Artificial intelligence method, device, equipment and storage medium for copywriting generation

Abstract

The present application provides an artificial intelligence method, device, equipment and storage medium for copywriting generation, which belongs to the field of data processing technology. The method includes: obtaining multiple traffic copywriting, their corresponding traffic values, several product topics of the advertised products and user browsing information, the user browsing information includes: several first historical browsing copywriting; determining the main copywriting topic based on all traffic copywriting and their corresponding traffic values; determining several second topics of the copywriting content to be generated based on the pre-constructed topic transfer description group, the main copywriting topic and all product topics, the pre-constructed topic transfer description group is used to describe the transfer probability between topics; obtaining the corresponding paragraph content of all second topics based on all second topics, all first historical browsing copywriting and the text generation model; generating the copywriting content to be generated based on the main copywriting topic, all second topics and their corresponding paragraph content. The present application can effectively improve the quality of the generated copywriting.

Description

Technical Field

[0001] The present application belongs to the field of data processing technology, and in particular relates to an artificial intelligence method, device, equipment and storage medium for copywriting generation. Background Art

[0002] With the development of the internet and digital media, the amount of information is exploding. In this fast-paced world, users' demand for personalized and high-quality content is also increasing, and copywriting has seen a new lease of life on online platforms. Copywriting, a graphic and textual presentation method and process for communicating creative ideas, is an indispensable component of marketing and promotional activities. Simultaneously, with the advancement and development of artificial intelligence, copywriting has also entered the digital and intelligent realm, resulting in the emergence of various new technologies and tools.

[0003] Currently, in order to expand product promotion efforts, relevant technologies will directly add product description information to the end of copy with high traffic to generate new copy content. However, when there is a large difference between the high-traffic copy content and the product, directly adding product description information may lead to unnatural transitions in the text. It can be seen that the quality of copy content generated by relevant technologies is poor. Summary of the Invention

[0004] In order to solve the above technical problems, the present application provides an artificial intelligence method, device, equipment and storage medium for copy generation.

[0005] In the first aspect, this application provides an artificial intelligence method for copywriting generation, which adopts the following technical solutions:

[0006] An artificial intelligence method for copywriting generation, comprising:

[0007] Acquire multiple traffic copywriting, their corresponding traffic values, several product topics of the advertised products, and user browsing information, wherein the user browsing information includes: several first historical browsing copywriting;

[0008] Determine a main copy topic based on all the traffic copies and their corresponding traffic values;

[0009] Determining a plurality of second topics for copy content to be generated based on a pre-constructed topic transition description group, the main copy topic, and all the product topics, wherein the pre-constructed topic transition description group is used to describe the transition probability between topics;

[0010] Obtaining text content corresponding to all of the second topics according to all of the second topics, all of the first historical browsing texts, and a text generation model;

[0011] Generate the to-be-generated copy content according to the main copy topic, all the second topics, and the corresponding paragraph contents;

[0012] The step of constructing the topic transfer description group includes:

[0013] Obtaining the corresponding occurrence frequencies of a plurality of second historical browsing texts and all the second historical text topics within a preset time period;

[0014] Determining a target second historical copy topic and a corresponding same-frequency topic from all the second historical copy topics, and mapping the target second historical copy topic to a first topic element, and mapping the same-frequency topic to a second topic element;

[0015] Obtaining the browsing time corresponding to each of the second historical documents, and generating a topic sequence according to all the browsing times, the first topic element, and the second topic element;

[0016] generating a plurality of topic element groups based on the first topic element and the second topic element, wherein the topic element groups are used to describe the existence status of the topic;

[0017] The existence frequencies corresponding to all the topic element groups are obtained, and the topic transfer description group is constructed according to the existence frequencies corresponding to all the topic element groups and the topic sequence.

[0018] In a preferred example, the present application may be further configured such that, based on all the second topics, all the first historical browsing texts, and the text generation model, the text segment contents corresponding to all the second topics are obtained, including:

[0019] For each of the second topics, obtaining the first historical copy topics corresponding to all the first historical browsing documents and the corresponding occurrence frequencies of the first historical copy topics;

[0020] Determining a plurality of reference topics corresponding to the second topic based on the occurrence frequencies corresponding to all the first historical copy topics and a preset frequency threshold;

[0021] Obtaining a number of keywords and keyword quantities corresponding to all the reference topics;

[0022] Determining a keyword emission probability threshold corresponding to each of the second topics based on all of the keywords and keyword quantities corresponding to each of the reference topics, wherein the keyword emission probability represents the probability of occurrence of a keyword in a topic, and the keyword emission probability threshold is the minimum probability of the keyword occurring;

[0023] All the text segment contents are obtained based on the number of keywords corresponding to all the second topics, the keyword emission probability thresholds corresponding to all the second topics, and the text generation model.

[0024] In a preferred example, the present application may be further configured as follows: obtaining all the text segment contents based on the number of keywords corresponding to all the second topics, the corresponding keyword emission probability thresholds, and the text generation model, including:

[0025] For each second topic, determining the richness of the second topic according to the number of keywords, and determining whether the richness is greater than a preset richness threshold;

[0026] If not, inputting the keyword emission probability threshold and the second topic into the text generation model to obtain all the paragraph contents;

[0027] If it is greater than, then adding a fusion processing layer to the text generation model to obtain a reconstructed text generation model;

[0028] The keyword emission probability threshold and the second topic are input into the reconstructed text generation model to obtain the text content corresponding to all the second topics.

[0029] In a preferred example, the present application can be further configured as follows: the reconstructed text generation model includes: an input layer, a processing layer, a fusion layer, and an output layer; the keyword emission probability threshold and the second topic are input into the reconstructed text generation model to obtain the text content corresponding to all the second topics, including:

[0030] Inputting the second topic into the input layer to obtain a topic vector corresponding to the second topic, and inputting the topic vector into the processing layer;

[0031] performing a linear transformation on the topic vector within the processing layer to obtain keyword information corresponding to the second topic, the keyword information including an initial keyword and a corresponding initial keyword emission probability;

[0032] In the fusion layer, the initial keyword emission probabilities corresponding to all the initial keywords and the keyword emission probability thresholds are fused to obtain a target keyword, and the target keyword is input into the output layer;

[0033] In the output layer, a text segment corresponding to the second topic is generated based on all the target keywords.

[0034] In a preferred example, the present application may be further configured such that, after generating the to-be-generated copy content based on all the second topics and the corresponding text content, the following is further included:

[0035] Identify all the text paragraphs corresponding to all the second topics, and obtain the corresponding main sentences;

[0036] Determining keywords for the subject sentences according to the subject sentences corresponding to the contents of all the paragraphs;

[0037] Obtaining a plurality of product images corresponding to the advertised product, and determining image tags corresponding to each of the product images;

[0038] Matching the subject sentence keywords corresponding to all the text segments with the image tags corresponding to all the product images to obtain the target product images corresponding to all the text segments;

[0039] Insert the target product image into the content of the copy to be generated.

[0040] In a second aspect, the present application provides a copywriting generation device based on artificial intelligence, which adopts the following technical solutions:

[0041] An artificial intelligence device based on copywriting generation, comprising:

[0042] an acquisition module, configured to acquire a plurality of traffic copywritings, their corresponding traffic values, a plurality of product topics of the advertised products, and user browsing information, wherein the user browsing information includes: a plurality of first historical browsing copywritings;

[0043] A main copy topic determination module, configured to determine a main copy topic based on all the traffic copies and their corresponding traffic values;

[0044] a second topic determination module, configured to determine a plurality of second topics for copy content to be generated based on a pre-constructed topic transition description group, the main copy topic, and all the product topics, wherein the pre-constructed topic transition description group is used to describe the transition probability between topics;

[0045] a paragraph content generation module, configured to obtain paragraph content corresponding to all the second topics according to all the second topics, all the first historical browsing texts, and a text generation model;

[0046] A copywriting generation module, configured to generate the to-be-generated copywriting content based on the main copywriting topic, all the second topics, and the corresponding paragraph contents;

[0047] The second topic determination module constructs the topic transfer description group, including:

[0048] Obtaining the corresponding occurrence frequencies of a plurality of second historical browsing texts and all the second historical text topics within a preset time period;

[0049] Determining a target second historical copy topic and a corresponding same-frequency topic from all the second historical copy topics, and mapping the target second historical copy topic to a first topic element, and mapping the same-frequency topic to a second topic element;

[0050] Obtaining the browsing time corresponding to each of the second historical documents, and generating a topic sequence according to all the browsing times, the first topic element, and the second topic element;

[0051] generating a plurality of topic element groups based on the first topic element and the second topic element, wherein the topic element groups are used to describe the existence status of the topic;

[0052] The existence frequencies corresponding to all the topic element groups are obtained, and the topic transfer description group is constructed according to the existence frequencies corresponding to all the topic element groups and the topic sequence.

[0053] In a third aspect, the present application provides an electronic device, which adopts the following technical solution:

[0054] at least one processor;

[0055] Memory;

[0056] At least one application is stored in the memory, and when the at least one application is executed by the at least one processor, the at least one processor is caused to execute the artificial intelligence method for copywriting generation as described in any one of the first aspects.

[0057] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution:

[0058] A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute the artificial intelligence method for copywriting generation as described in any one of the first aspects.

[0059] In summary, this application has the following beneficial technical effects:

[0060] Obtain multiple traffic copy, traffic value, product topic and the first historical browsing copy of the user; when the main content of the copy is closer to the current political hot spots, it is more attractive to users and can arouse users' reading interest. Therefore, it is necessary to determine the main copy topic based on the traffic copy and traffic value in order to attract readers to read based on the hot copy; when there is a high degree of correlation between topics, the corresponding copy has strong coherence and consistency, so it is necessary to determine the second topic of the copy content to be generated based on the topic transfer description group, the main copy topic and the product topic, so that there is a correlation between the second topic of the copy content to be generated to avoid topic jumping, and to improve the quality of the copy content to be generated from the topic correlation dimension; determine the user with reference to the historical browsing copy The preference for copywriting, and generating corresponding paragraph content based on the user's preferences on the basis of the high-traffic main copywriting topic can further enhance the user's reading interest. Therefore, it is necessary to obtain the paragraph content corresponding to the second topic according to the second topic, the first historical browsing copywriting and the text generation model, and then generate the copywriting content to be generated according to the main copywriting topic, the second topic and the paragraph content; compared with the related technology, this application uses the copywriting of high-traffic current political hot spots as the main copywriting topic, and extends the remaining topics based on this, ensuring the coherence and consistency between topics, and on this basis, generates more detailed paragraph content based on user preferences, so as to achieve the purpose of improving the quality of copywriting content, realize the technical effect of improving content quality, and solve the technical problem of poor copywriting content quality in the related technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 A schematic diagram of a scenario of an artificial intelligence method for copywriting generation provided in an embodiment of the present application;

[0062] Figure 2 A flowchart of an artificial intelligence method for copywriting generation provided in an embodiment of the present application;

[0063] Figure 3 A schematic diagram of the structure of an artificial intelligence device based on copywriting generation provided in an embodiment of the present application;

[0064] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0065] The following is combined with Figure 1 To the attached Figure 4 This application is described in further detail.

[0066] This specific embodiment is merely an explanation of the present application and is not a limitation of the present application. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed, but as long as they are within the scope of the present application, they are protected by patent law.

[0067] To make the objectives, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application are clearly and completely described. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0068] In this document, the term "and / or" simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document, unless otherwise specified, generally indicates an "or" relationship between the related objects.

[0069] like Figure 1 As shown, a scenario interaction diagram provided by an embodiment of the present application is provided. The user sends a copy generation request and an advertising product to the electronic device. After receiving the copy generation request, the electronic device integrates the user's browsing preferences, traffic copy and advertising products to generate copy content that is in line with current political hotspots and user preferences, effectively improving the quality of copy generation.

[0070] The embodiments of the present application are described in further detail below with reference to the accompanying drawings.

[0071] The embodiment of the present application provides an artificial intelligence method for copywriting generation, which is executed by an electronic device, which can be a server or a terminal device, wherein the server can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services. The terminal device can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc., but is not limited to this. The terminal device and the server can be directly or indirectly connected through wired or wireless communication. The embodiment of the present application does not limit this. Figure 2 As shown, the method includes step S101, step S102, step S103, step S104 and step S105, wherein:

[0072] Step S101: Acquire multiple traffic copywriting, their corresponding traffic values, several product topics of the advertised products, and user browsing information, where the user browsing information includes: several first historical browsing copywritings.

[0073] Specifically, a monitoring program is pre-integrated in the electronic device, and the monitoring program is used to monitor the triggering behavior of the acquisition request, and once it is monitored that the acquisition request is triggered, the acquisition operation is executed. Specifically, when the user determines to generate the copy content, an acquisition instruction will be automatically generated, wherein the method of confirming the acquisition instruction may include: the user confirms the acquisition by clicking the acquisition button on the electronic device or by voice, and when the electronic device detects that the user has triggered the acquisition request, the electronic device executes the acquisition command. Traffic copy representations are copy with a high number of clicks, forwarding volumes or related to current political hotspots. Traffic copy can be crawled from the hot topic lists corresponding to multiple social media platforms, and the copy with a preset sequence value (the embodiment of the present application does not limit the preset sequence value) can be selected as the traffic copy. For example, the copy with the top ten order values ​​can be selected as the traffic copy. A product topic represents the product information corresponding to the advertised product. The product information includes multiple product descriptions, each representing the product's descriptive content. Different product descriptions correspond to different descriptive dimensions. For example, if the advertised product is an outdoor backpack, the first product description corresponds to the second product description. The first product description describes the backpack from the perspective of its material, while the second product description describes the backpack from the perspective of its structure. User browsing information can be obtained using a web crawler. For each traffic copy, a traffic value can be determined based on page views, likes, and reposts. The process of determining the traffic value includes: obtaining page views, likes, and reposts for each traffic copy, and obtaining a first weight corresponding to the page views, a second weight corresponding to the likes, and a third weight corresponding to the reposts. The traffic value for each traffic topic is then calculated using the following formula: Traffic value = page views * first weight + likes * second weight + reposts * third weight. Preferably, the first weight > the second weight > the third weight.

[0074] Step S102: Determine the main copy topic based on all traffic copies and their corresponding traffic values.

[0075] Specifically, the main copy topic represents the core content discussed in the copy. It is understandable that the main copy topic can clarify the intention of the copy content, and extending it to other copy contents based on the intention of the copy content can effectively ensure the consistency and coherence of the entire content and improve the quality of the copy content. The specific process of determining the main copy topic based on all traffic copy and their corresponding traffic values ​​includes: comparing the traffic values ​​corresponding to all traffic copy, and taking the traffic copy corresponding to the highest traffic value as the main copy; identifying the semantics of the main copy, and determining the copy topic of the main copy based on the semantics of the main copy, and determining the above copy topic as the main copy topic.

[0076] Step S103: Determine the second topic for the copy content to be generated based on the pre-built topic transition description group, the main copy topic, and all product topics. The pre-built topic transition description group is used to describe the transition probability between topics.

[0077] Specifically, the copy content to be generated includes a main copy topic, a new copy topic, and a product topic. The second topic can be either a new copy topic or a product topic. A topic transition description group is used to describe the probability of transitioning from one topic to the next. In this embodiment of the present application, the topic transition description group is a topic transition probability matrix, which includes multiple topic transition probabilities. A higher topic transition probability between two topics indicates a higher probability of transitioning from one topic to the next. For example, for topic A, topic B, and topic C, if the topic transition probability between topic A and topic B is 90%, and the topic transition probability between topic A and topic C is 80%, then the content of topic A in the copy will be immediately followed by the content of topic B. When a user browses multiple copies of any category (e.g., military or economic) or any product, it indicates that the user is more interested in news or products in that category. Incorporating user-preferred copy topics or product topics into the generated copy can significantly increase the user's reading interest, thereby enhancing product promotion efforts. The specific construction process of the topic transition description group can be referenced in the following embodiments. It will be appreciated that determining the topic order value using the topic transition description group ensures consistency and rationality between topics, thereby generating higher-quality content. The specific process for determining the second topic includes: obtaining the current date and the construction date of the topic transition description group, determining a date difference based on the current date and the construction date, convolving the topic transition description group of the main copy topic based on the date difference, the number of convolutions corresponding to the date difference, i.e., convolving the topic transition description group of the main copy topic by the number of date differences to obtain a target topic transition description group with the main copy topic. Then, selecting the copy topic corresponding to the highest transition probability from the target topic transition description group of the main copy topic, and determining the copy topic with the highest transition probability as the next topic after the main copy topic (i.e., the second topic). It will be appreciated that as the date changes, the user's preference for topics also changes. To more clearly highlight the user's preference for topics, the topic transition description group needs to be convolved. Specifically, as the topic transition description group is convolved, the transition probability between topics changes accordingly. The next topic after the next topic of the main copy topic, i.e., the next topic of the second episode, is then determined. The specific determination process is the same as the above process and will not be further described in detail in this embodiment of the present application. A preset number of second topics (this embodiment of the present application does not limit the preset number) is then generated. A determination is then made as to whether any product topics exist among all second topics. If no product topics exist, a target product topic is selected, and the topic transition probabilities corresponding to the target product topic and all second topics are determined based on the topic transition description group of the target product topic and all second topics. The second topic corresponding to the highest topic transition probability is selected, and the target product topic is used as the next topic after the second topic with the highest topic transition probability.In the embodiment of the present application, the preset target product topic is the main product topic of the advertised product, that is, the target product topic is used to describe the main features of the advertised product.

[0078] Step S104: Obtain the text content corresponding to all second topics according to all second topics, all first historical browsing texts and the text generation model.

[0079] Specifically, the text generation model is pre-entered into the electronic device by a technician, and the text segment content represents the content of each second topic. The specific process of obtaining the text segment content corresponding to each second topic based on the second topic, all first historical browsing copy, and the text generation model can be referred to in the following embodiment. It can be understood that the text generation model can quickly and accurately generate text segment content, so that the content in each second topic is coherent and consistent, and the quality of the content of the to-be-generated copy is improved by improving the quality of the text segment content of each second topic.

[0080] Step S105: Generate the copy content to be generated based on the main copy topic, all the second topics and the corresponding text content.

[0081] Specifically, all second topics and their corresponding paragraph contents are matched one by one to obtain the complete paragraph content of each second topic; then the main copy topic, all second topics and their corresponding paragraph contents are arranged in sequence to generate the copy content to be generated, so as to generate copy that meets user preferences and is of higher quality.

[0082] Based on the above embodiment, multiple traffic copy, traffic value, commodity topic and the first historical browsing copy browsed by the user are obtained; when the main content of the copy is closer to the current political hot spots, it is more attractive to users and can arouse the user's reading interest. Therefore, it is necessary to determine the main copy topic according to the traffic copy and traffic value in order to attract readers to read based on the hot copy; when there is a high correlation between topics, the corresponding copy has strong coherence and consistency. Therefore, it is necessary to determine the second topic of the copy content to be generated according to the topic transfer description group, the main copy topic and the commodity topic, so that there is a correlation between the second topic of the copy content to be generated to avoid topic jumping, thereby improving the quality of the copy content to be generated from the topic correlation dimension; using the historical browsing copy as a reference to determine Determine the user's preference for copywriting, and generate corresponding paragraph content based on the user's preferences on the basis of the high-traffic main copywriting topic, which can further improve the user's reading interest. Therefore, it is necessary to obtain the paragraph content corresponding to the second topic according to the second topic, the first historical browsing copywriting and the text generation model, and then generate the copywriting content to be generated according to the main copywriting topic, the second topic and the paragraph content; compared with the relevant technology, this application uses the copywriting of high-traffic current political hot spots as the main copywriting topic, and extends the remaining topics based on this, ensuring the coherence and consistency between topics, and on this basis, generates more detailed paragraph content based on user preferences, so as to achieve the purpose of improving the quality of copywriting content, realize the technical effect of improving content quality, and solve the technical problem of poor copywriting content quality in the relevant technology.

[0083] A possible implementation of the embodiment of the present application is to construct a topic transfer description group through the following steps, including:

[0084] Obtaining the corresponding occurrence frequencies of several second historical browsing texts and all second historical text topics within a preset time period;

[0085] Determine a target second historical copy topic and corresponding co-frequency topics from all second historical copy topics, map the target second historical copy topic to a first topic element, and map the co-frequency topics to second topic elements respectively;

[0086] Obtaining the browsing time corresponding to each of the second historical texts, and generating a topic sequence based on all browsing times, the first topic element, and the second topic element;

[0087] generating a plurality of topic element groups based on the first topic element and the second topic element, wherein the topic element groups are used to describe the existence status of the topic;

[0088] The existence frequencies corresponding to all topic element groups are obtained, and a topic transfer description group is constructed according to the existence frequencies corresponding to all topic element groups and the topic sequence.

[0089] Specifically, the second historical browsing copy within a preset time period can be obtained. In the embodiment of the present application, the preset time period is pre-set by the technical personnel, such as the browsing copy within December 1st to December 5th. The embodiment of the present application does not limit the preset time period. Any two second historical copy topics can be selected, and any second historical copy topic can be selected from the two second historical copy topics as the target second historical copy topic; the same-frequency topic is the next copy topic adjacent to the target second historical copy topic. In the embodiment of the present application, the number of same-frequency topics corresponding to each target second historical copy topic is 1. For example, if a second historical copy includes in sequence: second historical copy topic 1, second historical copy topic 2, second historical copy topic 3 and second historical copy topic 6, then the second historical copy topic 1 can be selected as the target second historical copy topic, and the second historical copy topic 2 can be selected as the same-frequency topic of the second historical copy topic 1. The browsing time can be obtained from the browsing information library. The existence status of the target second historical copy topic and the existence status of the same-frequency topic can be mapped to a topic element. When the existence status is that the topic exists, the corresponding topic element is 1, and when the existence status is that the topic does not exist, the corresponding topic element is 0. For example, based on the browsing time, it can be determined that the user browsed the second historical browsing copy 1, the second historical browsing copy 2, the second historical browsing copy 3, the second historical browsing copy 4, the second historical browsing copy 5, and the second historical browsing copy 6 in sequence from December 1st to December 5th, wherein the content of copy 1 includes topic 1 and topic 2 in sequence, the content of the second historical browsing copy 2 includes topic 3 and topic 2 in sequence, the content of the second historical browsing copy 3 includes topic 1 and topic 6 in sequence, the content of the second historical browsing copy 4 includes topic 4 and topic 5 in sequence, the content of the second historical browsing copy 5 includes topic 3 and topic 4 in sequence, and the content of the second historical browsing copy 6 includes topic 6 and topic 2 in sequence, and thus the December 1st- The topic sequence within December 5th is 110110000001; taking topic 1 and topic 2 as examples (where topic 1 is the target second historical copy topic and topic 2 is the same-frequency topic of topic 1), the specific process of generating the topic element group includes: when the next topic of topic 1 is topic 2, the mapped topic element group is "11", when the next topic of topic 1 is not topic 2, the mapped topic element group is "10", when the previous topic of topic 2 is not topic 1, the mapped topic element group is "01", and when the topic in the copy is neither topic 1 nor topic 2, the mapped topic element group is "00". The topic element group is matched with the topic sequence to obtain the corresponding frequency of the topic element group, where C 00 =5, C represents the frequency of “00” as 5 (i.e., the frequency of topic 1 and topic 2 not appearing at the same time in the six texts browsed by the user is 5); C 01 =2, indicating that the frequency of occurrence of “01” is 2; C 10=2, indicating that the frequency of occurrence of "10" is 2; C 11 =2, indicating that the frequency of occurrence of "11" is 2. The specific process of constructing a topic transfer description group based on the corresponding existence frequencies and topic sequences of all topic element groups includes: determining a set of associated topic elements, each of which includes two topic element groups, such as the topic element group "00" and the topic element group "01" can constitute an associated topic element set; and then obtaining the matrix element corresponding to the above associated topic element set, that is, the matrix element P corresponding to the topic 1. 00 , P 01 , further, if it is topic 1, the corresponding matrix element is P 10 , P 11 , that is, the topic transfer description group corresponding to topic 1 and topic 2 is P0 Among them, the order of P0 is 1. When the order of P0 is 5, it means that the date difference between the current date and the construction date of the topic transfer description group is 5. It can be understood that when the frequency of simultaneous occurrence of topic 1 and topic 2 is higher, the correlation between topic 1 and topic 2 is higher, and the copy content generated based on this will also have a higher correlation and consistency.

[0090] Based on the above embodiment, the user's preferences may change over time, so it is necessary to obtain the second historical browsing copy within a preset time period to achieve real-time updates of the user's preferences; then determine the target second historical copy topic and the same-frequency topic, and map them to the first topic element and the second topic element respectively to simplify the workload; then obtain the browsing time of all second historical copies, and generate a topic sequence based on the browsing time, the first topic element and the second topic element to further simplify the topic; then generate a topic element group based on the first topic element and the second topic element to simplify the existence status of the topic; then obtain the existence frequency between the topic element groups, and construct a topic transfer description group based on the existence frequency and the topic sequence.

[0091] A possible implementation of the embodiment of the present application is to obtain the text content corresponding to all second topics based on all second topics, all first historical browsing texts, and a text generation model, including:

[0092] For each second topic, obtain the first historical copy topics corresponding to all first historical browsing documents and the corresponding occurrence frequencies of the first historical copy topics;

[0093] Determining a plurality of reference topics corresponding to the second topic based on the occurrence frequencies corresponding to all first historical copy topics and a preset frequency threshold;

[0094] Obtain several keywords and keyword quantities corresponding to all reference topics;

[0095] Determine a keyword emission probability threshold corresponding to each of the second topics based on all keywords and keyword quantities corresponding to each of the reference topics;

[0096] According to the number of keywords corresponding to all second topics, the corresponding keyword emission probability threshold and the text generation model, all paragraph contents are obtained.

[0097] Specifically, in the embodiment of the present application, the reference topic can be a topic that supports or supplements the second topic. The first historical copy topic and the corresponding frequency of occurrence can be obtained through a preset semantic recognition algorithm. The preset frequency threshold is pre-set by a technician. The specific process of determining the reference topic includes: obtaining multiple reference copy with similar content to the main traffic copy (copy corresponding to the main copy topic) from the browsing information library, matching each second topic with the third topic in all the first historical browsing copy through a preset semantic matching algorithm to obtain the topic similarity of the second topic and the third topic, and determining the third topic with a topic similarity higher than the topic similarity threshold (pre-set by a technician) as the reference topic; obtaining the text content of all reference topics, and determining the keyword process includes: in one feasible way, the keywords can be obtained by removing the preset prepositions, modal particles and interjections in the text content of the reference topic; in another feasible way: the keywords corresponding to the reference topic can be determined based on the correspondence between the preset copy topic and the keyword and the reference topic. In the embodiment of the present application, the keywords can be nouns or verbs, etc. The keyword emission probability characterizes the probability of occurrence of keywords in a certain topic. In the embodiment of the present application, the keyword emission probability threshold is the minimum probability of keyword occurrence. For example, the second topic is "tourist city", and the first keyword is: special snacks and special blocks. Among them, if the emission probability corresponding to special snacks is 80%, and the emission probability corresponding to special blocks is 85%, it means that when the text topic "tourist city" appears, "special snacks" has an 80% probability of occurrence and "special blocks" has an 85% probability of occurrence. For each second topic, the keyword information corresponding to all reference topics is obtained. The keyword information includes the existence frequency of each keyword and the sum of the existence frequencies corresponding to all keywords; the average existence frequency is determined based on the existence frequency of each keyword and the sum of the existence frequencies, all average existence frequencies are compared, and the minimum average existence frequency is determined as the keyword emission probability threshold. It can be understood that determining the minimum average existence frequency as the keyword emission probability threshold can increase the number of keywords in the second topic and avoid the poor quality of the text content of the second topic due to the small number of keywords. For example, the second topic is tourist city A. Users read reference topic 1 of tourist city A on December 1, reference topic 2 of tourist city A on December 2, and reference topic 3 of tourist city A on December 5. The keywords corresponding to reference topic 1 read on December 1 are special food, special streets, and seaside scenery. The keywords of reference topic 2 read by users on December 2 are special streets, internet celebrity check-in spots, and seaside scenery. The keywords of reference topic 3 read by users on December 5 are seaside scenery and mountain scenery. Among them, the average frequency of special food is , the average frequency of characteristic blocks is , the average frequency of seaside scenery is , the average frequency of Internet celebrity check-in spots is , the average frequency of mountain and river scenery is , and then we can The keyword emission probability threshold for tourist city A is used. The specific process of deriving all text content based on the number of keywords corresponding to each second topic, the corresponding keyword emission probability threshold, and the text generation model can be found in the following embodiments. Preferably, the text generation model is a Transformer model. It is understood that the Transformer model has high context modeling capabilities and can better adjust text content based on contextual information to ensure coherence and rationality, thereby improving the quality of the text content.

[0098] Based on the above embodiment, the first historical copy topics and corresponding occurrence frequencies corresponding to each of the first historical browsing copies are obtained so as to use the first historical browsing copies as a reference; the reference topic is determined based on the occurrence frequency of the first historical copy topic and the preset frequency threshold to avoid interference from accidental topics; the keywords and the number of keywords of the reference topic are then obtained, and the keyword emission probability threshold is determined based on the number of keywords; the paragraph content is then obtained based on the number of keywords corresponding to all the second topics, the keyword emission probability threshold and the text generation model, and the paragraph content is generated based on keywords with a high degree of correlation as a reference to effectively improve the generation quality of the paragraph content.

[0099] A possible implementation of the embodiment of the present application is to obtain all text segment contents based on the number of keywords corresponding to all second topics, the corresponding keyword emission probability thresholds, and the text generation model, including:

[0100] For each second topic, determining the richness of the second topic based on the number of keywords, and determining whether the richness is greater than a preset richness threshold;

[0101] If it is not greater than, the keyword emission probability threshold and the second topic are input into the text generation model to obtain the content of all paragraphs;

[0102] If it is greater than, then a fusion processing layer is added to the text generation model to obtain a reconstructed text generation model;

[0103] The keyword emission probability threshold and the second topic are input into the reconstructed text generation model to obtain the text content corresponding to all the second topics.

[0104] Specifically, in the embodiment of the present application, the richness of the second topic represents the amount of information contained in the second topic. As the richness increases, the greater the amount of information in the second topic, the greater the number of keywords corresponding to the keywords. The richness of the second topic can be determined based on the corresponding relationship between the preset number of keywords and the richness; the preset richness threshold can be pre-set by a technician. If it is not greater than the preset richness threshold, it indicates that the text content of the second topic has fewer keywords, so the text generation model can be used directly to generate text content with higher coherence and higher quality. If it is greater than the preset richness threshold, it indicates that the text content of the second topic has a large number of keywords. In order to ensure that the text content has a higher quality and avoid generating irrelevant keywords, it is necessary to add a fusion processing layer to screen the keywords. The specific process of generating text content by the reconstructed text generation model can refer to the following embodiment.

[0105] Based on the above embodiment, the richness of the second topic is determined according to the number of keywords, and it is judged whether the richness is greater than the preset richness threshold. If it is not greater, it indicates that the richness of the second topic is low. Otherwise, it indicates that the second topic has a high richness, and therefore it is necessary to add a fusion processing layer to output higher quality text content through keyword screening.

[0106] In one possible implementation of the embodiment of the present application, the reconstructed text generation model includes: an input layer, a processing layer, a fusion layer, and an output layer. The keyword emission probability threshold and the second topic are input into the reconstructed text generation model to obtain the text content corresponding to all the second topics, including:

[0107] Inputting the second topic into the input layer to obtain a topic vector corresponding to the second topic, and inputting the topic vector into the processing layer;

[0108] Performing a linear transformation on the topic vector in the processing layer to obtain keyword information corresponding to the second topic, the keyword information including the initial keyword and the corresponding initial keyword emission probability;

[0109] In the fusion layer, the initial keyword emission probabilities and keyword emission probability thresholds corresponding to all initial keywords are fused to obtain the target keyword, and the target keyword is input to the output layer;

[0110] In the output layer, the text content corresponding to the second topic is generated based on all target keywords.

[0111] Specifically, in an embodiment of the present application, the processing layer includes an encoder and a decoder. The second topic is input into the text generation model. The input layer in the text generation model first maps the second topic to an initial topic vector. The present application does not limit the specific process of the mapping. The encoder in the processing layer encodes the initial topic vector to obtain a higher-level feature element, which is then input into the decoder. The decoder uses a masked multi-head self-attention mechanism, an encoder-decoder attention mechanism, and feature elements to determine the association between the second topic and all initial keywords. The feature elements are then linearly transformed through an internal feedforward neural network to obtain initial keywords associated with the second topic and the corresponding initial keyword emission probability. Processing is performed in the fusion layer to obtain a target keyword. The process of determining the target keyword includes: obtaining a first weight value corresponding to a keyword emission probability threshold, and determining a second weight value of the initial keyword based on the first weight value, i.e., the second weight value = 1-first weight value, thereby obtaining a target keyword emission probability corresponding to each initial keyword, target keyword emission probability = keyword emission probability threshold * preset first weight value + second weight value * initial keyword emission probability. It is understandable that by introducing the weight value, the main role of the keyword emission probability threshold can be maximized, that is, the emission probability of the initial keyword is ensured to be at a high level, thereby obtaining a target keyword with a higher correlation with the second topic.

[0112] Based on the above embodiment, the second topic is input into the input layer to obtain the corresponding topic vector, so that the main idea of ​​the second topic can be captured more accurately through the topic vector; the topic vector is then input into the processor layer to obtain keyword information corresponding to the emission probability to determine the keyword information of the second topic; the initial keyword emission probability and the keyword emission probability threshold are then fused to obtain the final target keyword, and all keywords are then output to obtain text content that has a high degree of relevance and coherence with the second topic.

[0113] A possible implementation of the embodiment of the present application further includes, after generating the content to be generated based on all second topics and their corresponding text segments, the following steps are performed:

[0114] Identify all the text paragraphs corresponding to all the second topics and obtain the corresponding main sentences;

[0115] Determine the keywords of the main sentences according to the corresponding main sentences of all the paragraphs;

[0116] Obtain several product images corresponding to the advertised product and determine the image tags corresponding to all product images;

[0117] Match the subject sentence keywords corresponding to all the text segments with the image tags corresponding to all the product images to obtain the target product images corresponding to all the text segments;

[0118] Insert the target product image into the copy content to be generated.

[0119] Specifically, identifying the causal association words in the text content of each second topic and determining the main sentence based on the causal association words specifically includes: matching the text content with multiple preset causal association words through a preset text matching algorithm to obtain the target causal association words in the text content; then determining the sentence after the target word in the target causal association word as the main sentence, for example, if the target causal association word is "because-so", then the target word is "so", and the sentence after "so" is the main sentence. When there are no causal association words in the text content of the second topic, the first sentence or the last sentence of the paragraph can be selected as the main sentence. It can be understood that compared with directly analyzing the content of each paragraph, directly analyzing the main sentence can effectively reduce the workload and complexity of the work and improve the analysis efficiency. Among them, the target word can be a word that indicates a transition or a progressive relationship in the causal association words. The specific process of determining the main sentence keywords is the same as the process of determining the keywords of the reference topic mentioned above, and will not be repeated in the embodiments of this application. The main sentence keywords in each main sentence can be one or more. Multiple product images can be obtained from the advertising product information library; the product images can be input into a preset image annotation model, and the image annotation model outputs the product images and multiple image tags corresponding to the product images. The image annotation model is obtained by technicians training a neural network model based on multiple sample data. This application does not limit the specific training process. For each paragraph content, all keywords of each main sentence are semantically matched with each image tag of the product image to obtain the matching degree of each image tag with all keywords; the sum of the matching degrees corresponding to each image tag is calculated, and the first average matching degree is determined based on the number of keywords and the sum of the matching degrees; the second average matching degree corresponding to the product image and the paragraph is determined based on all image tags and their corresponding average matching degrees; all second average matching degrees are compared, and the product image corresponding to the maximum second average matching degree is determined as the target product image corresponding to the paragraph content, and then the target product image is inserted into any position in the corresponding paragraph content. It can be understood that by inserting pictures, the paragraph content can be made more vivid and vivid, so as to further enhance the reader's reading interest. The preset matching degree threshold is set by the technician based on work experience.

[0120] Based on the above embodiment, the main sentence can accurately and concisely summarize the content of the paragraph, so it is necessary to obtain the main sentence so as to directly analyze and process the main sentence; then determine the main sentence keywords, and the workload can be further reduced by analyzing the keywords in the main sentence; then obtain the product image corresponding to the advertised product and the corresponding image label, match the image label with the main sentence keyword to obtain the target product image, so as to ensure that the image and text have a high degree of adaptability, and insert the target product image into the corresponding paragraph; by adding pictures, the richness and quality of the content of the copy to be generated can be further improved.

[0121] The above embodiment introduces an artificial intelligence method for copywriting generation from the perspective of method flow. The following embodiment introduces an artificial intelligence device based on copywriting generation from the perspective of a virtual module or virtual unit. For details, please refer to the following embodiment.

[0122] The embodiment of the present application provides an artificial intelligence device based on copywriting generation, such as Figure 3 As shown, the artificial intelligence device based on copywriting generation may specifically include:

[0123] The acquisition module 201 is used to acquire a plurality of traffic copywriting, the traffic values ​​corresponding to each copywriting, a plurality of product topics of the advertised product, and user browsing information, wherein the user browsing information includes: a plurality of first historical browsing copywriting;

[0124] A main copy topic determination module 202 is configured to determine a main copy topic based on all traffic copies and their corresponding traffic values;

[0125] A second topic determination module 203 is configured to determine a number of second topics for copywriting content to be generated based on a pre-built topic transition description group, the main copywriting topic, and all product topics. The pre-built topic transition description group is used to describe the transition probability between topics.

[0126] The text segment content generation module 204 is configured to obtain text segment contents corresponding to all second topics based on all second topics, all first historical browsing texts, and the text generation model;

[0127] The copywriting generation module 205 is used to generate the copywriting content to be generated based on the main copywriting topic, all the secondary topics and the corresponding paragraph contents;

[0128] Among them, the topic transfer description group is constructed, specifically for:

[0129] Obtaining the corresponding occurrence frequencies of several second historical browsing texts and all second historical text topics within a preset time period;

[0130] Determine a target second historical copy topic and a corresponding co-frequency topic from all second historical copy topics, and map the target second historical copy topic to a first topic element, and map the co-frequency topic to a second topic element;

[0131] Obtaining the browsing time corresponding to each of the second historical texts, and generating a topic sequence based on all browsing times, the first topic element, and the second topic element;

[0132] generating a plurality of topic element groups based on the first topic element and the second topic element, wherein the topic element groups are used to describe the existence status of the topic;

[0133] The existence frequencies corresponding to all topic element groups are obtained, and a topic transfer description group is constructed according to the existence frequencies corresponding to all topic element groups and the topic sequence.

[0134] Based on the above embodiment, multiple traffic copy, traffic value, commodity topic and the first historical browsing copy browsed by the user are obtained; when the main content of the copy is closer to the current political hot spots, it is more attractive to users and can arouse the user's reading interest. Therefore, it is necessary to determine the main copy topic according to the traffic copy and traffic value in order to attract readers to read based on the hot copy; when there is a high correlation between topics, the corresponding copy has strong coherence and consistency. Therefore, it is necessary to determine the second topic of the copy content to be generated according to the topic transfer description group, the main copy topic and the commodity topic, so that there is a correlation between the second topics of the copy content to be generated to avoid topic jumping, thereby improving the quality of the copy content to be generated from the topic correlation dimension; using the historical browsing copy as a reference Determining the user's preference for copywriting, and generating corresponding paragraph content based on the user's preferences on the basis of the high-traffic main copywriting topic can further enhance the user's reading interest. Therefore, it is necessary to obtain the paragraph content corresponding to each of the second topics based on the second topic, the first historical browsing copywriting and the text generation model, and then generate the copywriting content to be generated based on the main copywriting topic, the second topic and the paragraph content; compared with the related technology, this application uses the copywriting of high-traffic current political hot spots as the main copywriting topic, and extends the remaining topics based on this, ensuring the coherence and consistency between topics, and on this basis, generates more detailed paragraph content based on user preferences, so as to achieve the purpose of improving the quality of copywriting content, realize the technical effect of improving content quality, and solve the technical problem of poor copywriting content quality in the related technology.

[0135] In one possible implementation of the embodiment of the present application, when the paragraph content generation module 204 obtains the paragraph content corresponding to all second topics based on all second topics, all first historical browsing documents, and the text generation model, it is configured to:

[0136] For each second topic, obtain the first historical copy topics corresponding to all first historical browsing documents and the corresponding occurrence frequencies of the first historical copy topics;

[0137] Determining a plurality of reference topics corresponding to the second topic based on the occurrence frequencies corresponding to all first historical copy topics and a preset frequency threshold;

[0138] Obtain several keywords and keyword quantities corresponding to all reference topics;

[0139] Determine a keyword emission probability threshold corresponding to each of the second topics based on all keywords and keyword quantities corresponding to each of the reference topics, wherein the keyword emission probability represents the probability of occurrence of a keyword in a topic, and the keyword emission probability threshold is the minimum probability of the keyword occurring;

[0140] According to the number of keywords corresponding to all second topics, the corresponding keyword emission probability threshold and the text generation model, all paragraph contents are obtained.

[0141] In one possible implementation of the embodiment of the present application, when the text segment content generation module 204 obtains all text segment contents based on the number of keywords corresponding to all second topics, the corresponding keyword emission probability thresholds, and the text generation model, it is configured to:

[0142] For each second topic, determining the richness of the second topic based on the number of keywords, and determining whether the richness is greater than a preset richness threshold;

[0143] If it is not greater than, the keyword emission probability threshold and the second topic are input into the text generation model to obtain the content of all paragraphs;

[0144] If it is greater than, then a fusion processing layer is added to the text generation model to obtain a reconstructed text generation model;

[0145] The keyword emission probability threshold and the second topic are input into the reconstructed text generation model to obtain the text content corresponding to all the second topics.

[0146] In one possible implementation of the embodiment of the present application, the reconstructed text generation model includes: an input layer, a processing layer, a fusion layer, and an output layer. When the text segment content generation module 204 inputs the keyword emission probability threshold and the second topic into the reconstructed text generation model to obtain the text segment content corresponding to all the second topics, it is used to:

[0147] Inputting the second topic into the input layer to obtain a topic vector corresponding to the second topic, and inputting the topic vector into the processing layer;

[0148] Performing a linear transformation on the topic vector in the processing layer to obtain keyword information corresponding to the second topic, the keyword information including the initial keyword and the corresponding initial keyword emission probability;

[0149] In the fusion layer, the initial keyword emission probabilities and keyword emission probability thresholds corresponding to all initial keywords are fused to obtain the target keyword, and the target keyword is input to the output layer;

[0150] In the output layer, the text content corresponding to the second topic is generated based on all target keywords.

[0151] A possible implementation of the embodiment of the present application is an artificial intelligence device based on document generation, further comprising:

[0152] The module for generating content of the copy to be generated is used to:

[0153] Identify all the text paragraphs corresponding to all the second topics and obtain the corresponding main sentences;

[0154] Determine the keywords of the main sentences according to the corresponding main sentences of all the paragraphs;

[0155] Obtain several product images corresponding to the advertised product and determine the image tags corresponding to all product images;

[0156] Match the subject sentence keywords corresponding to all the text segments with the image tags corresponding to all the product images to obtain the target product images corresponding to all the text segments;

[0157] Insert the target product image into the copy content to be generated.

[0158] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process of the artificial intelligence device based on text generation described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0159] An electronic device is provided in an embodiment of the present application, such as Figure 4 As shown, Figure 4 The electronic device shown includes a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, via a bus 302. Optionally, the electronic device may further include a transceiver 304. It should be noted that in practice, the number of transceivers 304 is not limited to one, and the structure of the electronic device does not constitute a limitation on the embodiments of the present application.

[0160] Processor 301 can be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 301 can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0161] Bus 302 may include a path for transmitting information between the above components. Bus 302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. Bus 302 may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, Figure 4 Only one thick line is used in the diagram, but it does not mean that there is only one bus or one type of bus.

[0162] The memory 303 may be a ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, a RAM (Random Access Memory) or other type of dynamic storage device that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0163] The memory 303 is used to store application code for executing the solution of the present application, and the execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 303 to implement the content shown in the above method embodiment.

[0164] Electronic devices include, but are not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. They may also include servers, etc. Figure 4 The electronic device shown is merely an example and should not limit the functions and scope of use of the embodiments of the present application.

[0165] An embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer-readable storage medium is run on a computer, the computer can execute the corresponding content in the aforementioned method embodiment, and obtain multiple traffic copywriting, traffic values, product topics and the first historical browsing copywriting browsed by the user compared with the related art; when the main content of the copywriting is closer to the current political hot spots, it is more attractive to the user and can arouse the user's reading interest. Therefore, it is necessary to determine the main copywriting topic based on the traffic copywriting and traffic value in order to attract readers to read based on the hot copywriting; when there is a high degree of correlation between topics, the corresponding copywriting has strong coherence and consistency. Therefore, it is necessary to determine the second topic of the copywriting content to be generated based on the topic transfer description group, the main copywriting topic and the product topic, so that there is correlation between the second topics of the copywriting content to be generated to avoid the occurrence of words Topic jumping is realized, which improves the quality of the copy content to be generated from the topic association dimension; the user's preference for copy is determined by referring to the historical browsing copy, and the corresponding paragraph content is generated based on the user's preference on the basis of the high-traffic main copy topic, which can further improve the user's reading interest. Therefore, it is necessary to obtain the paragraph content corresponding to the second topic according to the second topic, the first historical browsing copy and the text generation model, and then generate the copy content to be generated according to the main copy topic, the second topic and the paragraph content; compared with the relevant technology, this application uses the copy of the high-traffic current political hot spots as the main copy topic, and extends the remaining topics based on this, ensuring the coherence and consistency between the topics, and on this basis, generates more detailed paragraph content based on the user's preference, so as to achieve the purpose of improving the quality of the copy content, realize the technical effect of improving the content quality, and solve the technical problem of poor copy content quality in the relevant technology.

[0166] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown in sequence as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the flowcharts of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0167] The above description is only part of the implementation methods 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. An artificial intelligence method for copywriting generation, characterized in that: The following steps are involved: Acquire multiple traffic copywriting, their corresponding traffic values, several product topics of the advertised products, and user browsing information, wherein the user browsing information includes: several first historical browsing copywriting; Determine a main copy topic based on all the traffic copies and their corresponding traffic values; Determining a plurality of second topics for copy content to be generated based on a pre-constructed topic transition description group, the main copy topic, and all the product topics, wherein the pre-constructed topic transition description group is used to describe the transition probability between topics; Obtaining text content corresponding to all of the second topics according to all of the second topics, all of the first historical browsing texts, and a text generation model; Generate the to-be-generated copy content according to the main copy topic, all the second topics, and the corresponding paragraph contents; The step of constructing the topic transfer description group includes: Obtaining the corresponding occurrence frequencies of a plurality of second historical browsing texts and all the second historical text topics within a preset time period; Determining a target second historical copy topic and a corresponding same-frequency topic from all the second historical copy topics, and mapping the target second historical copy topic to a first topic element, and mapping the same-frequency topic to a second topic element; Obtaining the browsing time corresponding to each of the second historical documents, and generating a topic sequence according to all the browsing times, the first topic element, and the second topic element; generating a plurality of topic element groups based on the first topic element and the second topic element, wherein the topic element groups are used to describe the existence status of the topic; The existence frequencies corresponding to all the topic element groups are obtained, and the topic transfer description group is constructed according to the existence frequencies corresponding to all the topic element groups and the topic sequence.

2. The artificial intelligence method for copywriting generation according to claim 1, characterized in that: The step of obtaining the text content corresponding to each of the second topics based on all the second topics, all the first historical browsing texts, and the text generation model includes the following steps: For each of the second topics, obtaining the first historical copy topics corresponding to all the first historical browsing documents and the corresponding occurrence frequencies of the first historical copy topics; Determining a plurality of reference topics corresponding to the second topic based on the occurrence frequencies corresponding to all the first historical copy topics and a preset frequency threshold; Obtaining a number of keywords and keyword quantities corresponding to all the reference topics; Determining a keyword emission probability threshold corresponding to each of the second topics based on all of the keywords and keyword quantities corresponding to each of the reference topics, wherein the keyword emission probability represents the probability of occurrence of a keyword in a topic, and the keyword emission probability threshold is the minimum probability of the keyword occurring; All the text segment contents are obtained based on the number of keywords corresponding to all the second topics, the keyword emission probability thresholds corresponding to all the second topics, and the text generation model.

3. The artificial intelligence method for copywriting generation according to claim 2, characterized in that: The step of obtaining all the text segment contents according to the number of keywords corresponding to all the second topics, the keyword emission probability thresholds corresponding to all the second topics, and the text generation model comprises the following steps: For each second topic, determining the richness of the second topic according to the number of keywords, and determining whether the richness is greater than a preset richness threshold; If not, inputting the keyword emission probability threshold and the second topic into the text generation model to obtain all the paragraph contents; If it is greater than, then adding a fusion processing layer to the text generation model to obtain a reconstructed text generation model; The keyword emission probability threshold and the second topic are input into the reconstructed text generation model to obtain the text content corresponding to all the second topics.

4. The artificial intelligence method for copywriting generation according to claim 3, characterized in that: The reconstructed text generation model includes: an input layer, a processing layer, a fusion layer and an output layer. The step of inputting the keyword emission probability threshold and the second topic into the reconstructed text generation model to obtain the text content corresponding to all the second topics comprises the following steps: Inputting the second topic into the input layer to obtain a topic vector corresponding to the second topic, and inputting the topic vector into the processing layer; performing a linear transformation on the topic vector within the processing layer to obtain keyword information corresponding to the second topic, the keyword information including an initial keyword and a corresponding initial keyword emission probability; In the fusion layer, the initial keyword emission probabilities corresponding to all the initial keywords and the keyword emission probability thresholds are fused to obtain a target keyword, and the target keyword is input into the output layer; In the output layer, a text segment corresponding to the second topic is generated based on all the target keywords.

5. The artificial intelligence method for copywriting generation according to claim 1, characterized in that: After generating the to-be-generated copy content according to all the second topics and the corresponding paragraph contents, the following steps are also included: Identify all the text paragraphs corresponding to all the second topics, and obtain the corresponding main sentences; Determining keywords for the subject sentences according to the subject sentences corresponding to the contents of all the paragraphs; Obtaining a plurality of product images corresponding to the advertised product, and determining image tags corresponding to each of the product images; Matching the subject sentence keywords corresponding to all the text segments with the image tags corresponding to all the product images to obtain the target product images corresponding to all the text segments; Insert the target product image into the content of the copy to be generated.

6. An artificial intelligence device for copywriting generation, characterized in that: include: an acquisition module, configured to acquire a plurality of traffic copywritings, their corresponding traffic values, a plurality of product topics of the advertised products, and user browsing information, wherein the user browsing information includes: a plurality of first historical browsing copywritings; A main copy topic determination module, configured to determine a main copy topic based on all the traffic copies and their corresponding traffic values; a second topic determination module, configured to determine a plurality of second topics for copy content to be generated based on a pre-constructed topic transition description group, the main copy topic, and all the product topics, wherein the pre-constructed topic transition description group is used to describe the transition probability between topics; a paragraph content generation module, configured to obtain paragraph content corresponding to all the second topics according to all the second topics, all the first historical browsing texts, and a text generation model; A copywriting generation module, configured to generate the to-be-generated copywriting content based on the main copywriting topic, all the second topics, and the corresponding paragraph contents; The second topic determination module constructs the topic transfer description group, including: Obtaining the corresponding occurrence frequencies of a plurality of second historical browsing texts and all the second historical text topics within a preset time period; Determining a target second historical copy topic and a corresponding same-frequency topic from all the second historical copy topics, and mapping the target second historical copy topic to a first topic element, and mapping the same-frequency topic to a second topic element; Obtaining the browsing time corresponding to each of the second historical documents, and generating a topic sequence according to all the browsing times, the first topic element, and the second topic element; generating a plurality of topic element groups based on the first topic element and the second topic element, wherein the topic element groups are used to describe the existence status of the topic; The existence frequencies corresponding to all the topic element groups are obtained, and the topic transfer description group is constructed according to the existence frequencies corresponding to all the topic element groups and the topic sequence.

7. An electronic device, characterized in that: include: at least one processor; Memory; At least one application is stored in the memory, and when the at least one application is executed by the at least one processor, the at least one processor executes the artificial intelligence method for copywriting generation according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed in a computer, the computer is caused to execute the artificial intelligence method for generating text as described in any one of claims 1 to 5.

Images