based on a large language model, the text generation system comprising:

[0018] A response module, for responding to the demand information input by the user, inputting the demand information and the prompt word template into the large language model, and generating a target directory corresponding to the demand information, wherein the prompt word template includes guide information and an initial directory;

[0019] An acquisition module, used for acquiring text information corresponding to the target directory in a knowledge base, wherein the text information is embedded with non-text elements;

[0020] A generating module, used for inputting the text information and the target directory into a large language model to generate a target text consistent with the format of the target directory;

[0021] A replacement module is used to replace the non-text elements in the target text with non-text objects corresponding to the non-text elements.

[0022] Optionally, the acquisition module is specifically used to calculate the target similarity between the reference directory and the target directory, wherein the reference directory is any directory in the knowledge base, and the target similarity is determined based on the similarity between the name of the reference directory and the name of the target directory and the similarity between the text content corresponding to the reference directory and the name of the target directory; and the text information corresponding to the target directory is determined based on all the target similarities.

[0023] Optionally, the acquisition module also includes a calculation module, which is specifically used to weightedly determine the similarity between the name of the reference directory and the name of the target directory and the similarity between the text content corresponding to the reference directory and the name of the target directory, wherein the weight of the name of the reference directory is greater than the weight of the text content corresponding to the reference directory.

[0024] Optionally, the text generation system also includes a construction module, which is specifically used to obtain text objects and non-text objects in historical data; embed non-text elements corresponding to the non-text objects into text information of the text objects; and store the text information in a knowledge base.

[0025] Optionally, the generating module is further configured to update the target directory according to the modification information when modification information of the target directory is received.

[0026] A third aspect of the present disclosure provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and used to run on the processor, wherein when the processor executes the computer program, the text generation method based on the large language model as described in the first aspect is implemented.

[0027] Optionally, the electronic device further includes a dry cell battery.

[0028] A fourth aspect of the present disclosure is a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for generating text based on a large language model as described in the first aspect is implemented.

[0029] A fifth aspect of the present disclosure provides a computer program product, including a computer program, which, when executed by a processor, implements the text generation method based on a large language model as described in the first aspect.

[0030] On the basis of being in accordance with the common sense in the art, the above-mentioned optional conditions can be arbitrarily combined to obtain the preferred embodiments of the present disclosure.

[0031] The positive progressive effect of the present disclosure lies in: automatically generating a target directory based on user needs, then identifying text information containing non-text elements in the knowledge base based on the target directory, and combining the target directory and the text information using a large language model to generate a target text including non-text elements. Compared with traditional single-modal text content generation technology, the present disclosure realizes the query and text content generation of multi-modal data, and the generated target text not only meets the format requirements of the target directory but also has a high degree of accuracy, thereby improving user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1A flow chart of a text generation method based on a large language model provided in Embodiment 1 of the present disclosure;

[0033] Figure 2 A module schematic diagram of a text generation system based on a large language model provided in Embodiment 2 of the present disclosure;

[0034] Figure 3 A schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present disclosure. DETAILED DESCRIPTION

[0035] The present disclosure is further described below by way of examples, but the present disclosure is not limited to the scope of the examples.

[0036] Prefixes such as "first" and "second" are used in the embodiments of the present disclosure only to distinguish different description objects, and have no limiting effect on the position, order, priority, quantity or content of the described objects. The use of prefixes such as ordinal numbers to distinguish description objects in the embodiments of the present disclosure does not constitute a limitation on the described objects. For the statement of the described objects, please refer to the description in the context of the embodiments, and no unnecessary limitation should be constituted due to the use of such prefixes. In addition, in the description of the present embodiment, unless otherwise specified, the meaning of "plurality" is two or more.

[0037] In the embodiments of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0038] Example 1

[0039] This embodiment provides a text generation method based on a large language model, such as Figure 1 As shown, the text generation method includes:

[0040] S1. In response to demand information input by a user, the demand information and a prompt word template are input into a large language model to generate a target directory corresponding to the demand information, wherein the prompt word template includes guide information and an initial directory.

[0041] In a specific implementation, the demand information input by the user can be a specific text description or an uploaded file, and the prompt word template is a preset template, including guide information and an initial directory, wherein the guide information is used to help the large language model understand the nature of the task, and the initial directory is used to provide a basic directory structure framework. The large language model can generate a target directory that meets the user's needs based on the preset prompt word template and the demand information input by the user.

[0042] In a specific example, when generating a target directory for scientific research papers, the required information input by the user may be subject keywords or abstract information; when generating a target directory for bidding documents, the required information input by the user may be the technical specifications of the uploaded bidding documents.

[0043] In another specific example, the guiding information in the initial prompt word template is "Please generate a framework for a bid, including the following parts: project overview, overview background, project plan, staffing, technical advantages, project plan, budget and quotation, as well as risk assessment and management measures", and the initial directory is "1. Project Overview: 1.1 Project Overview, 1.2 Project Description, 1.3 Overall Project Objectives; 2. Overall Framework Design: 2.1 Overall Design Ideas, 2.2 Application Function Framework, 2.3 Overall Technical Architecture".

[0044] S2. Acquire text information corresponding to the target directory in a knowledge base, wherein non-text elements are embedded in the text information.

[0045] In a specific implementation, a query is performed in the knowledge base according to the target directory to retrieve text information related to the target directory. The returned text information includes not only text elements but also non-text elements related to the text elements, which are embedded in the text information.

[0046] In a specific example, the image element is represented by the figure special character tag and the image title information and embedded in the text information. Specifically, the obtained text information is "The intelligent coal blending system consists of five parts: data acquisition system, database system, intelligent optimization coal blending system, coke quality prediction system and fault diagnosis system.\n![figure](1-figure-1.jpg\" Figure 1-1 Application functional architecture\")", where "\n![figure](1-figure-1.jpg\" Figure 1-1 Application Capability Architecture\"" represents an embedded non-text element.

[0047] S3. Input the text information and the target directory into a large language model to generate a target text in a format consistent with the target directory.

[0048] In the specific implementation, the text information and the generated target directory are input into the large language model. The large language model parses the structure of the target directory and integrates the text information into the corresponding directory part. During the integration process, the large language model also performs semantic filling on the text information to ensure that the content of each part is semantically coherent and consistent with the subject corresponding to the directory of that part. The target text generated based on the large language model not only contains the integrated text information, but also conforms to the logical hierarchy and layout format of the target directory.

[0049] S4. Replace the non-text elements in the target text with non-text objects corresponding to the non-text elements.

[0050] In a specific implementation, the non-text elements in the target text are parsed, and the corresponding non-text objects are obtained from the knowledge base, database or external resource library according to the parsing results, and then the non-text elements in the target text are replaced with the actual non-text objects.

[0051] In a specific example, the target text includes a non-text element of the image type "[figure](1-figure-1.jpg\" Figure 1-1 Application Functional Architecture\")", according to the file name "1-figure-1.jpg" and description " Figure 1-1 Application Function Architecture", retrieves the corresponding image from the image database, and then replaces the image elements in the target text with the actually obtained image. The replaced target text will no longer contain non-text elements of the image type, but directly displays the corresponding image.

[0052] In an optional embodiment, the step of obtaining text information corresponding to the target directory in the knowledge base specifically includes: calculating the target similarity between the reference directory and the target directory, wherein the reference directory is any directory in the knowledge base, and the target similarity is determined based on the similarity between the name of the reference directory and the name of the target directory and the similarity between the text content corresponding to the reference directory and the name of the target directory; determining the text information corresponding to the target directory based on all the target similarities.

[0053] In the specific implementation, the similarity between the name of the reference directory and the name of the target directory, as well as the similarity between the text content corresponding to the reference directory name and the name of the target directory are calculated respectively, and the directory name similarity and the text content similarity are combined to obtain the target similarity, so as to retrieve the text information that best matches the target directory in the knowledge base, making it highly relevant to the target directory in both subject and content.

[0054] In a specific example, a string similarity algorithm is used to calculate the similarity between the name of the reference directory and the name of the target directory, and a TF-IDF algorithm or a cosine similarity algorithm is used to calculate the similarity between the text content corresponding to the reference directory name and the name of the target directory.

[0055] In an optional embodiment, the target similarity is determined by weighting the similarity between the name of the reference directory and the name of the target directory and the similarity between the text content corresponding to the reference directory and the name of the target directory, wherein the weight of the name of the reference directory is greater than the weight of the text content corresponding to the reference directory.

[0056] In a specific implementation, the similarity between the name of the reference directory and the name of the target directory and the similarity between the text content corresponding to the reference directory and the name of the target directory are combined in a weighted manner, wherein the similarity of the reference directory name is given a greater weight to ensure that the similarity of the directory names dominates the final result.

[0057] In a specific example, the similarity score between the name of the reference directory and the name of the target directory is expressed as score(directory), the similarity score between the text content corresponding to the reference directory and the name of the target directory is expressed as score(content), and the target similarity is expressed as score. Specifically, score = γ*score(directory)+(1-γ)*score(content), wherein γ is the weight of the similarity of the reference directory name, 1-γ is the weight of the similarity of the text content corresponding to the reference directory, and γ is greater than 0.5 during the weighted calculation process.

[0058] In an optional embodiment, the text generation method also includes constructing a knowledge base, and the step of constructing the knowledge base specifically includes: obtaining text objects and non-text objects in historical data; embedding non-text elements corresponding to the non-text objects into text information of the text objects; and storing the text information in the knowledge base.

[0059] In the specific implementation, since the historical data not only involves text objects but also some non-text related content, the non-text elements corresponding to the acquired non-text objects are embedded in the text information of the text objects when constructing the knowledge base, thereby realizing the index construction of multimodal heterogeneous elements and achieving efficient retrieval when the target directory acquires text information from the knowledge base.

[0060] In a specific example, the image element is represented by the figure special character tag and the image title information and embedded in the text information of the text object. Specifically, the text information after the image element is embedded in the knowledge base is "The intelligent coal blending system consists of five parts: data acquisition system, database system, intelligent optimization coal blending system, coke quality prediction system and fault diagnosis system.\n![figure](1-figure-1.jpg\" Figure 1-1 Application functional architecture\")".

[0061] In another specific example, the historical data also includes relatively independent information, which describes some factual statements and is stored independently in a docx file. Specifically, the text information embedded with the file element is expressed as "The classification of the coal quality evaluation system is shown in the following table: \n! [docx] (1-docx-0.docx)".

[0062] In another specific example, when building a knowledge base based on historical data, it is necessary to retain the original directory structure information in the historical data file to facilitate similarity matching of the target directory name.

[0063] In an optional implementation, the non-text object includes at least one of the following: a picture, a table, and a file.

[0064] In an optional implementation manner, before the step of inputting the text information and the target directory into the large language model, the step further includes: in response to modification information for the target directory, updating the target directory according to the modification information.

[0065] In a specific implementation, the user is allowed to view and modify the generated target directory, and the generated target directory is updated in a timely manner based on the modification information provided by the user.

[0066] In a specific example, a user interface is provided that can intuitively display the target directory. The user can add new directory names through the interface, delete directory names that are no longer needed, and modify the names in existing target directories. The target directory is updated based on the user's modification information to generate a target directory that better meets the user's needs.

[0067] Example 2

[0068] Corresponding to the text generation method embodiment based on a large language model in the aforementioned embodiment 1, the present disclosure also provides an embodiment of a text generation system based on a large language model, such as Figure 2 As shown, the text generation system includes:

[0069] A response module 101 is used to respond to the demand information input by the user, input the demand information and the prompt word template into the large language model, and generate a target directory corresponding to the demand information, wherein the prompt word template includes guide information and an initial directory;

[0070] The acquisition module 102 is used to acquire text information corresponding to the target directory in the knowledge base, wherein the text information is embedded with non-text elements;

[0071] A generating module 103, configured to input the text information and the target directory into a large language model to generate a target text consistent with the format of the target directory;

[0072] The replacement module 104 is used to replace the non-text elements in the target text with non-text objects corresponding to the non-text elements.

[0073] In an optional embodiment, the acquisition module is specifically used to calculate the target similarity between the reference directory and the target directory, wherein the reference directory is any directory in the knowledge base, and the target similarity is determined based on the similarity between the name of the reference directory and the name of the target directory and the similarity between the text content corresponding to the reference directory and the name of the target directory; and the text information corresponding to the target directory is determined based on all the target similarities.

[0074] In an optional embodiment, the acquisition module also includes a calculation module, which is specifically used to weightedly determine the similarity between the name of the reference directory and the name of the target directory and the similarity between the text content corresponding to the reference directory and the name of the target directory, wherein the weight of the name of the reference directory is greater than the weight of the text content corresponding to the reference directory.

[0075] In an optional embodiment, the text generation system also includes a construction module, which is specifically used to obtain text objects and non-text objects in historical data; embed non-text elements corresponding to the non-text objects into the text information of the text objects; and store the text information in a knowledge base.

[0076] In an optional implementation manner, the generating module is further configured to update the target directory according to the modification information when modification information of the target directory is received.

[0077] As for the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The system embodiment described above is only illustrative, wherein the units described as separate components may or may not be physically separated, and the components 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 modules may be selected according to actual needs to achieve the purpose of the disclosed solution.

[0078] Example 3

[0079] Figure 3This is a structural schematic diagram of an electronic device shown in an exemplary embodiment of the present disclosure, wherein the electronic device includes a memory, a processor, and a computer program stored in the memory and used to run on the processor, and when the processor executes the computer program, the text generation method based on the large language model of Example 1 is implemented. Figure 3 The electronic device 90 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0080] like Figure 3 As shown, the electronic device 90 may be in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 90 may include, but are not limited to: at least one processor 91, at least one memory 92, and a bus 93 connecting different system components (including the memory 92 and the processor 91).

[0081] The bus 93 includes a data bus, an address bus, and a control bus.

[0082] The memory 92 may include a volatile memory, such as a random access memory (RAM) 921 and / or a cache memory 922 , and may further include a read-only memory (ROM) 923 .

[0083] The memory 92 may also include a program tool 925 (or utility) having a set (at least one) of program modules 924, such program modules 924 including but not limited to: an operating system, one or more application programs, other program modules and program data, each of which or some combination may include an implementation of a network environment.

[0084] The processor 91 executes various functional applications and data processing by running the computer programs stored in the memory 92, such as the text generation method based on the large language model of the first embodiment of the present disclosure.

[0085] The electronic device 90 may also communicate with one or more external devices 94 (e.g., keyboards, pointing devices, etc.). Such communication may be performed via an input / output (I / O) interface 95. Furthermore, the electronic device 90 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 96. As shown, the network adapter 96 communicates with other modules of the electronic device 90 via a bus 93. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 90, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.

[0086] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.

[0087] Example 4

[0088] The embodiment of the present disclosure also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the text generation method based on the large language model of embodiment 1 is implemented.

[0089] The readable storage medium may include but is not limited to: a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device or any suitable combination of the above.

[0090] Example 5

[0091] The embodiments of the present disclosure also provide a computer program product, including a computer program, which implements the text generation method based on the large language model of embodiment 1 when executed by a processor.

[0092] Among them, the program code for executing the computer program product of the present disclosure can be written in any combination of one or more programming languages, and the program code can be executed completely on the user device, partially on the user device, as an independent software package, partially on the user device and partially on a remote device, or completely on the remote device.

[0093] Although the specific embodiments of the present disclosure are described above, those skilled in the art should understand that this is only an example, and the protection scope of the present disclosure is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present disclosure, but these changes and modifications all fall within the protection scope of the present disclosure.

Claims

1. A text generation method based on a large language model, characterized in that: The text generation method comprises: In response to the demand information input by the user, the demand information and the prompt word template are input into the large language model to generate a target directory corresponding to the demand information, wherein the prompt word template includes guide information and an initial directory; Acquire text information corresponding to the target directory in a knowledge base, wherein non-text elements are embedded in the text information; Inputting the text information and the target directory into a large language model to generate a target text consistent with the format of the target directory; The non-text elements in the target text are replaced with non-text objects corresponding to the non-text elements.

2. The text generation method according to claim 1, characterized in that: The step of obtaining text information corresponding to the target directory in the knowledge base specifically includes: Calculating a target similarity between a reference directory and the target directory, wherein the reference directory is any directory in a knowledge base, and the target similarity is determined according to a similarity between a name of the reference directory and a name of the target directory and a similarity between a text content corresponding to the reference directory and the name of the target directory; The text information corresponding to the target directory is determined according to all target similarities.

3. The text generation method according to claim 2, characterized in that: The target similarity is determined specifically by weighting the similarity between the name of the reference directory and the name of the target directory and the similarity between the text content corresponding to the reference directory and the name of the target directory, wherein the weight of the name of the reference directory is greater than the weight of the text content corresponding to the reference directory.

4. The text generation method according to claim 1, characterized in that: The text generation method further includes constructing a knowledge base, and the step of constructing a knowledge base specifically includes: Get text objects and non-text objects in historical data; Embedding a non-text element corresponding to the non-text object into the text information of the text object; The text information is stored in a knowledge base.

5. The text generation method according to claim 4, characterized in that: The non-text object includes at least one of the following: a picture, a table, and a file.

6. The text generation method according to any one of claims 1 to 5, characterized in that: Before the step of inputting the text information and the target directory into the large language model, the step further includes: In response to the modification information for the target directory, the target directory is updated according to the modification information.

7. A text generation system based on a large language model, characterized in that: The text generation system comprises: A response module, for responding to the demand information input by the user, inputting the demand information and the prompt word template into the large language model, and generating a target directory corresponding to the demand information, wherein the prompt word template includes guide information and an initial directory; An acquisition module, used for acquiring text information corresponding to the target directory in a knowledge base, wherein the text information is embedded with non-text elements; A generating module, used for inputting the text information and the target directory into a large language model to generate a target text consistent with the format of the target directory; A replacement module is used to replace the non-text elements in the target text with non-text objects corresponding to the non-text elements.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and used to run on the processor, characterized in that: When the processor executes the computer program, the text generation method according to any one of claims 1 to 6 is implemented.

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

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

Images