Generation method and device of continuous writing content and storage medium

Abstract

The invention relates to a continuous writing content generation method and device and a storage medium. The method comprises the following steps: extracting key information from received input information; generating continuous writing content corresponding to the key information by utilizing a retrieval enhancement generation mode; and outputting the continuous writing content corresponding to the key information. According to the method, the information amount of the key information can be increased through retrieval by utilizing a retrieval enhancement generation mode, and then the continuous writing content is generated according to the key information after the information amount is increased. Therefore, after the key information is retrieved and enhanced, the information to be expressed by the key information is more specific and more comprehensive, and when the continuous writing content is generated, the reference information is not limited to the key information, so that the generated continuous writing content better fits the intention of a user, and the usability of the continuous writing content is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device and storage medium for generating continued content. Background Art

[0002] When users are processing text, they sometimes encounter creative bottlenecks. To address this problem, R&D personnel have integrated the text continuation function into the large language model. Currently, when users use the text continuation function, they need to copy the edited text content into the dialog box corresponding to the text continuation function. The large language model generates the continuation text corresponding to the text content based on the text content in the dialog box for users to use.

[0003] However, the usability of the continuation text generated by the existing large language model is poor. The reason for this problem is that the existing large language model needs to first analyze the context information of the text content, and then generate the continuation text based on the context information of the text content. In this way, the usability of the continuation text depends on the understanding of the context information of the text content by the large language model. However, users who use the text continuation function are not only professionals in the field of natural language processing. The text content often has many problems such as poor text quality and short text length, which makes it impossible for the large language model to accurately understand the context of the text content, and then the generated continuation content has poor usability. For example, if the text quality of the text content is poor, such as: the expression is vague and not specific enough, the large language model will not be able to accurately analyze the context information of the text content, and the generated continuation text will be difficult to fit the user's intention, and then the use rate of the continuation content will be low. For another example, if the text content is short, the large language model will not have enough context information to refer to. When generating the continuation text, it is very easy to have the problem of inappropriate continuation text. Inappropriate continuation content will inevitably affect the user's use of the continuation content. Summary of the invention

[0004] The present application provides a method, device and storage medium for generating continued content to solve the problem of poor usability of continued text generated by a large language model.

[0005] In view of the above technical problems, the technical solution of this application is solved by the following embodiments:

[0006] An embodiment of the present application provides a method for generating continuation content, comprising: extracting key information from received input information; generating continuation content corresponding to the key information using a retrieval enhanced generation method; and outputting the continuation content corresponding to the key information.

[0007] Among them, extracting key information from the received input information includes: extracting target content from the input information and identifying the guidance type corresponding to the target content by performing multimodal analysis on the input information; extracting key content from the target content according to the guidance type corresponding to the target content; generating the key information according to the guidance type corresponding to the target content and the key content extracted from the target content.

[0008] Wherein, the identifying of the guidance type corresponding to the target content includes: when the target content is the first line of content, determining that the guidance type corresponding to the target content is the first title type; when the target content is content in a paragraph, or when the target content is the beginning of a paragraph and the content of the previous paragraph of the target content meets the preset title condition, determining that the guidance type corresponding to the target content is the main body type; when the target content is the beginning of a paragraph and the content of the previous paragraph of the target content does not meet the preset title condition, predicting the guidance type corresponding to the target content based on the target content.

[0009] Among them, predicting the guidance type corresponding to the target content according to the target content includes: using a preset type prediction model to predict the type of content of the next paragraph of the target content according to the target content; when it is predicted that the content of the next paragraph of the target content is a title, determining the guidance type corresponding to the target content is a second title type; when it is predicted that the content of the next paragraph of the target content is a main body, determining the guidance type corresponding to the target content is a main body type; when it is predicted that the content of the next paragraph of the target content is other viewpoints, determining the guidance type corresponding to the target content is other viewpoint types.

[0010] Among them, generating the key information according to the guidance type corresponding to the target content and the key content extracted from the target content includes: determining the guidance type corresponding to the target content and the key content extracted from the target content as the key information; or, extracting attribute information from the input information; using the attribute information, performing input enhancement processing on the key content, and determining the attribute information, the key content after the input enhancement processing, and the guidance type corresponding to the target content as the key information.

[0011] Among them, the use of the attribute information to perform input enhancement processing on the key content includes: obtaining memory information corresponding to the attribute information; wherein, the memory information includes: keywords corresponding to preset function plug-ins; when the key content matches the keywords corresponding to the function plug-in, calling the function plug-in, causing the function plug-in to execute the function task corresponding to the function plug-in according to the key content, obtaining the execution result returned by the function task, and using the execution result and the key content as the key content after input enhancement processing.

[0012] Among them, the use of retrieval enhanced generation method to generate the continuation content corresponding to the key information includes: querying auxiliary content matching the key information; generating a large model guidance instruction based on the key information; using a preset large language model to generate the continuation content corresponding to the key information based on the key information, the auxiliary content and the large model guidance instruction.

[0013] Among them, the query for auxiliary content matching the key information includes: starting from receiving the input information, every time the content volume of the input information increases by a preset second content volume threshold, querying the paradigm template matching the key information; starting from receiving the input information, every time the key information is extracted, querying the query information matching the key information.

[0014] Among them, the query information that matches the key information includes at least one of the following methods: according to the attribute information included in the key information, querying the memory information corresponding to the attribute information; according to the key content included in the key information, querying the pre-stored information that matches the key content in a preset knowledge retrieval system; according to the key content included in the key information, querying the historical continuation content that matches the key content in the cached historical continuation content.

[0015] The generating of the large model instruction according to the key information includes: generating the large model guidance instruction according to the guidance type corresponding to the target content included in the key information.

[0016] Among them, the large model guidance instructions are generated according to the guidance type corresponding to the target content included in the key information, including: when the guidance type corresponding to the target content is the first title type, a first title guidance instruction is generated; when the guidance type corresponding to the target content is the second title type, a second title guidance instruction is generated; when the guidance type corresponding to the target content is a table type or a body type, a body content guidance instruction is generated; when the guidance type corresponding to the target content is other viewpoint types, other viewpoint guidance instructions are generated.

[0017] Among them, the use of a preset large language model to generate the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guidance instructions includes: using a large language model corresponding to the attribute information in the key information to generate the continuation content corresponding to the key information according to the key content in the key information, the auxiliary content and the large model guidance instructions.

[0018] Among them, the use of a preset large language model to generate the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guidance instructions includes: using a preset large language model to generate the continuation content carrying a relevance label according to the key information, the auxiliary content and the large model guidance instructions; wherein the relevance label is the relevance between the continuation content and the key content in the key information.

[0019] Among them, the use of a preset large language model to generate the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guiding instructions includes: sending an interface request to the large language model; wherein the key information, the auxiliary content and the large model guiding instructions are carried in the interface request; the large language model generates the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guiding instructions carried in the interface request; reading the output content of the large language model; and when the output content meets the preset content generation conditions, confirming the output content of the large language model as the continuation content corresponding to the key information.

[0020] The content generation conditions include: no subsequent content is output after the output content; or the output content ends with a preset stop punctuation mark; or the content volume of the output content reaches a preset third content volume threshold.

[0021] Wherein, after generating the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guiding instruction by using the preset large language model, it also includes: determining the number of continuation contents corresponding to the key information generated by the large language model; and caching the continuation contents corresponding to the key information generated by the large language model when the number of continuation contents corresponding to the key information is less than a preset number threshold.

[0022] Among them, the outputting of the continued content corresponding to the key information includes: splicing the target content in the input information and the continued content waiting to be output into continuous content; performing content security verification on the spliced ​​continuous content; and outputting the continued content if the content security verification passes.

[0023] The embodiment of the present application also provides a device for generating continuation content, including: an extraction module for extracting key information from received input information; a generation module for generating continuation content corresponding to the key information by using a retrieval enhanced generation method; and an output module for outputting the continuation content corresponding to the key information.

[0024] An embodiment of the present application also provides a device for generating continuation content, comprising: at least one communication interface; at least one bus connected to the at least one communication interface; at least one processor connected to the at least one bus; and at least one memory connected to the at least one bus, wherein the processor is configured to: execute a program for generating continuation content stored in the memory to implement any of the methods for generating continuation content described above.

[0025] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are executed to implement any of the above-described methods for generating continued content.

[0026] The above technical solution provided by the embodiment of the present application has the following advantages over the prior art: the method provided by the embodiment of the present application can extract key information from the received input information; generate the continuation content corresponding to the key information by using the retrieval enhancement generation method; and output the continuation content corresponding to the key information. In the embodiment of the present application, the retrieval enhancement generation method can be used to first increase the amount of information of the key information by retrieval, and then generate the continuation content based on the key information with the increased amount of information. In this way, after the key information is searched and enhanced, the information to be expressed by the key information is more specific and comprehensive. When generating the continuation content, the referenced information is not limited to the key information itself, which can make the generated continuation content more in line with the user's intention, thereby improving the usability of the continuation content. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0028] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0029] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplified descriptions do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and unless otherwise stated, the figures in the drawings do not constitute proportional limitations.

[0030] Figure 1 A flowchart of a method for generating continued content according to an embodiment of the present application;

[0031] Figure 2 is a flowchart of the steps of extracting key information according to an embodiment of the present application;

[0032] Figure 3 A flowchart of the steps for generating the continued content according to an embodiment of the present application;

[0033] Figure 4 is a structural diagram of a device for generating continued content according to an embodiment of the present application;

[0034] Figure 5 It is a structural diagram of a device for generating continued content according to an embodiment of the present application. DETAILED DESCRIPTION

[0035] 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 making creative work are within the scope of protection of this application.

[0036] The disclosure below provides many different embodiments or examples to realize the different structures of the present application. In order to simplify the disclosure of the present application, the parts and settings of specific examples are described below. Of course, they are only examples, and the purpose is not to limit the present application. In addition, the present application can repeat reference numbers and / or letters in different examples. This repetition is for the purpose of simplification and clarity, and does not itself indicate the relationship between the various embodiments and / or settings discussed.

[0037] The embodiment of the present application provides a method for generating a continuation content. The execution subject of the embodiment of the present application is a continuation content generation terminal. The continuation content generation terminal can be deployed as a device in a server, or the continuation content generation terminal can be deployed in a device independent of the server. Figure 1 , which is a flow chart of a method for generating continued content according to an embodiment of the present application.

[0038] Step S110, extracting key information from the received input information.

[0039] Input information refers to the information entered in the preset client.

[0040] Key information refers to the information extracted from the input information and used to generate the continued content.

[0041] Specifically, the server synchronizes the content currently being processed by the client, the continuation position, and the currently set attribute information; determines the target content based on the synchronized content and the continuation position; generates input information based on at least the target content and sends it to the continuation content generation end; and the continuation content generation end extracts key information from the received input information.

[0042] Furthermore, the input information at least includes: target content. The target content refers to the content located before and / or after the continuation position. The types of target content include, but are not limited to: all the content being processed, part of the content being processed, and content of a preset content length before and / or after the cursor position. Furthermore, the target content may be content existing in a content carrier, such as a document, a picture, an application, etc. The target content is content already existing in the content carrier, such as text in a document, a picture in a document, text in a picture, edited information in a conversation program, etc. Among them, the continuation position refers to the cursor pen position, cursor position, and cursor display position in the client, that is, the continuation position selected by the cursor.

[0043] Step S120, using a retrieval augmented generation (RGA) method to generate the continuation content corresponding to the key information.

[0044] The search enhancement generation method refers to a method of generating the continuation content according to the extracted key information and the information retrieved using the key information. How to use the search enhancement generation method to generate the continuation content corresponding to the key information will be described later, so it will not be repeated here.

[0045] The continuation content refers to the content that is continuous with the target content in terms of content, and can be connected with the target content to form a coherent content, and the continuation content is obtained based on the target content. The types of the continuation content include but are not limited to: characters, text, pictures, etc.; the type of the continuation content can be different from the target type, for example, the target content can be a picture, and its continuation content can be text; the type of the continuation content can be the same as the target type, for example, the target content can be a picture, and its continuation content can be a picture.

[0046] Step S130: outputting the continued content corresponding to the key information.

[0047] In the case where the key information includes position information for outputting the continued content, the continued content is output at a position corresponding to the position information based on the position information. The position information here is used to display the position of the continued content. This position can be in front of, behind, above, below the target content, or can be displayed in a separate window. Regardless of the position, the target content and the continued content are associated with each other. When the target content changes, the continued content may also change.

[0048] When the continuation content generation end is deployed on a server, after generating the continuation content corresponding to the key information, the continuation content corresponding to the key information can be output to the client; when the continuation content generation end is deployed on an independent device, after generating the continuation content corresponding to the key information, the continuation content corresponding to the key information can be output to the client through the server; after the client receives the continuation content corresponding to the key information, the client can display the continuation content corresponding to the key information for the user to choose and use.

[0049] In an embodiment of the present application, key information is extracted from the received input information; a search-enhanced generation method is used to generate continuation content corresponding to the key information; and the continuation content corresponding to the key information is output. In an embodiment of the present application, the search-enhanced generation method can be used to first increase the amount of information in the key information through search, and then generate continuation content based on the key information with the increased amount of information. In this way, after the key information is search-enhanced, the information to be expressed by the key information is more specific and comprehensive. When generating the continuation content, the referenced information is not limited to the key information itself, which can make the generated continuation content more in line with the user's intention, thereby improving the usability of the continuation content.

[0050] In order to make the embodiments of the present application easier to understand, the generation process of the continuation content of the embodiments of the present application will be further described below.

[0051] In the embodiment of the present application, the method for generating the continuation content is applied at the continuation content generating end.

[0052] In an embodiment of the present application, a long connection is established between a client having a content continuation requirement and a server for generating input information, and a long connection is established between the server for generating input information and a continuation content generation end for generating continuation content. Based on the long connection, the client synchronizes the content being processed and the continuation position to the server, as well as the attribute information set by the synchronization client; and, based on the long connection, the continuation content generation end receives the input information sent by the server. Furthermore, in the case where the continuation content generation end is deployed on the server, a long connection can be established only between the client and the server.

[0053] In the embodiment of the present application, the server generates input information and sends it to the continuation content generation end at least according to the target content currently being processed by the client and the continuation position. In this way, the input information at least includes the target content. Of course, the input information can also include the attribute information currently set by the client side. The attribute information includes: user information and / or setting information.

[0054] User information refers to the user attributes corresponding to the client. User information includes but is not limited to: account code and device information. In order to make the continued content generated by the server more in line with the user's intention, the server can store historical information corresponding to the user information, such as: historical calls, historical continued content, etc., and use this historical information to increase the amount of key information. For details on how to increase the amount of key information, please refer to the following description, which will not be repeated here.

[0055] Setting information refers to the properties of the generated continuation content. Setting information includes, but is not limited to: type information of the continuation content. Among them, in order to make the generated continuation content more in line with the user's intention, the type information used to generate the continuation content can be pre-set on the client; when obtaining the content currently being processed, the type information is obtained; and the content currently being processed and the type information are synchronized to the server. Of course, in order to improve the efficiency of content continuation, after establishing a long connection with the server, the client can synchronize the content currently being processed and the most recently set type information when synchronizing information for the first time. When synchronizing information subsequently, if the type information does not change, the type information will no longer be synchronized.

[0056] Type information includes, but is not limited to, writing style information and user type information.

[0057] The behavior style information is the content style of the continued content. The types of behavior style information include but are not limited to: rigorous, humorous, written language and colloquial. Furthermore, default behavior style information can be set, for example: written language.

[0058] User type information is the usage scenario of the continued content. The types of user type information include but are not limited to: general, teacher, administration and operation. Furthermore, default user type information can be set, for example: administration.

[0059] In the embodiment of the present application, after receiving the input information, the continued content generation end can extract key information from the input information. Figure 2 , which is a flow chart of the key information extraction steps according to an embodiment of the present application.

[0060] Step S210 , extracting target content from the input information and identifying the guidance type corresponding to the target content by performing multimodal analysis on the input information.

[0061] Multimodal parsing can parse input information into continued content to generate usable data formats.

[0062] The guiding type is the content type that the follow-up content corresponding to the target content should have.

[0063] Types of guide types include, but are not limited to, table types, title types, body types, and other viewpoint types. Among them, multiple levels of title types can be set according to needs, such as the first title type and the second title type.

[0064] The table type means that the continuation content is the content in the table.

[0065] The title type refers to the continued content being the title. Among them, the first title type can be a supplementary title type. The second title type can be a subtitle type.

[0066] The main body type refers to the continuation of the content as the main text.

[0067] Other viewpoint types refer to the continued content having a different viewpoint from the target content.

[0068] Furthermore, by performing multimodal analysis on the input information, the information carried by the input information can be extracted. At least the target content can be extracted from the input information. In the case where the input information also includes other information, other information can also be extracted. The other information includes but is not limited to: user information and / or attribute information.

[0069] Furthermore, by performing multimodal analysis on the input information, the type of lead corresponding to the target content can be identified. Specifically, when the target content is the first line of content, the type of lead corresponding to the target content is determined to be the first title type; when the target content is the content in a paragraph, or when the target content is the beginning of a paragraph and the content of the previous paragraph of the target content meets the preset title condition, the type of lead corresponding to the target content is determined to be the main body type; when the target content is the beginning of a paragraph and the content of the previous paragraph of the target content does not meet the preset title condition, the type of lead corresponding to the target content is predicted based on the target content.

[0070] The title condition is used to measure whether the content is a title. The title condition includes at least one of the following: the content style is the title style, and the content format is the title format. Among them, the title style includes: the content is a preset font size, or the content font is different from the font of the main body type content. The title format includes: the content is a separate paragraph, and the content is at the beginning of the paragraph or in the middle of the paragraph.

[0071] For example: in the input information received for the first time, the target content 1 extracted is "Performance Report", and the target content 1 is the first line of content edited by the user. At this time, it can be determined that the target content 1 is a major title; in the input information received again, the target content 2 is extracted, and the target content 2 is the content in the next paragraph of the target content 1. At this time, it can be determined that the target content 2 is the main text; in the input information received subsequently, the target content 3 is extracted, and the target content 3 is the beginning content of the next paragraph of the target content 2. Since the target content 2 is not a title, and the target content 3 is the beginning content of a paragraph, it cannot be determined whether the target content 3 is a main text paragraph or a subheading, so it is necessary to predict the guide type corresponding to the target content 3.

[0072] Further, predicting the guidance type corresponding to the target content according to the target content includes: predicting the type of content of the next paragraph of the target content according to the target content using a preset type prediction model; wherein the types include title, main body and other viewpoints; in the case where it is predicted that the content of the next paragraph of the target content is a title, determining the guidance type corresponding to the target content to be a second title type; in the case where it is predicted that the content of the next paragraph of the target content is a main body, determining the guidance type corresponding to the target content to be a main body type; in the case where it is predicted that the content of the next paragraph of the target content is other viewpoints, determining the guidance type corresponding to the target content to be other viewpoint types.

[0073] Step S220: extract key content from the target content according to the guidance type corresponding to the target content.

[0074] When the guide type of the target content is a table type or a title type, all the content in the target content is used as the key content; when the guide type of the target content is a main body type or other viewpoint types, all the content in the target content is used as the key content, or part of the content is extracted from the target content as the key content.

[0075] Furthermore, extracting part of the target content as the key content includes: extracting a content segment of a preset length at a preset position in the target content as the key content; or, when the content volume of the target content is greater than a preset first content volume threshold, inputting the target content into a preset content extraction model, and extracting the key content from the target content using the content extraction model.

[0076] The first content volume threshold may be an empirical value or a value obtained through experiments. For example, when the number of words in the target content is greater than 500, the target content is input into a content extraction model, and the content extraction model is used to extract key content from the target content. The content extraction model is a pre-trained model for extracting key content. The type of model can be selected according to needs. For example, a deep learning model is selected as a content extraction model.

[0077] Further, extracting a content segment of a preset length at a preset position in the target content includes: extracting a content segment of a preset length from the last character in the target content; wherein the preset length may be a preset word value. The preset length may be an empirical value or a value obtained through experiments. The content in the content segment is in different paragraphs, in the same paragraph, or in at least one associated content frame.

[0078] For example: add two associated text boxes on the client, after the text entered in the first text box reaches the maximum text amount corresponding to the first text box, the text entered again will be entered into the second text box, and the server can generate target content based on the content synchronized from the client, and the target content can include the contents of the two text boxes. In this way, when extracting a content segment of a preset length, the content segment can also include the contents of the two text boxes.

[0079] Step S230: generating the key information according to the guidance type corresponding to the target content and the key content extracted from the target content.

[0080] The guidance type corresponding to the target content and the key content extracted from the target content are determined as the key information; or, attribute information is extracted from the input information; input enhancement processing is performed on the key content using the attribute information, and the attribute information, the key content after the input enhancement processing, and the guidance type corresponding to the target content are determined as the key information.

[0081] Furthermore, in order to make the key content more specific and complete, the attribute information can be used to perform input enhancement processing on the key content. Specifically, the memory information corresponding to the attribute information can be obtained; wherein, the memory information includes: keywords corresponding to the preset function plug-in; when the key content matches the keyword corresponding to the function plug-in, the function plug-in is called to make the function plug-in perform the function task corresponding to the function plug-in according to the key content, obtain the execution result returned by the function task, and use the execution result and the key content as the key content after input enhancement processing.

[0082] Memory information is a record corresponding to user information. The types of memory information may include: short-term memory information and long-term memory information. Memory information may be stored in a cache to improve query efficiency.

[0083] Short-term memory information is the memory information that is easy to change corresponding to the user information. For example: the setting information used last time (writing style information and user type information), the continued content adopted by the user, and the keywords corresponding to the function plug-ins that the user calls most. Among them, the function plug-in is used to perform the preset function tasks. For example: the order acquisition plug-in is used to obtain the user's order number, and the weather query plug-in is used to query weather information.

[0084] Long-term memory information is the memory information corresponding to the user information that is not easily changed, for example: the user's gender, occupation, and the environment of the client's device.

[0085] Whether the memory information is easy to change can be determined by setting a preset time length. The memory information that changes within the preset time length is short-term memory information, and the memory information that does not change within the preset time length is long-term memory information.

[0086] When performing input enhancement processing, it can be determined whether there is a keyword of the functional plug-in that matches the key content in the short-term memory information. The key content matches the keyword corresponding to the functional plug-in, which means that the keyword corresponding to the functional plug-in is included in the key content, or the similarity between the key content and the keyword of the functional plug-in is greater than a preset similarity threshold.

[0087] For example: the key content is "The weather in area A is changeable, so you always have to take an umbrella with you when you go out to feel at ease." The memory information includes the keyword corresponding to the weather query plug-in, and the keyword is "weather." Then it can be determined that the key content matches the keyword corresponding to the weather query plug-in, and the weather query plug-in can be called. The weather query plug-in can be used to query "the weather in area A" and return the execution result, that is, the specific weather conditions in area A.

[0088] In an embodiment of the present application, after extracting key information from the input information, a search-enhanced generation method can be used to generate continued content corresponding to the key information.

[0089] Specifically, before generating the continuation content corresponding to the key information, a knowledge retrieval system can be pre-set. The information stored in the knowledge retrieval system includes, but is not limited to: paradigm templates, pre-stored information, memory information stored corresponding to user information, and historical continuation content. Of course, the memory information and historical continuation content corresponding to the user information can also be stored separately in other storage spaces. For example, the memory information and historical continuation content are stored in a preset buffer.

[0090] A paradigm template is a pre-set content structure information, for example, the outline of an article.

[0091] The memory information has been described above and will not be repeated here.

[0092] The pre-stored information is pre-collected standardized information, for example: ancient poems, high-quality sentences, and the beginning of articles of a preset type (such as Dear So-and-So).

[0093] The historical continuation content is the continuation content corresponding to the user information before the current time. The historical continuation content may be the adopted continuation content.

[0094] In order to improve the comprehensiveness and accuracy of the information in the knowledge retrieval system, the knowledge retrieval system may be updated at preset time intervals or after a preset number of continued contents are generated.

[0095] After the knowledge retrieval system is set up, information matching the key information can be retrieved in the knowledge retrieval system, and the continuation content corresponding to the key information is generated according to the key information and the information matching the key information. According to different scenarios, two methods of generating continuation content are given below.

[0096] Method 1: When the key information is extracted for the first time, the guide type corresponding to the target content included in the key information is queried; when the guide type corresponding to the target content is the first title type, the paradigm template matching the key information is queried and the paradigm template is determined as the continued content.

[0097] For example, when a user only inputs a title, the user may not be sure how to write the subsequent content. In this case, the user can query the paradigm template that matches the title and output the paradigm template as the continuation content. When the paradigm template is displayed on the client, the user can write according to the structure of the paradigm template. Furthermore, if the target content is the title "personal resume", the continuation content generation end can query the paradigm template corresponding to the personal resume and output it to the client. The paradigm template includes: a writing framework for personal introduction, a writing framework for experience information, a writing framework for project information, etc. The user can refer to the paradigm template to describe the personal resume.

[0098] Method 2, query the auxiliary content that matches the key information; generate a large model guidance instruction based on the key information; use the preset large language model to generate the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guidance instruction. Furthermore, when the key information is extracted for the first time and the guidance type corresponding to the target content is not the first title type, and when the key information is extracted subsequently, query the auxiliary content that matches the key information; generate a large model guidance instruction based on the key information; use the preset large language model to generate the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guidance instruction. In method 2, the auxiliary content can be used to increase the amount of information in the key information, so that the continuation content generated by the large language model is more in line with the user's intention.

[0099] The following further describes the method for generating the continuation content of method 2. Figure 3 As shown, it is a flowchart of the steps of generating the continued content according to an embodiment of the present application.

[0100] Step S310: query auxiliary content matching the key information.

[0101] The auxiliary information includes at least one of the following: a paradigm template and query information. The query information includes at least one of the following: memory information, pre-stored information, and historical continuation content.

[0102] Specifically, starting from receiving the input information, each time the content volume of the input information increases by a preset second content volume threshold, a paradigm template matching the key information is queried; starting from receiving the input information, each time the key information is extracted, query information matching the key information is queried.

[0103] The second content volume threshold is an empirical value or a value obtained through experiments. For example, every time the target content in the input information increases by 300 words, a paradigm template matching the key information extracted from the input information is searched once.

[0104] Furthermore, each time input information is received, key information is extracted, and each time key information is extracted, query information matching the key information is searched. The query information matching the key information includes at least one of the following methods: searching for memory information corresponding to the attribute information according to the attribute information included in the key information; searching for pre-stored information matching the key content in a preset knowledge retrieval system according to the key content included in the key information; searching for historical continuation content matching the key content in the cached historical continuation content according to the key content included in the key information.

[0105] Step S320: Generate a large model guidance instruction based on the key information.

[0106] The large model prompt instruction (Prompt) has a corresponding relationship with the prompt type, and the large model prompt instruction is used to guide the large language model to generate the continuation content of the corresponding prompt type.

[0107] Specifically, a large model guidance instruction is generated according to the guidance type corresponding to the target content included in the key information.

[0108] Further, when the guide type corresponding to the target content is the first title type, a first title guide instruction is generated; when the guide type corresponding to the target content is the second title type, a second title guide instruction is generated; when the guide type corresponding to the target content is the table type or the body type, a main content guide instruction is generated; when the guide type corresponding to the target content is other viewpoint types, other viewpoint guide instructions are generated. Among them, the first title guide instruction can be a complete title prompt, the second title guide instruction can be a subtitle prompt, the main content guide instruction can be a content prompt, and other viewpoint guide instructions can be other viewpoint prompts.

[0109] Step S330, using a preset large language model, generating the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guidance instruction.

[0110] Using the large language model corresponding to the attribute information in the key information, the continuation content corresponding to the key information is generated according to the key content in the key information, the auxiliary content and the large model guidance instruction. Further, using the preset large language model, the continuation content carrying the relevance tag can be generated according to the key information, the auxiliary content and the large model guidance instruction.

[0111] The relevance tag is the relevance between the continued content and the key content in the key information. The greater the relevance, the higher the relevance between the continued content and the key content, and the closer the continued content is to the expression idea of ​​the target content, that is, the better the content continuity between the continued content and the target content. The lower the relevance, the lower the relevance between the continued content and the key content, and the further the continued content is from the expression idea of ​​the target content, that is, the worse the content continuity between the continued content and the target content.

[0112] Furthermore, an application programming interface (API) request is sent to the large language model; wherein the key information, the auxiliary content and the large model guiding instruction are carried in the interface request; the large language model generates the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guiding instruction carried in the interface request; the output content of the large language model is read; and when the output content meets the preset content generation conditions, the output content of the large language model is confirmed as the continuation content corresponding to the key information.

[0113] The interface request may be an application programming interface (Application Programming Interface, API for short) request.

[0114] When reading the output content of the large language model, the output content of the large language model can be read through Server SentEvents (SSE for short).

[0115] The content generation condition is used to measure whether the output content can be used as a continuation content. The content generation condition includes: no subsequent content is output after the output content; or the output content ends with a preset stop punctuation mark; or the content volume of the output content reaches a preset third content volume threshold.

[0116] No subsequent content output means that after the input content, there is no continuous content output of the output content, wherein there is no stop punctuation mark between the output content and the continuous content.

[0117] Stop punctuation marks are used to indicate the end of content, including but not limited to: period ".", exclamation mark "!", question mark "?" and tilde "~".

[0118] The third content volume threshold is used to measure whether the content is finished. The third content volume threshold is an empirical value or a value obtained through experiments. For example, the third content volume threshold is 20 words.

[0119] In an embodiment of the present application, since there may be content in the key content and the auxiliary content that cannot be recognized by the large language model, in order to enable the large language model to normally generate the continuation content and enable the generated continuation content to be recognized by the client, before generating the continuation content corresponding to the key information by using the large language model corresponding to the attribute information in the key information, according to the key content in the key information, the auxiliary content and the large model guidance instruction, the first content can be identified in the key content and the auxiliary content, and the first content appearing in the key content and the auxiliary content can be replaced with the corresponding second content, so that the large language model generates the continuation content corresponding to the key information according to the replaced key content and the auxiliary content and the large model guidance instruction; after generating the continuation content corresponding to the key information, the second content can be identified in the generated continuation content and the second content can be replaced with the first content.

[0120] The first content is the content that cannot be recognized by the large language model and needs to be blocked.

[0121] The second content is content that can be recognized by the large language model.

[0122] Unrecognizable content includes, but is not limited to, preset symbols. Each unrecognizable content has a corresponding second content. When replacing, the unrecognizable content is replaced with the second content corresponding to the unrecognizable content. For example, in the key content and auxiliary content, the line break symbol is "\r", while the line break symbol recognizable by the large language model is "\n". Therefore, the line break symbols in the key content and auxiliary content need to be replaced from "\r" to "\n". After the continuation content is generated, in order for the client to recognize the line break symbol, the line break symbol in the continuation content also needs to be replaced from "\n" to "\r".

[0123] The content that needs to be shielded includes but is not limited to: the user's private information. The content that needs to be shielded can be replaced with a second content with preset content. For example, if the content that needs to be shielded is the ID card number, the ID card number can be replaced with a group of symbols, and after the continued content is generated, the group of symbols can be replaced with the ID card number. Of course, according to different needs, the content that needs to be shielded can also not be restored.

[0124] In the embodiment of the present application, for the same key information, auxiliary content and large model guidance instructions, the large language model can generate a preset number (quantity threshold) of continued content, and determine the relevance between each continued content and the key content in the key information. The quantity threshold is the number of continued contents that need to be generated. The quantity threshold can be an empirical value or a value obtained through experiments.

[0125] Furthermore, after the continuation content corresponding to the key information is generated by using the preset large language model according to the key information, the auxiliary content and the large model guidance instruction, the number of continuation contents corresponding to the key information generated by the large language model can be determined; if the number of continuation contents corresponding to the key information is less than the preset quantity threshold, the continuation contents corresponding to the key information generated by the large language model are cached until the number of cached continuation contents corresponding to the key information reaches the quantity threshold. Among them, a resident thread can be used to cache the continuation contents corresponding to the key information generated by the large language model and determine whether the number of cached continuation contents has reached the quantity threshold.

[0126] In an embodiment of the present application, in order to increase the accuracy and readability of the continued content, the target keyword can be queried in the continued content before outputting the continued content corresponding to the key information; when the continued content includes the target keyword, the target keyword is replaced with the target replacement word corresponding to the target keyword.

[0127] Target keywords refer to the content that needs to be replaced or completed in the continued content. Target keywords include but are not limited to: preset names and preset titles.

[0128] The target replacement word is the standard expression of the content corresponding to the target keyword. For example, if the continued content includes "authority", "authority" is the target keyword, and its corresponding target replacement word is "claim". In this case, "authority" in the continued content can be replaced with "claim". After the replacement, the expression of the continued content is more accurate and easy to understand.

[0129] In an embodiment of the present application, outputting the continuation content corresponding to the key information includes: splicing the target content in the input information and the continuation content to be output into continuous content; performing content security verification on the spliced ​​continuous content; and outputting the continuation content if the content security verification is passed. The content security verification is used to verify whether there is sensitive information in the spliced ​​continuous content; if sensitive information exists, it is determined that the content security verification fails, and the output of the continuation content is prohibited; if sensitive information does not exist, it is determined that the content security verification passes, and the output of the continuation content is allowed. Sensitive information includes, but is not limited to, information that is likely to mislead users and information that is expressly prohibited. For example: allowing the public to trample on the ornamental lawn at will. In this way, it can be avoided that after the continuation content is selected by the user, the continuous content formed by connecting with the target content includes sensitive information.

[0130] The present application also provides a device for generating continued content. Figure 4, which is a structural diagram of a device for generating continued content according to an embodiment of the present application.

[0131] The device for generating the continued content includes:

[0132] The extraction module 410 is used to extract key information from the received input information.

[0133] The generation module 420 is used to generate the continuation content corresponding to the key information by using the search enhanced generation method.

[0134] The output module 430 is used to output the continued content corresponding to the key information.

[0135] The functions of the device described in the embodiment of the present application have been described in the above method embodiment. Therefore, for any details not provided in the description of this embodiment, please refer to the relevant descriptions in the above embodiments and will not be repeated here.

[0136] The present application also provides a device for generating continued content, such as Figure 5 , which is a structural diagram of a device for generating continued content according to an embodiment of the present application.

[0137] The device for generating the continued content includes: a processor 510 , a communication interface 520 , a memory 530 and a communication bus 540 . The processor 510 , the communication interface 520 , and the memory 530 communicate with each other via the communication bus 540 .

[0138] The memory 530 is used to store computer programs.

[0139] In one embodiment of the present application, the processor 510, when executing the program stored in the memory 530, implements the method for generating the continuation content provided by any of the aforementioned method embodiments, including: extracting key information from the received input information; generating the continuation content corresponding to the key information by using the retrieval enhanced generation method; and outputting the continuation content corresponding to the key information.

[0140] Among them, extracting key information from the received input information includes: extracting target content from the input information and identifying the guidance type corresponding to the target content by performing multimodal analysis on the input information; extracting key content from the target content according to the guidance type corresponding to the target content; generating the key information according to the guidance type corresponding to the target content and the key content extracted from the target content.

[0141] Wherein, the identifying of the guidance type corresponding to the target content includes: when the target content is the first line of content, determining that the guidance type corresponding to the target content is the first title type; when the target content is content in a paragraph, or when the target content is the beginning of a paragraph and the content of the previous paragraph of the target content meets the preset title condition, determining that the guidance type corresponding to the target content is the main body type; when the target content is the beginning of a paragraph and the content of the previous paragraph of the target content does not meet the preset title condition, predicting the guidance type corresponding to the target content based on the target content.

[0142] Among them, predicting the guidance type corresponding to the target content according to the target content includes: using a preset type prediction model to predict the type of content of the next paragraph of the target content according to the target content; when it is predicted that the content of the next paragraph of the target content is a title, determining the guidance type corresponding to the target content is a second title type; when it is predicted that the content of the next paragraph of the target content is a main body, determining the guidance type corresponding to the target content is a main body type; when it is predicted that the content of the next paragraph of the target content is other viewpoints, determining the guidance type corresponding to the target content is other viewpoint types.

[0143] Among them, generating the key information according to the guidance type corresponding to the target content and the key content extracted from the target content includes: determining the guidance type corresponding to the target content and the key content extracted from the target content as the key information; or, extracting attribute information from the input information; using the attribute information, performing input enhancement processing on the key content, and determining the attribute information, the key content after the input enhancement processing, and the guidance type corresponding to the target content as the key information.

[0144] Among them, the use of the attribute information to perform input enhancement processing on the key content includes: obtaining memory information corresponding to the attribute information; wherein, the memory information includes: keywords corresponding to preset function plug-ins; when the key content matches the keywords corresponding to the function plug-in, calling the function plug-in, causing the function plug-in to execute the function task corresponding to the function plug-in according to the key content, obtaining the execution result returned by the function task, and using the execution result and the key content as the key content after input enhancement processing.

[0145] Among them, the use of retrieval enhanced generation method to generate the continuation content corresponding to the key information includes: querying auxiliary content matching the key information; generating a large model guidance instruction based on the key information; using a preset large language model to generate the continuation content corresponding to the key information based on the key information, the auxiliary content and the large model guidance instruction.

[0146] Among them, the query for auxiliary content matching the key information includes: starting from receiving the input information, every time the content volume of the input information increases by a preset second content volume threshold, querying the paradigm template matching the key information; starting from receiving the input information, every time the key information is extracted, querying the query information matching the key information.

[0147] Among them, the query information that matches the key information includes at least one of the following methods: according to the attribute information included in the key information, querying the memory information corresponding to the attribute information; according to the key content included in the key information, querying the pre-stored information that matches the key content in a preset knowledge retrieval system; according to the key content included in the key information, querying the historical continuation content that matches the key content in the cached historical continuation content.

[0148] The generating of the large model instruction according to the key information includes: generating the large model guidance instruction according to the guidance type corresponding to the target content included in the key information.

[0149] Among them, the large model guidance instructions are generated according to the guidance type corresponding to the target content included in the key information, including: when the guidance type corresponding to the target content is the first title type, a first title guidance instruction is generated; when the guidance type corresponding to the target content is the second title type, a second title guidance instruction is generated; when the guidance type corresponding to the target content is a table type or a body type, a body content guidance instruction is generated; when the guidance type corresponding to the target content is other viewpoint types, other viewpoint guidance instructions are generated.

[0150] Among them, the use of a preset large language model to generate the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guidance instructions includes: using a large language model corresponding to the attribute information in the key information to generate the continuation content corresponding to the key information according to the key content in the key information, the auxiliary content and the large model guidance instructions.

[0151] Among them, the use of a preset large language model to generate the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guidance instructions includes: using a preset large language model to generate the continuation content carrying a relevance label according to the key information, the auxiliary content and the large model guidance instructions; wherein the relevance label is the relevance between the continuation content and the key content in the key information.

[0152] Among them, the use of a preset large language model to generate the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guiding instructions includes: sending an interface request to the large language model; wherein the key information, the auxiliary content and the large model guiding instructions are carried in the interface request; the large language model generates the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guiding instructions carried in the interface request; reading the output content of the large language model; and when the output content meets the preset content generation conditions, confirming the output content of the large language model as the continuation content corresponding to the key information.

[0153] The content generation conditions include: no subsequent content is output after the output content; or the output content ends with a preset stop punctuation mark; or the content volume of the output content reaches a preset third content volume threshold.

[0154] Wherein, after generating the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guiding instruction by using the preset large language model, it also includes: determining the number of continuation contents corresponding to the key information generated by the large language model; and caching the continuation contents corresponding to the key information generated by the large language model when the number of continuation contents corresponding to the key information is less than a preset number threshold.

[0155] Among them, the outputting of the continued content corresponding to the key information includes: splicing the target content in the input information and the continued content waiting to be output into continuous content; performing content security verification on the spliced ​​continuous content; and outputting the continued content if the content security verification passes.

[0156] The embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method for generating the continuation content provided in any of the above method embodiments are implemented. Since the method for generating the continuation content has been described in detail above, for the details not fully described in the description of this embodiment, please refer to the relevant description in the above embodiments, and no further description will be given here.

[0157] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0158] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0159] It should be understood that the terms used herein are only for the purpose of describing specific example embodiments and are not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms "one", "an" and "said" as used herein may also be meant to include plural forms. The terms "include", "comprise", "contain", and "have" are inclusive, and therefore specify the existence of stated features, steps, operations, elements and / or parts, but do not exclude the existence or addition of one or more other features, steps, operations, elements, parts, and / or combinations thereof. The method steps, processes, and operations described herein are not interpreted as necessarily requiring them to be performed in the specific order described or illustrated, unless the execution order is clearly indicated. It should also be understood that additional or alternative steps may be used.

[0160] The above description is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest range consistent with the principles and novel features applied for herein.

Claims

1. A method for generating continued content, characterized in that: include: Extract key information from the received input information; Generate the continuation content corresponding to the key information by using the search enhancement generation method; Output the continuation content corresponding to the key information.

2. The method according to claim 1, characterized in that: Extracting key information from the received input information includes: By performing multimodal analysis on the input information, target content is extracted from the input information and a guidance type corresponding to the target content is identified; Extracting key content from the target content according to the guide type of the target content; The key information is generated according to the guide type of the target content and the key content extracted from the target content.

3. The method according to claim 2, characterized in that The identifying the guidance type corresponding to the target content includes: In the case where the target content is the first line of content, determining that the guide type corresponding to the target content is the first title type; In the case where the target content is the content in the paragraph, or in the case where the target content is the content at the beginning of a paragraph and the content of the previous paragraph of the target content meets the preset title condition, determining that the guide type corresponding to the target content is the main body type; In the case that the target content is the beginning content of a paragraph and the content of a previous paragraph of the target content does not meet the preset title condition, the guidance type corresponding to the target content is predicted according to the target content.

4. The method according to claim 3, characterized in that The predicting, according to the target content, a guidance type corresponding to the target content includes: Predicting the type of content of the next paragraph of the target content according to the target content using a preset type prediction model; In the case where it is predicted that the next paragraph of the target content is a title, determining that the guide type corresponding to the target content is a second title type; In the case where it is predicted that the next paragraph of the target content is the main body, determining the guide type corresponding to the target content as the main body type; When it is predicted that the content of the next paragraph of the target content is other viewpoints, it is determined that the guidance type corresponding to the target content is other viewpoint type.

5. The method according to claim 2, characterized in that: The generating the key information according to the guidance type corresponding to the target content and the key content extracted from the target content includes: determining the guidance type corresponding to the target content and the key content extracted from the target content as the key information; or Attribute information is extracted from the input information; input enhancement processing is performed on the key content using the attribute information, and the attribute information, the key content after the input enhancement processing, and the guidance type corresponding to the target content are determined as the key information.

6. The method according to claim 5, characterized in that The step of utilizing the attribute information to perform input enhancement processing on the key content includes: Acquire memory information corresponding to the attribute information; wherein the memory information includes: keywords corresponding to the preset function plug-in; In the case where the key content matches the keyword corresponding to the function plug-in, the function plug-in is called to make the function plug-in execute the function task corresponding to the function plug-in according to the key content, obtain the execution result returned by the function task, and use the execution result and the key content as the key content after input enhancement processing.

7. The method according to claim 1, characterized in that The method of using the search enhancement generation method to generate the continued content corresponding to the key information includes: Querying auxiliary content matching the key information; Generate a large model guidance instruction according to the key information; By using a preset large language model, the continuation content corresponding to the key information is generated according to the key information, the auxiliary content and the large model guiding instruction.

8. The method according to claim 7, characterized in that The query for auxiliary content matching the key information includes: Starting from receiving the input information, when the content volume of the input information increases by a preset second content volume threshold, searching for a paradigm template matching the key information; Starting from receiving the input information, each time the key information is extracted, query information matching the key information is searched.

9. The method according to claim 8, characterized in that The query information matching the key information includes at least one of the following methods: According to the attribute information included in the key information, query the memory information corresponding to the attribute information; According to the key content included in the key information, searching the pre-stored information matching the key content in the preset knowledge retrieval system; According to the key content included in the key information, the cached historical continuation content is searched for the historical continuation content that matches the key content.

10. The method according to claim 7, characterized in that The step of generating a large model instruction according to the key information includes: A large model guidance instruction is generated according to the guidance type corresponding to the target content included in the key information.

11. The method according to claim 10, characterized in that The generating of a large model guidance instruction according to the guidance type corresponding to the target content included in the key information includes: In the case where the guidance type corresponding to the target content is a first title type, generating a first title guidance instruction; In the case where the guidance type corresponding to the target content is a second title type, generating a second title guidance instruction; In the case where the guidance type corresponding to the target content is a table type or a body type, generating a body content guidance instruction; When the guidance type corresponding to the target content is other viewpoint types, other viewpoint guidance instructions are generated.

12. The method according to claim 7, characterized in that The using of the preset large language model to generate the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guidance instruction includes: Using a large language model corresponding to the attribute information in the key information, the continuation content corresponding to the key information is generated according to the key content in the key information, the auxiliary content and the large model guiding instructions.

13. The method according to claim 7, characterized in that The using of the preset large language model to generate the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guidance instruction includes: Using a preset large language model, a continuation content carrying a relevance tag is generated according to the key information, the auxiliary content and the large model guidance instruction; wherein the relevance tag is the relevance between the continuation content and the key content in the key information.

14. The method according to claim 7, characterized in that The using of the preset large language model to generate the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guidance instruction includes: Sending an interface request to the large language model; wherein the interface request carries the key information, the auxiliary content, and the large model guiding instruction; the large language model generates the continuation content corresponding to the key information according to the key information, the auxiliary content, and the large model guiding instruction carried in the interface request; Reading the output content of the large language model; When the output content meets the preset content generation condition, the output content of the large language model is confirmed as the continued content corresponding to the key information.

15. The method according to claim 14, characterized in that The content generation conditions include: There is no subsequent content output after the output content; or, The output content ends with a preset stop punctuation mark; or, The content volume of the output content reaches a preset third content volume threshold.

16. The method according to claim 7, characterized in that After the preset large language model is used to generate the continuation content corresponding to the key information according to the key information, the auxiliary content and the large model guiding instruction, the method further includes: Determine the amount of continued content corresponding to the key information generated by the large language model; When the number of the continued contents corresponding to the key information is less than a preset number threshold, the continued contents corresponding to the key information generated by the large language model are cached.

17. The method according to claim 1, characterized in that The outputting of the continued content corresponding to the key information includes: splicing the target content in the input information and the continued content waiting to be output into continuous content; Performing content security verification on the spliced ​​continuous content; When the content security verification is passed, the continued content is output.

18. A device for generating continued content, characterized in that: include: An extraction module, used to extract key information from the received input information; A generation module, used to generate the continuation content corresponding to the key information by using a search-enhanced generation method; The output module is used to output the continued content corresponding to the key information.

19. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are executed to implement the method for generating continued content according to any one of claims 1 to 17.

Images