Email marketing title generation method and its device, equipment and medium

Abstract

The present application relates to a method for generating an email marketing title in the field of e-commerce technology and its device, equipment, and medium, the method comprising: inferring multiple title generation templates based on at least one sample example, the sample example comprising a keyword set and an email marketing title corresponding to its semantics; replacing the keyword set in each title generation template with the same keyword set to reconstruct multiple first title generation templates, generating a first email marketing title corresponding to each first title generation template; determining the quality score of each first email marketing title, screening out the first email marketing title with a quality score that meets the preset conditions and reconstructing the sample example with its first title generation template, iterating the above process at least once; obtaining the target keyword set to replace the first title generation template corresponding to the first email marketing title with the highest final quality score, reconstructing the second title generation template to generate its corresponding target email marketing title. In this way, an email marketing title that meets marketing expectations can be produced.

Description

Technical Field

[0001] The present application relates to the field of e-commerce technology, and in particular to a method for generating an email marketing title and its corresponding device, computer equipment, and computer-readable storage medium. Background Art

[0002] In today's highly competitive e-commerce market, email marketing has become an important promotion method for many e-commerce companies. Email marketing is a low-cost promotion method that sends marketing emails to potential or existing customers, which can continuously enhance brand awareness, increase product exposure, attract potential customers, and guide existing customers to purchase.

[0003] In traditional technology, email marketing titles are usually generated based on a small number of keywords. This implementation method usually makes it difficult to ensure the quality of the titles, resulting in a high probability of failing to attract users to open the email to learn more detailed information, and failing to achieve beneficial marketing results.

[0004] In view of the shortcomings of traditional technology, the applicant has been engaged in research in related fields for a long time, and has taken a different approach to solve industry problems in the e-commerce field. Summary of the invention

[0005] The primary purpose of the present application is to solve at least one of the above problems and provide a method for generating an email marketing title and its corresponding device, computer equipment, and computer-readable storage medium.

[0006] In order to meet the various objectives of this application, this application adopts the following technical solutions:

[0007] A method for generating an email marketing title adapted to one of the purposes of this application includes the following steps:

[0008] Inferring multiple title generation templates based on at least one sample example, the sample example includes a keyword set and an email marketing title corresponding to its semantics, the title generation template includes a general qualified description of the title to be generated and a keyword set in the sample example, the keyword set includes a plurality of preset keywords;

[0009] Replacing the keyword set in each title generation template with the same keyword set to reconstruct multiple first title generation templates, and generating a first email marketing title corresponding to each first title generation template;

[0010] Determine the quality score of each first email marketing title, select the first email marketing titles whose quality scores meet the preset conditions and reconstruct the sample examples with their first title generation templates, and iterate the above process at least once;

[0011] A target keyword set is obtained to replace the first title generation template corresponding to the first email marketing title with the highest final quality score, a second title generation template is reconstructed, and a target email marketing title of the second title generation template is generated, wherein the target keyword set includes multiple keywords input by the user, or multiple keywords describing the product specified by the user.

[0012] A further embodiment, inferring multiple title generation templates based on at least one sample example, comprises the following steps:

[0013] Using the language model as a reasoner, based on a set of keywords in at least one sample example and the email marketing title corresponding to their semantics, extract multiple deep semantic information representing a limited description of the corresponding title;

[0014] Multiple title generation templates are generated based on multiple deep semantic information.

[0015] In a further embodiment, determining the quality score of each first email marketing title includes the following steps:

[0016] Use the language model as a scorer and use the thinking chain method to guide the scorer to score each first email marketing title from multiple quality dimensions to obtain the corresponding score for each quality dimension;

[0017] The quality score is obtained by combining the scores corresponding to all quality dimensions.

[0018] A further embodiment includes the following steps before obtaining the target keyword set:

[0019] Get the product title of the product specified by the user;

[0020] A preset named entity recognition model is used to identify multiple target keywords in the product title and construct a target keyword set.

[0021] In a further embodiment, generating a target email marketing title of a second title generation template comprises the following steps:

[0022] Using the language model as a generator to generate a plurality of second email marketing titles of a second title generation template, wherein the plurality of second email marketing titles have similar semantics;

[0023] The language model is used as a scorer to score each second email marketing title, to obtain a quality score corresponding to each second email marketing title, and to screen out target email marketing titles whose quality scores meet preset conditions.

[0024] A further embodiment, before inferring multiple title generation templates based on at least one sample example, includes the following steps:

[0025] Obtain multiple keyword sets and email marketing titles corresponding to the semantics of each keyword set;

[0026] The email marketing titles corresponding to the keyword set and their semantics constitute sample examples.

[0027] A further embodiment, after obtaining multiple keyword sets and the email marketing titles semantically corresponding to each keyword set, includes:

[0028] The language model is used as a generator to generate multiple similar email marketing titles corresponding to each email marketing title, and the similar email marketing titles are semantically similar to the corresponding email marketing title.

[0029] On the other hand, an email marketing title generation device provided to meet one of the purposes of the present application includes a generation template reasoning module, a first title generation module, a sample example reconstruction module and a target title determination module, wherein the generation template reasoning module is used to infer multiple title generation templates based on at least one sample example, the sample example includes a keyword set and an email marketing title corresponding to its semantics, the title generation template includes a general limited description of the title to be generated and a keyword set in the sample example, and the keyword set includes a plurality of preset keywords; the first title generation module is used to replace the keyword set in each title generation template with the same keyword set to reconstruct multiple first title generation templates, and generate a first email marketing title corresponding to each first title generation template; the sample example reconstruction module is used to determine the quality score of each first email marketing title, screen out the first email marketing title whose quality score meets the preset conditions and reconstruct the sample example with its first title generation template, and iterate the above process at least once; the target title determination module is used to obtain a target keyword set to replace the first title generation template corresponding to the first email marketing title with the highest final quality score, reconstruct the second title generation template, and generate a target email marketing title for the second title generation template, wherein the target keyword set includes multiple keywords input by the user, or multiple keywords describing the product specified by the user.

[0030] In a further embodiment, the template generation reasoning module includes: an encoding stage sub-module, which is used to use the language model as a reasoner to extract multiple deep semantic information representing the limited description of the generated corresponding title based on a keyword set in at least one sample example and the email marketing title corresponding to its semantics; a decoding stage sub-module, which is used to generate multiple title generation templates based on the multiple deep semantic information.

[0031] In a further embodiment, the sample example reconstruction module includes: a multi-dimensional scoring sub-module, which is used to use the language model as a scorer, and adopt a thinking chain method to guide the scorer to score each first email marketing title from multiple quality dimensions to obtain a score corresponding to each quality dimension; a comprehensive scoring sub-module, which is used to integrate the scores corresponding to all quality dimensions to obtain a quality score.

[0032] In a further embodiment, the target title determination module includes: a title acquisition submodule for acquiring the product title of the product specified by the user; and a keyword recognition submodule for using a preset named entity recognition model to recognize multiple target keywords in the product title and construct a target keyword set.

[0033] In a further embodiment, the target title determination module includes: a first similar title generation sub-module, used to use the language model as a generator to generate multiple second email marketing titles of the second title generation template, and the multiple second email marketing titles have similar semantics; a title selection sub-module, used to use the language model as a scorer to score each second email marketing title, obtain a quality score corresponding to each second email marketing title, and screen out target email marketing titles whose quality scores meet preset conditions.

[0034] A further embodiment, before generating the template inference module, includes: a data acquisition submodule, used to acquire multiple keyword sets and email marketing titles corresponding to the semantics of each keyword set; a sample example construction submodule, used to construct sample examples from the keyword sets and the email marketing titles corresponding to their semantics.

[0035] A further embodiment, after obtaining the data acquisition submodule, includes: a second similar title generation submodule, which is used to use the language model as a generator to generate multiple similar email marketing titles corresponding to each email marketing title, and the similar email marketing titles are semantically similar to the corresponding email marketing title.

[0036] On the other hand, a computer device provided to meet one of the purposes of the present application includes a central processing unit and a memory, wherein the central processing unit is used to call and run a computer program stored in the memory to execute the steps of the email marketing title generation method described in the present application.

[0037] On the other hand, a computer-readable storage medium is provided to meet another purpose of the present application, which stores a computer program implemented according to the email marketing title generation method in the form of computer-readable instructions, and when the computer program is called and executed by a computer, it executes the steps included in the method.

[0038] The technical solution of this application has many advantages, including but not limited to the following aspects:

[0039] The present application infers multiple title generation templates based on at least one sample example, wherein the sample example includes a keyword set and an email marketing title corresponding to its semantics, and the title generation template includes a general limited description of the title to be generated and a keyword set in the sample example, and replaces the keyword set in each title generation template with the same keyword set to reconstruct multiple first title generation templates, generates a first email marketing title corresponding to each first title generation template, determines the quality score of each first email marketing title, selects the first email marketing title with a quality score that meets the preset conditions and reconstructs the sample example with its first title generation template, iterates the above process at least once, obtains the target keyword set to replace the first title generation template corresponding to the first email marketing title with the highest final quality score, and reconstructs the second title generation template to generate its corresponding target email marketing title. On the one hand, the title generation template is continuously optimized through iteration, so that after the high-quality first title generation template is finally obtained to replace the target keyword set, an email marketing title that can produce a beneficial marketing effect is generated, thereby attracting users who receive the corresponding email to open the email. On the other hand, the obtained high-quality first title generation template has universality, and can be universal by replacing the required keyword set. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0041] Figure 1 This is a flowchart of a typical embodiment of the method for generating an email marketing title of the present application;

[0042] Figure 2 A schematic diagram of a process of inferring a title generation template based on at least one sample example in an embodiment of the present application;

[0043] Figure 3 A schematic diagram of a process for determining a quality score of a first email marketing title in an embodiment of the present application;

[0044] Figure 4 A schematic diagram of a process for constructing a target keyword set in an embodiment of the present application;

[0045] Figure 5 A schematic diagram of a process for generating high-quality targeted email marketing titles in an embodiment of the present application;

[0046] Figure 6 A schematic diagram of a flow chart for constructing a sample example in an embodiment of the present application;

[0047] Figure 7 This is a functional block diagram of the device for generating email marketing titles of the present application;

[0048] Figure 8 A schematic diagram of the structure of a computer device used in this application. DETAILED DESCRIPTION

[0049] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present application.

[0050] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0051] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

[0052] It will be understood by those skilled in the art that the "client", "terminal" and "terminal device" used herein include both devices with wireless signal receivers, which are devices with only wireless signal receivers without transmission capabilities, and devices with receiving and transmitting hardware, which are devices with receiving and transmitting hardware capable of two-way communication on a two-way communication link. Such devices may include: cellular or other communication devices such as personal computers, tablet computers, which have single-line displays or multi-line displays or cellular or other communication devices without multi-line displays; PCS (Personal Communications Service, personal communication system), which can combine voice, data processing, fax and / or data communication capabilities; PDA (Personal Digital Assistant, personal digital assistant), which may include a radio frequency receiver, pager, Internet / intranet access, web browser, notepad, calendar and / or GPS (Global Positioning System, global positioning system) receiver; conventional laptop and / or palmtop computers or other devices, which have and / or include a conventional laptop and / or palmtop computer or other device with and / or including a radio frequency receiver. The "client", "terminal" and "terminal device" used herein may be portable, transportable, installed in a vehicle (air, sea and / or land), or suitable and / or configured to run locally, and / or in a distributed form, at any other location on the earth and / or in space. The "client", "terminal" and "terminal device" used herein may also be a communication terminal, an Internet terminal, a music / video playback terminal, such as a PDA, a MID (Mobile Internet Device) and / or a mobile phone with a music / video playback function, or a smart TV, a set-top box and other devices.

[0053] The hardware referred to by the names such as "server", "client", and "service node" in this application is essentially an electronic device with capabilities equivalent to those of a personal computer. It is a hardware device that has the necessary components revealed by the von Neumann principle, such as a central processing unit (including an arithmetic unit and a controller), a memory, an input device, and an output device. The computer program is stored in its memory, and the central processing unit calls the program stored in the external memory into the internal memory for execution, executes the instructions in the program, and interacts with the input and output devices to complete specific functions.

[0054] It should be pointed out that the concept of "server" referred to in this application can also be expanded to apply to server clusters. According to the network deployment principle understood by those skilled in the art, the servers should be logically divided. In physical space, these servers can be independent of each other but can be called through interfaces, or integrated into a physical computer or a set of computer clusters. Those skilled in the art should understand this flexibility, and should not use it to restrict the implementation of the network deployment method of this application.

[0055] Unless expressly specified, one or more technical features of the present application can be deployed on a server for implementation and accessed by a client through a remote call to obtain an online service interface provided by the server, or can be directly deployed and run on a client for access.

[0056] The neural network models referenced or may be referenced in this application, unless expressly specified, can be deployed on a remote server and remotely called on the client, or can be deployed and directly called on a client with sufficient device capabilities. In some embodiments, when it runs on the client, its corresponding intelligence can be obtained through transfer learning to reduce the requirements for the client's hardware operating resources and avoid excessive occupation of the client's hardware operating resources.

[0057] Unless explicitly specified, the various data involved in this application can be stored remotely on a server or on a local terminal device, as long as it is suitable for being called by the technical solution of this application.

[0058] Those skilled in the art should be aware that, although the various methods of the present application are described based on the same concept and thus present commonality to each other, unless otherwise specified, these methods can be independently executed. Similarly, for each embodiment disclosed in the present application, they are all proposed based on the same inventive concept, therefore, concepts with the same expression, and concepts that are appropriately changed for convenience despite different expressions, should be understood as equivalent.

[0059] Unless the mutually exclusive relationship between the embodiments to be disclosed in this application is explicitly stated, the relevant technical features involved in each embodiment can be cross-combined to flexibly construct a new embodiment, as long as such combination does not deviate from the creative spirit of this application and can meet the needs of the prior art or solve certain deficiencies in the prior art. Those skilled in the art should be aware of this flexibility.

[0060] A method for generating an email marketing title in the present application can be programmed as a computer program product and deployed in a client or server for execution. For example, in the exemplary application scenario of the present application, it can be deployed and implemented in a server of an e-commerce platform, thereby accessing an interface opened after the computer program product is run and performing human-computer interaction with the process of the computer program product through a graphical user interface to execute the method.

[0061] See also Figure 1 The email marketing title generation method of the present application, in its typical embodiment, comprises the following steps:

[0062] Step S1100: inferring a plurality of title generation templates based on at least one sample example, wherein the sample example includes a keyword set and an email marketing title corresponding to its semantics, wherein the title generation template includes a general qualified description of the title to be generated and a keyword set in the sample example, wherein the keyword set includes a plurality of preset keywords;

[0063] A service of generating email marketing titles based on a small number of keywords can be provided to users of e-commerce enterprises. Specifically, when users provide a small number of keywords, such as 2-5 keywords, these keywords can be used as the basis to generate email titles with marketing nature that correspond to the semantics of these keywords, namely, email marketing titles.

[0064] Usually, for marketing emails sent by e-commerce companies, it is possible to track whether the emails are opened by the customers who receive the emails and whether the e-commerce links in the email content are clicked by the customers who receive the emails. The e-commerce links can be website access links of e-commerce online stores or access links to product display pages of e-commerce online stores. It is not difficult to understand that the emails can be opened by the customers who receive the emails, and even the e-commerce links in the email content can be clicked by the customers who receive the emails after opening. These customer behaviors are the beneficial marketing effects that e-commerce companies hope to achieve by sending emails to customers. However, the triggering of these customer behaviors is primarily due to the fact that the titles of the emails, i.e., the email marketing titles, are attractive enough to attract customers. Accordingly, the emails corresponding to the emails opened by the customers who receive the emails and / or the e-commerce links in the email content are clicked by the customers who receive the emails are collected, the email marketing titles of these emails are obtained, and the multiple keywords based on which the email marketing titles are generated are used as presets to form a keyword set. Further, a single keyword set and a single email marketing title corresponding to its semantics constitute the sample example.

[0065] The language model is used as a reasoner, and a few-shot method is adopted. Based on at least one sample example, with the help of a thruster, it is inferred how the email marketing title in the sample example should be limited, and the corresponding limited description is output as a title generation template. Specifically, the thruster encodes the keyword set in at least one sample example and the email marketing title corresponding to its semantics, and correspondingly extracts multiple deep semantic information representing the limited description of the corresponding title. For the deep semantic information represented by a single vectorization, the general limited description of the single title and the keywords in the sample example are generated. It can be understood that the general limited description has universality and can be any one or more of limiting the number of words in the generated title, limiting the effect that the generated title wants to achieve, such as attracting the attention of the reader of the title to stimulate his action, limiting the scene of the generated title to adapt to the scene to generate the corresponding title, etc. The keywords in the sample example can be understood as more specific limited descriptions of the generated title relative to the general limited description. Further, the deep semantic information is decoded to generate corresponding multiple description texts as the title generation template.

[0066] The language model is a large language model suitable for the field of NLP. It is pre-trained with an extremely large corpus until convergence, acquires the ability to generate human language, and has accurate text semantic understanding and logical reasoning capabilities. The language model can be selected from the GPT series model, the Bert model, the Encoder-Decoder model, the Transformer model, etc., and those skilled in the art can select one as needed.

[0067] Step S1200: Replacing the keyword set in each title generation template with the same keyword set to reconstruct multiple first title generation templates, and generating a first email marketing title corresponding to each first title generation template;

[0068] A keyword set in a single sample example is selected to replace the keyword set in each title generation template, so that the specific limiting description of the title generated by each title generation template is the same keyword set. On this basis, the quality of the titles generated by the corresponding different general limiting descriptions is examined, and accordingly, the corresponding multiple first title generation templates are obtained.

[0069] The language model is used as a generator, and a single first title generation template is used as an input, the first title generation template is encoded, and the corresponding vectorized deep semantic information is extracted, wherein the deep semantic information represents a limited description of the generated title. Further, the deep semantic information is decoded to generate a corresponding first email marketing title.

[0070] Step S1300: determine the quality score of each first email marketing title, select the first email marketing titles with quality scores that meet the preset conditions and reconstruct the sample examples with their first title generation templates, and iterate the above process at least once;

[0071] The language model is used as a scorer, and a single first email marketing title is used as input. The first email marketing title is encoded to extract the corresponding vectorized deep semantic information, which represents the score corresponding to the title based on different quality dimensions of marketing. The different quality dimensions can be any number such as semantic richness, attractiveness, and fluency of expression. Those skilled in the art can flexibly set them according to business needs. Further, the deep semantic information is decoded to generate a quality score that summarizes the sum of the scores corresponding to all quality dimensions.

[0072] In one embodiment, a first email marketing title with a quality score higher than a preset threshold is screened out, and a single sample example is reconstructed using a single first email marketing title and a first title generation template that generates the title. The preset threshold can be flexibly set by a person skilled in the art based on prior knowledge or experimental data.

[0073] In another embodiment, the first email marketing titles are sorted in descending order according to their quality scores, and the top N first email marketing titles are screened out. A single sample example is reconstructed with a single first email marketing title and a first title generation template for generating the title. The N can be flexibly set by those skilled in the art according to business needs.

[0074] Further, all operations of steps S1100-1300 are iterated at least once. It can be understood that based on continuous iteration and from a result-oriented perspective, the quality of the generated first email marketing title is continuously improved, that is, the general qualified description in the first title generation template is continuously optimized, and the first title generation template capable of generating high-quality titles can be effectively obtained.

[0075] Step S1400, obtaining a target keyword set to replace the first title generation template corresponding to the first email marketing title with the highest final quality score, reconstructing a second title generation template, and generating a target email marketing title for the second title generation template, wherein the target keyword set includes multiple keywords input by the user, or multiple keywords describing the product specified by the user.

[0076] When users of e-commerce companies use the service of generating email marketing titles, in one embodiment, users can enter a small number of keywords, such as 2-5 keywords, and use these keywords as the basis for generating titles, thereby obtaining these keywords to form a target keyword set. In another embodiment, users can specify a product, thereby using multiple keywords in the product title of the product as the basis for generating titles, thereby obtaining these keywords to form a target keyword set. The determination of keywords in the product title can be flexibly implemented by those skilled in the art, for example, by using a term weight algorithm, a rule matching algorithm, a named entity recognition algorithm, etc.

[0077] The target keyword set is used to replace the keyword set in the first title generation template corresponding to the first email marketing title with the highest final quality score, to obtain a second title generation template. The language model is used as a generator, and a single second title generation template is used as input to encode the second title generation template, and extract the corresponding vectorized deep semantic information, wherein the deep semantic information represents the limited description of the generated title. Further, the deep semantic information is decoded to generate the corresponding target email marketing title.

[0078] It can be known from the typical embodiments of the present application that the technical solution of the present application has many advantages, including but not limited to the following aspects:

[0079] The present application infers multiple title generation templates based on at least one sample example, wherein the sample example includes a keyword set and an email marketing title corresponding to its semantics, and the title generation template includes a general limited description of the title to be generated and a keyword set in the sample example, and replaces the keyword set in each title generation template with the same keyword set to reconstruct multiple first title generation templates, generates a first email marketing title corresponding to each first title generation template, determines the quality score of each first email marketing title, selects the first email marketing title with a quality score that meets the preset conditions and reconstructs the sample example with its first title generation template, iterates the above process at least once, obtains the target keyword set to replace the first title generation template corresponding to the first email marketing title with the highest final quality score, and reconstructs the second title generation template to generate its corresponding target email marketing title. On the one hand, the title generation template is continuously optimized through iteration, so that after the high-quality first title generation template is finally obtained to replace the target keyword set, an email marketing title that can produce a beneficial marketing effect is generated, thereby attracting users who receive the corresponding email to open the email. On the other hand, the obtained high-quality first title generation template has universality, and can be universal by replacing the required keyword set.

[0080] See also Figure 2In a further embodiment, step S1100, inferring multiple title generation templates based on at least one sample example, comprises the following steps:

[0081] Step S1110: Using the language model as a reasoner, based on a keyword set in at least one sample example and an email marketing title corresponding to its semantics, extract a plurality of deep semantic information representing a limited description of the corresponding title;

[0082] A few-shot method is used to edit a few-shot prompt text with at least one sample example and a task description. The task description is used to instruct the language model to describe the restrictions required to generate a title based on at least one sample example (the restrictions can also be understood as requirements). Those skilled in the art can flexibly set the task description based on the disclosure herein. For easy understanding, an exemplary example is given below:

[0083] A few sample prompts were compiled with three sample examples and task descriptions: “Think of a test paper. Here are some questions and corresponding answers. How should each question be described?

[0084] Topic 1: Men's clothing, Thanksgiving, big sales

[0085] Answer 1: Men's Thanksgiving Carnival!

[0086] Topic 2: Women's clothing, Thanksgiving, promotion

[0087] Answer 2: Women’s clothing feast, Thanksgiving shopping season!

[0088] Topic 3: Home appliances, Black Friday, promotions

[0089] Answer 3: Black Friday SALE! Home appliances are waiting for you to buy! "

[0090] Among them, "Think of a test question. Here are some questions and corresponding answers. How should each question be described better?" is the task description, and a single question (keyword set) and its single answer (email marketing title corresponding to the semantics of the keyword set) are sample examples.

[0091] In one embodiment, the language model adopts a GPT-4 model, takes the few sample prompt text as the input of the model, performs word segmentation on it, and obtains a corresponding word segmentation sequence. Then, the word segmentation sequence is input into the encoding end of the model, and each word segmentation in the word segmentation sequence is encoded by stacking multiple layers of multi-head self-attention layers and fully connected layers. Specifically, for each word segmentation, a multi-head attention calculation is performed on it when passing through the multi-head attention layer, so that self-attention is weighted on different dimensions of the word segmentation to obtain a corresponding weighted vector representation, and then after passing through the fully connected layer, the encoding vector representation corresponding to the word segmentation is obtained, that is, multiple deep semantic information representing the limited description of the corresponding title is extracted, and the deep semantic information of a single vectorized representation represents the general limited description of the single title and the keywords in the sample example. It can be understood that the general limited description has universality, which can be any one or more of limiting the number of words in the generated title, limiting the effect of the generated title, such as attracting the attention of the reader of the title to stimulate his action, limiting the scene of the generated title to adapt to the scene to generate the corresponding title, etc. The keywords in the sample example can be understood as a more specific limited description of the generated title relative to the general limited description.

[0092] Step S1120: Generate multiple title generation templates based on multiple deep semantic information.

[0093] Furthermore, the encoding representation corresponding to each word in the word segmentation sequence is input into the decoding end of the model, and the encoding representation corresponding to each word segmentation is decoded. Specifically, during the decoding process, the generation probability of each generated word is calculated according to the generated words, the current word position and the encoding representation corresponding to the word segmentation, and the word with the highest generation probability is selected. The words generated each time are sequentially spliced ​​to obtain a title generation template. For easy understanding, an exemplary example is given:

[0094] Title generation template 1: Based on the given keyword set, edit the email marketing title. The number of words in the title should not exceed 10 to ensure that it can attract readers and stimulate their interest in reading. Keyword set: men's clothing, Thanksgiving, big sale.

[0095] Title generation template 2: Your company is currently facing a major challenge to increase brand awareness and attention. You need to write an email marketing title based on a given set of keywords to ensure that they can attract customers' attention and inspire them to take action. Keyword set: women's clothing, Thanksgiving, promotion.

[0096] Title generation template 3: During Black Friday, you need to design an email marketing title based on a given set of keywords to ensure that the title attracts consumers' attention and stimulates their desire to buy. Keyword set: home appliances, Black Friday, promotion.

[0097] In this embodiment, by adopting a few-sample approach, the model is guided to infer the corresponding title generation template based on at least one sample example. The model's accurate text understanding and logical reasoning capabilities can be utilized to automatically derive the title generation template, thereby ensuring to a certain extent that the titles generated by it can produce beneficial marketing effects.

[0098] See also Figure 3 In a further embodiment, step S1300, determining the quality score of each first email marketing title, comprises the following steps:

[0099] Step S1310: Using the language model as a scorer, guiding the scorer to score each first email marketing title from multiple quality dimensions using a thought chain method, and obtaining a score corresponding to each quality dimension;

[0100] Using the chain of thought method, each first email marketing title and task description to be scored is used to edit the guidance prompt text. The task description is used to instruct the language model to think step by step, and the score corresponding to the title is evaluated from different quality dimensions of marketing. Those skilled in the art can flexibly set the task description according to the disclosure herein. The different quality dimensions can be any number such as semantic richness, attractiveness, and fluency of expression. For easy understanding, an exemplary example is given:

[0101] Edit the guidance text with the three first email marketing titles and task descriptions: "Q: Think step by step, analyze the following email marketing titles generated based on the same keyword set from the perspective of semantic richness, attractiveness, and expression fluency in marketing. Which one is most attractive to customers? The corresponding semantic richness, attractiveness, and expression fluency are all full marks of 10 points. Score accordingly, briefly explain the reasons, and output in json format.

[0102] Keyword set: "women's clothing, Thanksgiving, promotion"

[0103] Title 1. Women's Thanksgiving Carnival!

[0104] Title 2. Women's clothing feast, Thanksgiving shopping season!

[0105] Title 3. Thanksgiving exclusive: Huge discount on women's clothing! A great opportunity for high-cost shopping!

[0106] A: {Title 1 contains the keywords women's clothing and Thanksgiving, uses metaphors, and conveys the meaning of promotion, but the language is relatively bland. ”

[0107] Among them, except for titles 1, 2, and 3 which correspond to the corresponding first email marketing titles, the rest are task descriptions.

[0108] In one embodiment, the language model adopts the GPT-4 model, takes the guide prompt text as the input of the model, performs word segmentation on it, obtains the corresponding word segmentation sequence, and then inputs the word segmentation sequence into the encoding end of the model, and encodes each word segmentation in the word segmentation sequence by stacking multiple layers of multi-head self-attention layers and fully connected layers. Specifically, for each word segmentation, multi-head attention calculation is performed on it when passing through the multi-head attention layer, so as to perform self-attention weighting on different dimensions of the word segmentation, obtain the corresponding weighted vector representation, and then after passing through the fully connected layer, obtain the encoding vector representation corresponding to the word segmentation, that is, extract the score corresponding to the title of the different quality dimensions based on marketing. Further, the encoding representation corresponding to each word segmentation in the word segmentation sequence is input into the decoding end of the model, and the encoding representation corresponding to each word segmentation is decoded. Specifically, during the decoding process, the generation probability of each generated word is calculated according to the generated words, the current word position and the encoding representation corresponding to the word segmentation, and the word with the highest generation probability is selected. The words generated each time are sequentially spliced ​​to obtain the score corresponding to each quality dimension. For the convenience of understanding, the following exemplary examples are given:

[0109] {"Title 1": {"Semantic Richness Score": 5,"Attractiveness Score": 6,"Expression Fluency Score": 9,"Reason":"The title contains the keywords women's clothing and Thanksgiving, uses metaphors, conveys the meaning of promotion, and is fluent and easy to read, but the language is relatively plain."},

[0110] "Title 2": {"Semantic Richness Score": 7,"Attractiveness Score": 8,"Expression Fluency Score": 8,"Reason":"The title uses the metaphor of "feast", which expresses rich semantics, increases its appeal, and clarifies the theme and timing of the promotion, but its fluency is poor."},

[0111] "Title 3": {"Semantic Richness Score": 9,"Attractiveness Score": 9,"Expression Fluency Score": 9,"Reason":"The title contains the keywords women's clothing, Thanksgiving, and promotion, and uses specific words such as great discounts and high cost performance, which is more attractive. The expression fluency is higher than the first two titles, the semantic richness is the highest, and it is most likely to attract customers."}}

[0112] Step S1320: Comprehensively calculate the scores corresponding to all quality dimensions to obtain a quality score.

[0113] In one embodiment, taking a single first email marketing title as an example, the sum of the scores corresponding to all its quality dimensions is aggregated to obtain a quality score.

[0114] In another embodiment, taking a single first email marketing title as an example, a weight corresponding to each quality dimension is preset according to the different effects of each quality dimension on producing a beneficial marketing effect, and then the score corresponding to each quality dimension is matched with the corresponding weight, multiplied, and then added together to obtain a quality score.

[0115] In this embodiment, by adopting a thought chain approach, corresponding scores are determined based on multiple quality dimensions, and the scores corresponding to all quality dimensions are integrated to obtain quality scores, thereby ensuring an accurate evaluation of the beneficial marketing effects that the title can produce.

[0116] See also Figure 4 In a further embodiment, before step S1400, obtaining the target keyword set, the following steps are included:

[0117] Step S2400: Obtain the product title of the product specified by the user;

[0118] When a user of an e-commerce enterprise uses the service of generating a title for an email marketing, the user can specify a product, thereby triggering a title generation request to be sent to a server. The server receives the title generation request and responds to it, obtaining the product title of the product.

[0119] Step S2410: Use a preset named entity recognition model to identify multiple target keywords in the product title and construct a target keyword set.

[0120] The named entity recognition model is suitable for named entity recognition tasks. The specific model selection can be RoBERTa+CRF, BiLSTM+CRF, IDCNN+CRF, Bert+BiLSTM+CRF, FLAT, etc. The technical personnel in this field can select one selection as needed. After the named entity recognition model is trained to convergence, it can learn the ability to recognize the target keywords in the product title. In view of the fact that the training process of RoBERTa+CRF, BiLSTM+CRF, IDCNN+CRF, Bert+BiLSTM+CRF, and FLAT is known in the art, it will not be described in detail.

[0121] The target keywords may be any one or more of the categories, attributes, functions, effects, applicable scenarios, etc. of the product, and may be flexibly set by technicians in this field according to business requirements.

[0122] In this embodiment, by adopting the named entity recognition model, the target keywords in the product title can be quickly and accurately identified without user input, thereby improving the user experience.

[0123] See also Figure 5In a further embodiment, step S1400, generating a target email marketing title of a second title generation template, comprises the following steps:

[0124] Step S1410: Using the language model as a generator to generate a plurality of second email marketing titles of a second title generation template, wherein the plurality of second email marketing titles have similar semantics;

[0125] The instruction text is edited with the second title generation template and the task description. The task description is used to instruct the model to generate multiple second email marketing titles with the same semantics based on the second title generation template. Those skilled in the art can flexibly set the task description according to the disclosure herein. For ease of understanding, an exemplary example is given below:

[0126] Instruction text: Generate multiple email marketing titles with similar semantics based on the given title generation template. Title generation template: Your company is currently facing a major challenge and needs to increase brand awareness and attention. You need to edit the email marketing title based on the given keyword set. The title should not exceed 10 words to ensure that it can attract customers and stimulate their impulse to buy. Keyword set: men's clothing, Thanksgiving, big sale.

[0127] Among them: except for the title generation template which is the corresponding second title generation template, the rest are task descriptions.

[0128] In one embodiment, the language model adopts the GPT-4 model, and the second title generation template is used as the input of the model, and the word segmentation is performed to obtain the corresponding word segmentation sequence. Then, the word segmentation sequence is input to the encoding end of the model, and each word segmentation in the word segmentation sequence is encoded by stacking multiple layers of multi-head self-attention layers and fully connected layers. Specifically, for each word segmentation, a multi-head attention calculation is performed on it when passing through the multi-head attention layer, so that the self-attention weighting is performed on the different dimensions of the word segmentation to obtain the corresponding weighted vector representation, and then after passing through the fully connected layer, the encoding vector representation corresponding to the word segmentation is obtained, that is, the deep semantic information representing the semantics of the title to be generated is extracted. Further, the encoding representation corresponding to each word segmentation in the word segmentation sequence is input to the decoding end of the model, and the encoding representation corresponding to each word segmentation is decoded. Specifically, during the decoding process, the generation probability of each generated word is calculated according to the generated words and the current word position and the encoding representation corresponding to the word segmentation, and the word with the highest generation probability is selected. The words generated each time are sequentially spliced ​​to obtain multiple second email marketing titles with similar semantics. For the convenience of understanding, the following exemplary examples are given:

[0129] Title 1: Men's clothing prices are crazy reduced! Thanksgiving special offer!

[0130] Title 2: Popular men's clothing, extra discounts for Thanksgiving!

[0131] Title 3: Thanksgiving Men’s Clothing Sale!

[0132] Step S1420: Use the language model as a scorer to score each second email marketing title, obtain a quality score corresponding to each second email marketing title, and filter out target email marketing titles whose quality scores meet preset conditions.

[0133] Referring to steps S1310-1320, the quality score corresponding to each second email marketing title is determined. In one embodiment, the second email marketing title with the highest quality score is selected as the target email marketing title. In another embodiment, the second email marketing title is sorted from high to low according to the quality score of each second email marketing title, and the top N second email marketing titles are selected as the target email marketing titles, respectively. The N can be flexibly set by those skilled in the art according to business needs.

[0134] In this embodiment, by generating a plurality of second email marketing titles with similar semantics based on the second title generation template, a target email marketing title with better quality is selected to ensure that a high-quality title is obtained.

[0135] See also Figure 6 In a further embodiment, before step S1100, inferring multiple title generation templates based on at least one sample example, the following steps are included:

[0136] Step S1000: obtaining multiple keyword sets and email marketing titles semantically corresponding to each keyword set;

[0137] A service of generating email marketing titles based on a small number of keywords can be provided to users of e-commerce enterprises. Specifically, when users provide a small number of keywords, such as 2-5 keywords, these keywords can be used as the basis to generate email titles with marketing nature that correspond to the semantics of these keywords, namely, email marketing titles.

[0138] Usually, for marketing emails sent by e-commerce companies, it is possible to track whether the emails are opened by the customers who receive the emails and whether the e-commerce links in the email content are clicked by the customers who receive the emails. The e-commerce links can be website access links of e-commerce online stores or access links to product display pages of e-commerce online stores. It is not difficult to understand that the emails can be opened by the customers who receive the emails, and even the e-commerce links in the email content can be clicked by the customers who receive the emails after opening. These customer behaviors are the beneficial marketing effects that e-commerce companies hope to achieve by sending emails to customers. However, the triggering of these customer behaviors is primarily due to the fact that the titles of the emails, i.e., the email marketing titles, are attractive enough to attract customers. Accordingly, the corresponding emails opened by the customers who receive the emails and / or the e-commerce links in the email content are clicked by the customers who receive the emails are collected, and the email marketing titles of these emails are obtained, and multiple keywords based on which the email marketing titles are generated are used as presets to form a keyword set.

[0139] Step S1010: The keyword set and the email marketing titles corresponding to their semantics are combined into sample examples.

[0140] The sample example is composed of a single keyword set and a single email marketing title corresponding to its semantics.

[0141] In this embodiment, the construction process of the sample examples is explained, and a beneficial marketing effect is generated based on the email marketing title in the sample examples, so that a title generation template is subsequently derived based on at least one of the sample examples, which ensures that the title generated therefrom can produce a beneficial marketing effect to a certain extent.

[0142] In a further embodiment, after step S1000, obtaining multiple keyword sets and email marketing titles semantically corresponding to each keyword set, the method further includes:

[0143] Step S1001: Use the language model as a generator to generate multiple similar email marketing titles corresponding to each email marketing title, where the similar email marketing titles are semantically similar to the corresponding email marketing title.

[0144] In one embodiment, the language model adopts the GPT-4 model, takes the single email marketing title as the input of the model, performs word segmentation on it, obtains the corresponding word segmentation sequence, and then inputs the word segmentation sequence into the encoding end of the model, and encodes each word segmentation in the word segmentation sequence by stacking multiple layers of multi-head self-attention layers and fully connected layers. Specifically, for each word segmentation, multi-head attention calculation is performed on it when passing through the multi-head attention layer, so as to perform self-attention weighting on different dimensions of the word segmentation, obtain the corresponding weighted vector representation, and then after passing through the fully connected layer, obtain the encoding vector representation corresponding to the word segmentation, that is, extract the deep semantic information representing the semantics of the title to be generated. Further, the encoding representation corresponding to each word segmentation in the word segmentation sequence is input into the decoding end of the model, and the encoding representation corresponding to each word segmentation is decoded. Specifically, during the decoding process, the generation probability of each generated word is calculated according to the generated words, the current word position and the encoding representation corresponding to the word segmentation, and the word with the highest generation probability is selected. The words generated each time are sequentially spliced ​​to obtain multiple email marketing titles with similar semantics.

[0145] In this embodiment, a model is used to generate multiple email marketing titles with similar semantics, thereby reasonably enriching the number of titles, and then subsequently expanding the number of sample examples.

[0146] See also Figure 7, an email marketing title generation device provided to meet one of the purposes of the present application is a functional embodiment of the email marketing title generation method of the present application. On the other hand, the device is an email marketing title generation device provided to meet one of the purposes of the present application, including a generation template reasoning module 1100, a first title generation module 1200, a sample example reconstruction module 1300 and a target title determination module 1400, wherein the generation template reasoning module 1100 is used to infer multiple title generation templates based on at least one sample example, the sample example includes a keyword set and its semantically corresponding email marketing title, the title generation template includes a general limited description of the title to be generated and a keyword set in the sample example, the keyword set includes multiple preset keywords; the first title generation module 1200 is used to The keyword set in each title generation template is replaced with the same keyword set to reconstruct multiple first title generation templates, and a first email marketing title corresponding to each first title generation template is generated; a sample example reconstruction module 1300 is used to determine the quality score of each first email marketing title, screen out the first email marketing title whose quality score meets the preset conditions and reconstruct the sample example with its first title generation template, and iterate the above process at least once; a target title determination module 1400 is used to obtain a target keyword set to replace the first title generation template corresponding to the first email marketing title with the highest final quality score, reconstruct a second title generation template, and generate a target email marketing title for the second title generation template, wherein the target keyword set includes multiple keywords input by the user, or multiple keywords describing the product specified by the user.

[0147] In a further embodiment, the template generation reasoning module 1100 includes: an encoding stage sub-module, which is used to use the language model as a reasoner to extract multiple deep semantic information representing the limited description of the generated corresponding title according to the keyword set in at least one sample example and the email marketing title corresponding to its semantics; a decoding stage sub-module, which is used to generate multiple title generation templates based on the multiple deep semantic information.

[0148] In a further embodiment, the sample example reconstruction module 1300 includes: a multi-dimensional scoring sub-module, which is used to use the language model as a scorer, and adopt a thinking chain method to guide the scorer to score each first email marketing title from multiple quality dimensions to obtain a score corresponding to each quality dimension; a comprehensive scoring sub-module, which is used to integrate the scores corresponding to all quality dimensions to obtain a quality score.

[0149] In a further embodiment, the target title determination module 1400 includes: a title acquisition submodule for acquiring the product title of the product specified by the user; and a keyword recognition submodule for using a preset named entity recognition model to recognize multiple target keywords in the product title and construct a target keyword set.

[0150] In a further embodiment, the target title determination module 1400 includes: a first similar title generation submodule, used to use the language model as a generator to generate multiple second email marketing titles of the second title generation template, and the multiple second email marketing titles have similar semantics; a title selection submodule, used to use the language model as a scorer to score each second email marketing title, obtain a quality score corresponding to each second email marketing title, and screen out target email marketing titles whose quality scores meet preset conditions.

[0151] In a further embodiment, before the template inference module 1100 is generated, it includes: a data acquisition submodule, used to acquire multiple keyword sets and email marketing titles corresponding to the semantics of each keyword set; a sample example construction submodule, used to construct sample examples from the keyword sets and the email marketing titles corresponding to their semantics.

[0152] A further embodiment, after obtaining the data acquisition submodule, includes: a second similar title generation submodule, which is used to use the language model as a generator to generate multiple similar email marketing titles corresponding to each email marketing title, and the similar email marketing titles are semantically similar to the corresponding email marketing title.

[0153] In order to solve the above technical problems, the present application also provides a computer device. Figure 8 As shown, a schematic diagram of the internal structure of a computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. Among them, the computer-readable storage medium of the computer device stores an operating system, a database, and computer-readable instructions. The database may store a control information sequence. When the computer-readable instructions are executed by the processor, the processor can implement a method for generating an email marketing title. The processor of the computer device is used to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device may store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor can execute the email marketing title generation method of the present application. The network interface of the computer device is used to connect and communicate with a terminal. Those skilled in the art can understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0154] In this embodiment, the processor is used to execute Figure 7The memory stores the program code and various data required to execute the above modules or submodules. The network interface is used to transmit data between user terminals or servers. The memory in this embodiment stores the program code and data required to execute all modules / submodules in the email marketing title generation device of this application, and the server can call the program code and data of the server to execute the functions of all submodules.

[0155] The present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the email marketing title generation method of any embodiment of the present application.

[0156] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments of the present application can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a computer-readable storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0157] In summary, this application generates a title template through iterative optimization, thereby producing a title that meets the expected marketing effect.

[0158] Those skilled in the art will appreciate that the various operations, methods, steps, measures, and schemes in the processes discussed in this application may be alternated, altered, combined, or deleted. Further, other steps, measures, and schemes in the various operations, methods, and processes discussed in this application may also be alternated, altered, rearranged, decomposed, combined, or deleted. Further, the steps, measures, and schemes in the prior art that are similar to those disclosed in this application may also be alternated, altered, rearranged, decomposed, combined, or deleted.

[0159] The above description is only a partial implementation method of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for generating an email marketing title, characterized in that: The steps include: Inferring multiple title generation templates based on at least one sample example, the sample example includes a keyword set and an email marketing title corresponding to its semantics, the title generation template includes a general qualified description of the title to be generated and a keyword set in the sample example, the keyword set includes a plurality of preset keywords; Replacing the keyword set in each title generation template with the same keyword set to reconstruct multiple first title generation templates, and generating a first email marketing title corresponding to each first title generation template; Determine the quality score of each first email marketing title, select the first email marketing titles whose quality scores meet the preset conditions and reconstruct the sample examples with their first title generation templates, and iterate the above process at least once; Obtain a target keyword set to replace the first title generation template corresponding to the first email marketing title with the highest final quality score, reconstruct a second title generation template, and generate a target email marketing title of the second title generation template, wherein the target keyword set includes multiple keywords input by the user, or multiple keywords describing the product specified by the user; The inferring multiple title generation templates based on at least one sample example includes: Using the language model as a reasoner, based on a set of keywords in at least one sample example and the email marketing title corresponding to their semantics, extract multiple deep semantic information representing a limited description of the corresponding title; Multiple title generation templates are generated based on multiple deep semantic information.

2. The method for generating an email marketing title according to claim 1, characterized in that: Determine the quality score of each first email marketing title, including the following steps: Use the language model as a scorer and use the thinking chain method to guide the scorer to score each first email marketing title from multiple quality dimensions to obtain the corresponding score for each quality dimension; The quality score is obtained by combining the scores corresponding to all quality dimensions.

3. The method for generating an email marketing title according to claim 1, characterized in that: Before obtaining the target keyword set, the following steps are included: Get the product title of the product specified by the user; A preset named entity recognition model is used to identify multiple target keywords in the product title and construct a target keyword set.

4. The method for generating an email marketing title according to claim 1, characterized in that: Generating the target email marketing title of the second title generation template includes the following steps: Using the language model as a generator to generate a plurality of second email marketing titles of a second title generation template, wherein the plurality of second email marketing titles have similar semantics; The language model is used as a scorer to score each second email marketing title, to obtain a quality score corresponding to each second email marketing title, and to screen out target email marketing titles whose quality scores meet preset conditions.

5. The method for generating an email marketing title according to claim 1, characterized in that: Before inferring multiple title generation templates based on at least one sample example, the following steps are included: Obtain multiple keyword sets and email marketing titles corresponding to the semantics of each keyword set; The email marketing titles corresponding to the keyword set and their semantics constitute sample examples.

6. The method for generating an email marketing title according to claim 5, characterized in that: After obtaining multiple keyword sets and the email marketing titles corresponding to the semantics of each keyword set, including: The language model is used as a generator to generate multiple similar email marketing titles corresponding to each email marketing title, and the similar email marketing titles are semantically similar to the corresponding email marketing title.

7. A device for generating a title for email marketing, characterized in that: include: A template generation inference module is used to infer multiple title generation templates based on at least one sample example, wherein the sample example includes a keyword set and an email marketing title corresponding to its semantics, and the title generation template includes a general qualified description of the title to be generated and a keyword set in the sample example, wherein the keyword set includes a plurality of preset keywords; A first title generation module, used for replacing the keyword set in each title generation template with the same keyword set to reconstruct multiple first title generation templates, and generating a first email marketing title corresponding to each first title generation template; A sample example reconstruction module is used to determine the quality score of each first email marketing title, select the first email marketing title whose quality score meets the preset conditions and reconstruct the sample example with its first title generation template, and iterate the above process at least once; a target title determination module, configured to obtain a target keyword set to replace a first title generation template corresponding to a first email marketing title with the highest final quality score, reconstruct a second title generation template, and generate a target email marketing title of the second title generation template, wherein the target keyword set includes a plurality of keywords input by a user, or a plurality of keywords describing a product specified by a user; The inferring multiple title generation templates based on at least one sample example includes: Using the language model as a reasoner, based on a set of keywords in at least one sample example and the email marketing title corresponding to their semantics, extract multiple deep semantic information representing a limited description of the corresponding title; Multiple title generation templates are generated based on multiple deep semantic information.

8. A computer device comprising a central processing unit and a memory, characterized in that: The central processing unit is used to call and run the computer program stored in the memory to execute the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: It stores a computer program implemented according to the method described in any one of claims 1 to 6 in the form of computer-readable instructions, and when the computer program is called and executed by a computer, the steps included in the corresponding method are executed.

Images