Method and device for generating advertisement script and storage medium

Abstract

The invention discloses a method and device for generating an advertisement script and a storage medium. The method comprises the steps that a first cue of a user is received by a large language model, a task list is generated based on the first cue, the task list comprises description content of multiple steps, and the multiple steps are at least related to characteristics and advantages of recognized products; searching information related to the product on the network based on keywords extracted from the description contents of the multiple steps, and establishing a local knowledge base by utilizing the searched information; receiving, by the large language model, content from the local knowledge base, and generating a report based on the received content, the report including at least characteristics and advantages of the product; an advertising script for the product is generated by a large language model based on the report.

Description

Technical Field

[0001] The present disclosure generally relates to the field of information services, and more specifically, to a method and device for automatically generating an advertisement script, and a storage medium for storing related programs. Background Art

[0002] Recently, large language models (LLMs) have been greatly developed. Large language models refer to deep learning models trained with large amounts of text data. They can understand the meaning of language text or generate natural language text, and can handle a variety of natural language tasks such as text classification, question answering, and dialogue. There are currently many large language models, such as OpenAI's GPT series (such as GPT-3.5, ChatGPT, and GPT-4), Google's PaLM 2, Metaverse's LLaMA, Baidu's Wenxinyiyan, and so on.

[0003] With the development of LLM technology, it has been increasingly applied to the generation of advertising scripts in the modern advertising industry. Advertising scripts are textual expressions of advertising creativity, usually including shot descriptions, copy content, narration content, etc. Compared with the traditional creation process of manually writing advertising scripts, LLM can generate advertising scripts in a highly automated way. This is because LLM has mastered rich language knowledge and patterns through pre-training in a massive corpus.

[0004] Although LLM has advantages in improving efficiency and saving manpower, it still has some shortcomings. For example, in traditional methods, people usually first understand the unique characteristics and advantages of the product, and then focus on these characteristics and advantages to write the product's advertising script. However, LLM usually does not know the personalized characteristics and advantages of a specific product, and may not even know what the product is. Therefore, the advertising script generated by LLM may not highlight the characteristics and advantages of the product, and may even contain content that is inconsistent with the product or does not comply with laws and regulations. Therefore, the advertising script generated by LLM often requires manual editing and optimization. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the application of LLM in generating advertising scripts, the present disclosure proposes a new solution for generating advertising scripts using LLM. In the present disclosure, information about a product and its competitors is first collected to generate an analysis report containing the characteristics and advantages of the product that are different from its competitors, and then LLM generates an advertising script based on the content of the analysis report. The advertising script generated according to the solution of the present disclosure will not contain content that is inconsistent with the product, and more importantly, it can highlight the unique characteristics and advantages of the product, thereby achieving better publicity effects.

[0006] According to one aspect of the present disclosure, a method for generating an advertising script using a large language model is provided, comprising: receiving a first prompt from a user by the large language model, and generating a task list based on the first prompt, the task list including descriptions of multiple steps, the multiple steps at least being related to identifying features and advantages of a product; searching for information related to the product on the Internet based on keywords extracted from the descriptions of the multiple steps, and establishing a local knowledge base using the searched information; receiving content from the local knowledge base by the large language model, and generating a report based on the received content, the report at least including features and advantages of the product; and generating an advertising script for the product by the large language model based on the report.

[0007] According to another aspect of the present disclosure, a device for generating an advertising script is provided, the device communicates with a large language model, and includes a memory storing a program and one or more processors. The processor is configured to perform the following operations by executing the program: input a first prompt of a user to the large language model; instruct the large language model to generate a task list according to the first prompt, wherein the task list includes descriptions of multiple steps, and the multiple steps are at least related to identifying the characteristics and advantages of a product; search for information related to the product on the network based on keywords extracted from the descriptions of the multiple steps, and establish a local knowledge base using the searched information; input the content of the local knowledge base to the large language model; instruct the large language model to generate a report based on the content in the local knowledge base, wherein the report at least includes the characteristics and advantages of the product; instruct the large language model to generate an advertising script for the product based on the report.

[0008] According to another aspect of the present disclosure, a non-transitory computer-readable medium storing a program is provided. When the program is executed by a computer, the computer executes the method as described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 The overall framework diagram of the advertisement script generation system according to the present disclosure is schematically shown.

[0010] Figure 2 The diagram schematically shows the prompts input by the user to the large language model.

[0011] Figure 3 The process of generating a report based on a task list and a local knowledge base is schematically illustrated.

[0012] Figure 4The process of generating an advertisement script based on a report and user prompts is schematically shown.

[0013] Figure 5 A flow chart of a method for generating an advertisement script according to the present disclosure is shown.

[0014] Figure 6 A modular block diagram of an advertisement script generating system according to the present disclosure is shown.

[0015] Figure 7 An exemplary configuration block diagram of computer hardware for implementing the present disclosure is shown. DETAILED DESCRIPTION

[0016] In the present disclosure, if you want to create an advertising script for a product, you need to obtain a report that at least contains the unique features and advantages of the product, so that the large language model can generate an advertising script that can highlight the features and advantages of the product based on the report. More specifically, the report may include the following content: information about the product, information about other products that compete with the product in the market (hereinafter referred to as "competing products"), comparison and analysis of the product and competing products, the features and advantages of the product that are different from competing products, and so on.

[0017] Figure 1 The overall framework diagram of the advertising script generation system according to the present disclosure is schematically shown. Figure 1 As shown, the user inputs appropriate prompts to the large language model LLM so that the model LLM outputs a task list, which includes descriptions of multiple steps, which are related to the various parts of the report as described above. As an example, these steps may include: steps of collecting information about the product and competing products, steps of identifying the strengths and weaknesses of competing products, steps of discovering the unique features and advantages of the product by comparing the product with the competing products, steps of drawing conclusions, and so on.

[0018] The following is an example of a task list generated by the model LLM, in this case the product is KindleScribe. The task list includes the following five steps:

[0019] “Step 1 > Search the market > Search the market and discover the current trends as well as the competitive landscape of Kindle Scribe.

[0020] Step 2 > Analyze competitors > Find the strengths and weaknesses of competitors and compare the features and advantages of Kindle Scribe with its competitors.

[0021] Step 3 > Analyze the product > Analyze the features and advantages of Kindle Scribe.

[0022] Step 4 > Identify Opportunities > Identify opportunities for Kindle Scribe to differentiate itself from its competitors.

[0023] Step 5 > Conclusion > Draw your conclusion about Kindle Scribe and its competitors. ”

[0024] It should be noted that this is only an exemplary task list, and the task list generated by the model LLM according to the present disclosure does not necessarily include such five steps. However, the task list according to the present disclosure should at least include steps related to discovering the characteristics and advantages of the product that are different from competing products.

[0025] Figure 2 The prompts input by the user to the model LLM are schematically shown, which may include, for example, the following four parts:

[0026] - Definition of the objectives to be achieved, for example, in order to obtain an analytical report on the features and advantages of a product, the model LLM is required to develop a list of tasks to achieve the objective;

[0027] -Definition of the task flow, for example, how many steps does the task list include and whether it includes sub-steps;

[0028] - The format of the task list to be output is limited, for example, the task list is output in the format of "step number > step name > step description content";

[0029] - Limitation of the role of the model LLM, for example, let the model LLM generate content as a salesperson or advertising staff.

[0030] It should be noted that the prompts input to the model LLM do not necessarily include Figure 2 The four aspects shown in the figure can be selected by a technician in the field according to actual needs, or some or all of them can be selected, or input Figure 2 The present disclosure does not limit the requirements of other aspects not shown in the present disclosure.

[0031] Return to reference Figure 1, extract multiple keywords from the step description content of each step in the task list, and then search for product-related information on the network (such as the Internet) based on the extracted keywords, preferably based on the combination of keywords and product names (such as Kindle Scribe). Download the searched information locally to form a local knowledge base related to the product. As an example, keywords can be extracted by the model LLM, and public search engines (such as Google, Baidu) can be used to perform searches, or other customized search tools can be used. In addition, in addition to searching content from the Internet, content provided by users (such as product owners) can also be used to establish a local knowledge base. The formed local knowledge base contains a large number of documents related to the product.

[0032] Then, the model LLM uses the content in the local knowledge base to generate a report. Specifically, for a step in the task list (e.g., step 1), the similarity between each document in the local knowledge base and the step description content of the step (e.g., "search the market, discover current trends and the competitive landscape of Kindle Scribe") is calculated, and one or more documents with a similarity greater than a predetermined threshold are selected.

[0033] Then, the selected document is input to the model LLM, and the model LLM is instructed to generate content corresponding to the step (e.g., content specifically describing the current competitive landscape of Kindle Scribe) based on the content in the input document, which will constitute part of the required report. In this way, corresponding content is generated for each step in the task list in turn. Finally, the content corresponding to steps 1 to 5 is combined together to form the required report.

[0034] In terms of calculating the similarity between the document and the step description content, and thus selecting the document with greater similarity, those skilled in the art may use any appropriate known method to implement. A specific implementation of the present disclosure will be given below.

[0035] In the field of natural language processing and large language models, a token usually refers to the smallest discrete unit in a language text, which can be a word, punctuation mark, number or other language element. A sentence or text is usually divided into a string of tokens, which are then input into a large language model, which attempts to predict the next most likely token. Tokens are the basic units for large language models to train and generate text. In practice, there is usually an upper limit on the number of tokens that can be input into a large language model at one time. To address this situation, the present disclosure provides the following methods: Figure 3 The specific implementation shown.

[0036] Reference Figure 3 , each document in the local knowledge base is divided into multiple sub-documents, so that the size of each sub-document can meet the requirements of the model LLM used for the number of input tokens, that is, so that each sub-document can be input into the model LLM. Next, each sub-document is vectorized, such as word embedding. The vector representations of each sub-document obtained constitute a vector database. On the other hand, the step description content of a step in the task list is also vectorized to obtain its vector representation. In the present disclosure, for example, word2vector or OpenAI's open source vectorization model can be used to perform word embedding.

[0037] Next, the similarity between the vector representation of the step description content and the vector representation of each sub-document in the vector database is calculated. A person skilled in the art may use any appropriate known method to calculate the similarity. As an example, the cosine distance may be calculated, but the present disclosure is not limited thereto. Then, one or more sub-documents ( Figure 3 The model LLM generates the content corresponding to the step using the content contained in the input sub-document. The corresponding content is generated for all steps in the task plan in sequence, and then the content corresponding to each step is combined together to form a report.

[0038] Return to reference Figure 1 , when a report has been generated, since the report itself contains the comparison and analysis of the product and the competing products, the characteristics and advantages of the product compared with the competing products, etc., as a simple implementation, the model LLM can be instructed to automatically generate an advertising script based on the report. In the present disclosure, since the model LLM can learn the characteristics and advantages of the product from the report, the generated advertising script is more targeted.

[0039] As another optional implementation, the model LLM can be instructed to generate reports based not only on the generated reports, but also on the user's specific requirements (such as Figure 1 In this case, the user's specific requirements may include, for example:

[0040] - Requirements regarding advertising, such as advertising style, target audience, length of advertising, etc.;

[0041] - Output format of the advertisement script, for example, outputting the advertisement script in JSON format or other easily processable formats;

[0042] - Include product feature descriptions in the ad script, for example, include product features described in concise words or phrases in the ad script for each shot or scene;

[0043] -The role of the model LLM, for example, let the model LLM generate advertising scripts as a salesperson or advertising staff.

[0044] In this embodiment, if Figure 4 As shown, the generated report can be divided into multiple segments, and each segment is vectorized (e.g., word embedding). On the other hand, the specific requirements (prompts) input by the user are also vectorized. Then, the similarity between the vector representation of the prompt and the vector representation of each segment of the report is calculated. Those skilled in the art can use any appropriate known method to calculate the similarity. As an example, the cosine distance can be calculated, but the present disclosure is not limited thereto. Then, one or more segments ( Figure 4 The similar segment shown in FIG. 1 is input to the model LLM. Then, the model LLM is instructed to generate an advertisement script based on the input similar segment and the user's specific requirements (prompt words).

[0045] Compared with the simple implementation described above, the advertising script generated by the model LLM in this implementation can further meet the user's special requirements for advertising. For example, product features can be emphasized in the form of keywords in the advertising script (for each shot).

[0046] Figure 5 A flowchart of a method for generating an advertisement script according to the present disclosure is shown. Figure 6 A modular block diagram of an advertisement script generating system according to the present disclosure is shown.

[0047] like Figure 5 As shown, in step S510, the model LLM receives the user's first prompt and generates a task list. The first prompt may include, for example, Figure 2 The task list includes descriptions of multiple steps, at least the steps related to identifying the features and advantages of the product.

[0048] In step S520, information related to the product is searched on the Internet based on the keywords extracted from the description content of each step, and a local knowledge base is established using the searched information. In addition, when establishing the local knowledge base, information provided by the owner of the product can also be used. This step can be performed by Figure 6 The search unit 610 and the knowledge base building unit 620 shown in FIG. 6 are used to perform the search.

[0049] In step S530, appropriate content in the local knowledge base is input to the model LLM, and the model LLM generates a report based on the input content, and the report at least includes the characteristics and advantages of the product.

[0050] In step S540, the model LLM generates an advertising script that can highlight the characteristics and advantages of the product based on the report. Preferably, the model LLM is further input with a second prompt of the user, the second prompt including the user's specific requirements for the advertising script to be generated, and the model LLM generates the advertising script based on the report and the second prompt.

[0051] The above has described the scheme of automatically generating advertising scripts using a large language model according to the present disclosure in combination with specific embodiments. Compared with the traditional method of manually writing advertising scripts, the scheme of the present disclosure greatly improves efficiency and reduces labor costs. Compared with the existing method of generating advertising scripts using LLM, since LLM in the present disclosure can generate advertising scripts based on report content containing product information and product features and advantages, it can avoid generating advertising content that is inconsistent with the product due to LLM's lack of knowledge of the product, and more importantly, it can make the generated advertising script highlight the characteristics and advantages of the product that are different from its competitors, thereby achieving better publicity effects.

[0052] The methods described in the above embodiments may be implemented by software, hardware, or a combination of software and hardware. The programs included in the software may be stored in advance in a storage medium provided inside or outside the device. As an example, during execution, these programs are written to a random access memory (RAM) and executed by a processor (e.g., a CPU) to implement the various methods and processes described herein.

[0053] Figure 7 The block diagram shows an example configuration of computer hardware that executes the method of the present disclosure according to a program, and the computer hardware is an example of an apparatus for generating an advertisement script according to the present disclosure.

[0054] like Figure 7 As shown, in a computer 700 , a central processing unit (CPU) 701 , a read only memory (ROM) 702 , and a random access memory (RAM) 703 are connected to one another via a bus 704 .

[0055] The input / output interface 705 is further connected to the bus 704. The input / output interface 705 is connected with the following components: an input device 706 formed with a keyboard, a mouse, a microphone, etc.; an output device 707 formed with a display, a speaker, etc.; a storage device 708 formed with a hard disk, a nonvolatile memory, etc.; a communication device 709 formed with a network interface card (such as a local area network (LAN) card, a modem, etc.); and a drive 710 that drives a removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory.

[0056] In the computer having the above-described structure, the CPU 701 loads the program stored in the storage device 708 into the RAM 703 via the input / output interface 705 and the bus 704 and executes the program so as to perform the method described above.

[0057] The program to be executed by the computer (CPU 701) may be recorded on a removable medium 711 as a package medium, which is formed in the form of, for example, a magnetic disk (including a floppy disk), an optical disk (including a compact disc-read only memory (CD-ROM)), a digital versatile disc (DVD), etc.), a magneto-optical disk, or a semiconductor memory. In addition, the program to be executed by the computer (CPU 701) may also be provided via a wired or wireless transmission medium such as a local area network, the Internet, or digital satellite broadcasting.

[0058] When the removable medium 711 is mounted in the drive 710, the program can be installed in the storage device 708 via the input / output interface 705. In addition, the program can be received by the communication device 709 via a wired or wireless transmission medium and installed in the storage device 708. Alternatively, the program can be pre-installed in the ROM 702 or the storage device 708.

[0059] The program executed by the computer may be a program that performs processing according to the sequence described in this specification, or may be a program that performs processing in parallel or performs processing when necessary such as when called.

[0060] The modules or devices described herein are only in a logical sense and do not strictly correspond to physical devices or entities. For example, the functions of each module described herein may be implemented by multiple physical entities, or the functions of multiple modules described herein may be implemented by a single physical entity. In addition, the features, components, elements, steps, etc. described in one embodiment are not limited to the embodiment, but may also be applied to other embodiments, such as replacing specific features, components, elements, steps, etc. in other embodiments, or combining them.

[0061] The scope of the present disclosure is not limited to the specific embodiments described herein. It should be understood by those skilled in the art that, depending on design requirements and other factors, various modifications or changes may be made to the embodiments herein without departing from the principles of the present disclosure. The scope of the present disclosure is defined by the appended claims and their equivalents.

[0062] In addition, the present disclosure may also include the following embodiments.

[0063] (1) A method for generating an advertising script using a large language model, comprising:

[0064] The large language model receives a first prompt from a user, and generates a task list based on the first prompt, wherein the task list includes descriptions of multiple steps, and the multiple steps are at least related to identifying features and advantages of a product;

[0065] Searching for information related to the product on the Internet based on keywords extracted from the description contents of the plurality of steps, and establishing a local knowledge base using the searched information;

[0066] The large language model receives content from the local knowledge base, and generates a report based on the received content, wherein the report at least includes features and advantages of the product;

[0067] An advertisement script for the product is generated by the large language model based on the report.

[0068] (2) The method according to (1), wherein the first prompt includes one or more of the following:

[0069] Definition of a goal to be achieved by the task list, wherein the goal is to generate the report;

[0070] Definitions about the steps in the task list;

[0071] Definitions on the format of the task list;

[0072] Definition of the role of the large language model.

[0073] (3) The method according to (1), wherein the plurality of steps include:

[0074] Steps to collect information about the product in question and its competitors;

[0075] Steps to identify the strengths and weaknesses of said competitor;

[0076] Steps to compare the product with the competitors to identify the features and advantages of the product;

[0077] Steps to draw conclusions.

[0078] (4) The method according to (1) further includes: using information related to the product provided by the owner of the product to establish the local knowledge base.

[0079] (5) The method according to (1), further comprising:

[0080] For each of the multiple steps, calculate the similarity between each document in the local knowledge base and the description content of the step, input one or more documents with similarity greater than a first threshold into the large language model, and generate content corresponding to the step based on the input documents by the large language model,

[0081] The generated contents respectively corresponding to the plurality of steps are combined together to form the report.

[0082] (6) The method according to (5), further comprising:

[0083] In the case where a document in the local knowledge base cannot be input into the large language model, segmenting the document into a plurality of sub-documents, each sub-document having a size that enables it to be input into the large language model;

[0084] Vectorize each sub-document;

[0085] For each of the multiple steps, the description content of the step is vectorized, and the similarity between the vector representation of the description content of the step and the vector representation of each sub-document is calculated, and one or more sub-documents with similarity greater than a second threshold are input into the large language model, and the large language model generates content corresponding to the step based on the input sub-documents,

[0086] The generated contents respectively corresponding to the plurality of steps are combined together to form the report.

[0087] (7) The method according to (1), further comprising:

[0088] The large language model receives a second prompt from the user, where the second prompt includes the user's requirement for an advertisement script to be generated;

[0089] generating an advertisement script for the product based on the report and the second prompt by the large language model,

[0090] The generated advertising script includes product features described in the form of keywords.

[0091] (8) The method according to (7), further comprising:

[0092] Splitting the report into a plurality of segments, and vectorizing each segment;

[0093] vectorizing the second prompt;

[0094] Calculating the similarity between the vector representation of the second prompt and the vector representation of each segment of the report;

[0095] inputting one or more segments having a similarity greater than a third threshold into the large language model;

[0096] The advertisement script is generated by the large language model based on the input segment and the second prompt.

[0097] (9) A device for generating an advertising script, the device communicating with a large language model and comprising:

[0098] a memory storing a program; and

[0099] One or more processors, the processors being configured to perform the following operations by executing the program:

[0100] Inputting a first prompt of a user into the large language model;

[0101] Instructing the large language model to generate a task list according to the first prompt, wherein the task list includes descriptions of multiple steps, and the multiple steps are at least related to identifying features and advantages of a product;

[0102] Searching for information related to the product on the Internet based on keywords extracted from the description contents of the plurality of steps, and establishing a local knowledge base using the searched information;

[0103] Inputting the content in the local knowledge base into the large language model;

[0104] Instructing the large language model to generate a report based on the content in the local knowledge base, wherein the report at least includes features and advantages of the product;

[0105] The large language model is instructed to generate an advertisement script for the product based on the report.

[0106] (10) The device according to (9), wherein the first prompt includes one or more of the following:

[0107] Definition of a goal to be achieved by the task list, wherein the goal is to generate the report;

[0108] Definitions about the steps in the task list;

[0109] Definitions on the format of the task list;

[0110] Definition of the role of the large language model.

[0111] (11) The apparatus according to (9), wherein the plurality of steps comprises:

[0112] Steps to collect information about the product in question and its competitors;

[0113] Steps to identify the strengths and weaknesses of said competitor;

[0114] Steps to compare the product with the competitors to identify the features and advantages of the product;

[0115] Steps to draw conclusions.

[0116] (12) The apparatus according to (9), wherein the operation further comprises: establishing the local knowledge base using information related to the product provided by an owner of the product.

[0117] (13) The apparatus according to (9), wherein the operation further comprises:

[0118] For each of the multiple steps, calculate the similarity between each document in the local knowledge base and the description content of the step, input one or more documents with similarity greater than a first threshold into the large language model, and instruct the large language model to generate content corresponding to the step based on the input documents,

[0119] The large language model is instructed to combine the generated contents respectively corresponding to the plurality of steps to form the report.

[0120] (14) The apparatus according to (13), wherein the operation further comprises:

[0121] In the case where a document in the local knowledge base cannot be input into the large language model, segmenting the document into a plurality of sub-documents, each sub-document having a size that enables it to be input into the large language model;

[0122] Vectorize each sub-document;

[0123] For each of the multiple steps, vectorize the description content of the step, calculate the similarity between the vector representation of the description content of the step and the vector representation of each sub-document, input one or more sub-documents with similarity greater than a second threshold into the large language model, and instruct the large language model to generate content corresponding to the step based on the input sub-documents,

[0124] The large language model is instructed to combine the generated contents respectively corresponding to the plurality of steps to form the report.

[0125] (15) The apparatus according to (9), wherein the operation further comprises:

[0126] Inputting a second prompt of the user into the large language model, wherein the second prompt includes the user's requirement for the advertisement script to be generated;

[0127] instructing the large language model to generate an advertisement script for the product based on the report and the second prompt,

[0128] The generated advertising script includes product features described in the form of keywords.

[0129] (16) The apparatus according to (15), wherein the operation further comprises:

[0130] Splitting the report into a plurality of segments, and vectorizing each segment;

[0131] vectorizing the second prompt;

[0132] Calculating the similarity between the vector representation of the second prompt and the vector representation of each segment of the report;

[0133] inputting one or more segments having a similarity greater than a third threshold into the large language model;

[0134] The large language model is instructed to generate the advertisement script based on the input segment and the second prompt.

[0135] (17) A non-transitory computer-readable medium storing a program, wherein when the program is executed by a computer, the computer executes the method for generating an advertisement script according to any one of (1) to (8).

Claims

1. A method for generating an advertisement script using a large language model, comprising: The large language model receives a first prompt from a user, and generates a task list based on the first prompt, wherein the task list includes descriptions of multiple steps, and the multiple steps are at least related to identifying features and advantages of a product; Searching for information related to the product on the Internet based on keywords extracted from the description contents of the plurality of steps, and establishing a local knowledge base using the searched information; The large language model receives content from the local knowledge base, and generates a report based on the received content, wherein the report at least includes features and advantages of the product; An advertisement script for the product is generated by the large language model based on the report.

2. The method according to claim 1, wherein: The first prompt includes one or more of the following: Definition of a goal to be achieved by the task list, wherein the goal is to generate the report; Definitions about the steps in the task list; Definitions on the format of the task list; Definition of the role of the large language model.

3. The method according to claim 1, wherein: The multiple steps include: Steps to collect information about the product in question and its competitors; Steps to identify the strengths and weaknesses of said competitor; Steps to compare the product with the competitors to identify the features and advantages of the product; Steps to draw conclusions.

4. The method according to claim 1, further comprising: The local knowledge base is built using information related to the product provided by an owner of the product.

5. The method according to claim 1, further comprising: For each of the multiple steps, calculate the similarity between each document in the local knowledge base and the description content of the step, input one or more documents with similarity greater than a first threshold into the large language model, and generate content corresponding to the step based on the input documents by the large language model, The generated contents respectively corresponding to the plurality of steps are combined together to form the report.

6. The method according to claim 5, further comprising: In the case where a document in the local knowledge base cannot be input into the large language model, segmenting the document into a plurality of sub-documents, each sub-document having a size that enables it to be input into the large language model; Vectorize each sub-document; For each of the multiple steps, the description content of the step is vectorized, and the similarity between the vector representation of the description content of the step and the vector representation of each sub-document is calculated, and one or more sub-documents with similarity greater than a second threshold are input into the large language model, and the large language model generates content corresponding to the step based on the input sub-documents, The generated contents respectively corresponding to the plurality of steps are combined together to form the report.

7. The method according to claim 1, further comprising: The large language model receives a second prompt from the user, where the second prompt includes the user's requirement for an advertisement script to be generated; generating an advertisement script for the product based on the report and the second prompt by the large language model, The generated advertising script includes product features described in the form of keywords.

8. The method according to claim 7, further comprising: Splitting the report into a plurality of segments, and vectorizing each segment; vectorizing the second prompt; Calculating the similarity between the vector representation of the second prompt and the vector representation of each segment of the report; inputting one or more segments having a similarity greater than a third threshold into the large language model; The advertisement script is generated by the large language model based on the input segment and the second prompt.

9. An apparatus for generating an advertising script, the apparatus being in communication with a large language model and comprising: A memory storing a program; as well as One or more processors, the processors being configured to perform the following operations by executing the program: Inputting a first prompt of a user into the large language model; Instructing the large language model to generate a task list according to the first prompt, wherein the task list includes descriptions of multiple steps, and the multiple steps are at least related to identifying features and advantages of a product; Searching for information related to the product on the Internet based on keywords extracted from the description contents of the plurality of steps, and establishing a local knowledge base using the searched information; Inputting the content in the local knowledge base into the large language model; Instructing the large language model to generate a report based on the content in the local knowledge base, wherein the report at least includes features and advantages of the product; The large language model is instructed to generate an advertisement script for the product based on the report.

10. A non-transitory computer-readable medium storing a program, wherein when the program is executed by a computer, the computer executes the method for generating an advertisement script according to any one of claims 1 to 8.

Images