Advertising text generation methods, devices, equipment, and media

Abstract

This application relates to a method, apparatus, device, and medium for generating advertising text in the field of computer technology. The method includes: obtaining the title of an advertised product and determining multiple selling point keywords matching the product title from a selling point database of the product category to which the advertised product belongs; constructing a first sentence pair by combining the product title with each selling point keyword; inputting each first sentence pair into a preset first-sentence generation model to obtain corresponding first-sentence text, thus constructing a first-sentence set; concatenating the product title and each first-sentence text in the first-sentence set, and then constructing a second sentence pair by combining them with each selling point keyword; inputting each second sentence pair into a preset second-sentence generation model to obtain corresponding second-sentence text, thus constructing a second-sentence set; and concatenating the first-sentence text in the first-sentence set with the second-sentence text in the second-sentence set. This application can generate advertising text that accurately describes the selling points of a product.

Description

Technical Field

[0001] This application relates to the field of e-commerce technology, and in particular to a method for generating advertising text and the corresponding apparatus, computer equipment, and computer-readable storage medium. Background Technology

[0002] In e-commerce, advertising is a crucial way to attract potential customers and drive purchases. One of the key elements presented to users during advertising is the advertising copy. In practice, creating effective advertising copy requires a thorough understanding of the product and target market, as well as familiarity with advertising copywriting guidelines. Based on this understanding, one must extract the product's selling points, employ appropriate expression techniques, and craft copy that resonates with the market and maximizes commercial success. This clearly demonstrates the significant time and effort required for advertising copy creation. Furthermore, writing advertising copy demands a high level of expertise, which merchants often lack, further complicating the process.

[0003] Traditional advertising text generation technology typically inputs information about the product itself, such as the product title and details, and generates corresponding advertising text based on a pre-trained and converged encoder-decoder network. However, because it lacks the inclusion of relevant product selling points as a basis for generation, the generated advertising text does not adequately and accurately describe the corresponding product selling points, resulting in unsatisfactory advertising text quality that fails to attract customers or induce purchase behavior.

[0004] Given the shortcomings of traditional technologies, the applicant has made corresponding explorations. Summary of the Invention

[0005] The primary objective of this application is to solve at least one of the aforementioned problems by providing a method for generating advertising text and corresponding apparatus, computer equipment, and computer-readable storage medium.

[0006] To achieve the various objectives of this application, the following technical solution is adopted:

[0007] An advertising text generation method provided for one of the purposes of this application includes the following steps:

[0008] Obtain the product title of the advertised product, and identify multiple selling point keywords that match the product title from the selling point database of the product category to which the advertised product belongs;

[0009] The product title is combined with each selling point keyword to form a first sentence pair. Each first sentence pair is input into a preset first half sentence generation model to obtain the corresponding first half sentence text, and a first half sentence set is constructed.

[0010] The product title and each first half of the ...

[0011] The first half of the sentence text in the first half of the sentence set is concatenated with the second half of the sentence text in the second half of the sentence set.

[0012] On the other hand, an advertising text generation device provided to meet one of the purposes of this application includes a data acquisition module, a first half sentence generation module, a second half sentence generation module, and an advertising text generation module. The data acquisition module is used to acquire the product title of the advertised product and determine multiple selling point keywords matching the product title from a selling point database of the product category to which the advertised product belongs. The first half sentence generation module is used to construct a first sentence pair by combining the product title with each selling point keyword, and input each first sentence pair into a preset first half sentence generation model to obtain corresponding first half sentence text, thus constructing a first half sentence set. The second half sentence generation module is used to concatenate the product title and each first half sentence text in the first half sentence set, and then construct a second sentence pair by combining them with each selling point keyword, and input each second sentence pair into a preset second half sentence generation model to obtain corresponding second half sentence text, thus constructing a second half sentence set. The advertising text generation module is used to concatenate the first half text in the first half sentence set with the second half text in the second half sentence set.

[0013] In another aspect, a computer device provided for one of the purposes of this application includes a central processing unit and a memory, the central processing unit being used to invoke and run a computer program stored in the memory to perform the steps of the advertising text generation method described in this application.

[0014] In another aspect, a computer-readable storage medium is provided to suit another purpose of this application, which stores, in the form of computer-readable instructions, a computer program implemented according to the described advertising text generation method, which, when invoked by a computer, performs the steps included in the method.

[0015] The technical solution of this application has many advantages, including but not limited to the following aspects:

[0016] This application implements a two-stage generation scheme. First, multiple first-sentence pairs are constructed using the product title and its matching selling point keywords. A first-sentence generation model is used to generate corresponding first-sentence texts. Then, the product title and selling point keywords are combined with each first-sentence text to construct a second-sentence pair. A second-sentence generation model is used to obtain corresponding second-sentence texts. Finally, each first-sentence text is concatenated with its corresponding second-sentence text to obtain complete advertising texts. On the one hand, the selling point features of the advertised product are introduced in both stages of this two-stage generation process, ensuring that both the generated first-sentence and second-sentence texts maintain a strong correlation with the selling point features and accurately describe the selling points of the advertised product. Thus, the description of the complete advertising text can fully focus on the selling points of the advertised product. On the other hand, the two-segment generation scheme achieves intelligent generation. The generation of the second half of the text depends on the first half of the text as input, and the two still maintain semantic coherence, which can ensure that the expression of the resulting advertising text is fluent. Moreover, it can generate diverse first half of the text based on multiple selling points and product titles related to the advertised product, and then generate diverse second half of the text accordingly, so that the final advertising text is rich enough to provide sufficient reference and effectively assist in the creation of advertising text. Attached Figure Description

[0017] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0018] Figure 1 This is a flowchart illustrating a typical embodiment of the advertising text generation method of this application;

[0019] Figure 2 This is a flowchart illustrating the process of constructing a database of selling points corresponding to multiple product categories in an embodiment of this application;

[0020] Figure 3 This is a schematic diagram illustrating the process of preparing training samples corresponding to the first half-sentence generation model and the second half-sentence generation model in an embodiment of this application.

[0021] Figure 4 This is a flowchart illustrating the process of obtaining the product title corresponding to the historically advertised product and its historical advertising text, and determining the first selling point keyword in the historical advertising text, as described in an embodiment of this application.

[0022] Figure 5 This is a schematic diagram of the training process of the first half of the sentence generation model in the embodiments of this application;

[0023] Figure 6 This is a schematic diagram illustrating the training process of the second half sentence generation model in an embodiment of this application;

[0024] Figure 7 This is a schematic block diagram of the advertising text generation device of this application;

[0025] Figure 8 This is a schematic diagram of the structure of a computer device used in this application. Detailed Implementation

[0026] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0027] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated 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 say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0028] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0029] Those skilled in the art will understand that the terms "client," "terminal," and "terminal device" as used herein include both devices that receive wireless signals, devices that only possess wireless signal receiver capabilities without transmission capabilities, and devices with receiving and transmitting hardware, devices that have receiving and transmitting hardware capable of bidirectional communication over a bidirectional communication link. Such devices may include: cellular or other communication devices such as personal computers or tablets, having single-line displays, multi-line displays, or cellular or other communication devices without multi-line displays; PCS (Personal Communications Service) that can combine voice, data processing, fax, and / or data communication capabilities; PDA (Personal Digital Assistant) that may include a radio frequency receiver, pager, internet / intranet access, web browser, notepad, calendar, and / or GPS (Global Positioning System) receiver; and conventional laptops and / or handheld computers or other devices that have and / or include radio frequency receivers. As used herein, "client," "terminal," and "terminal device" can be portable, transportable, installed in a means of transportation (air, sea, and / or land), or suitable and / or configured to operate locally and / or in a distributed manner, operating in any other location on Earth and / or in space. "Client," "terminal," and "terminal device" as used herein can also be a communication terminal, an internet access terminal, or a music / video playback terminal, such as a PDA, a MID (Mobile Internet Device), and / or a mobile phone with music / video playback capabilities, or a smart TV, set-top box, etc.

[0030] The hardware referred to by the names "server," "client," and "service node" in this application is essentially an electronic device with the equivalent capabilities of a personal computer. It is a hardware device with the necessary components revealed by the von Neumann architecture, such as a central processing unit (including an arithmetic logic unit and a control unit), memory, input devices, and output devices. The computer program is stored in its memory, and the central processing unit loads the program stored in the secondary storage into the main memory to run it, execute the instructions in the program, and interact with the input and output devices to complete specific functions.

[0031] It should be noted that the concept of "server" used in this application can also be extended to the case of server clusters. Based on the network deployment principles understood by those skilled in the art, the servers should be logically divided. Physically, these servers can be independent of each other but accessible through interfaces, or they can be integrated into a single physical computer or a computer cluster. Those skilled in the art should understand this flexibility and should not use it to constrain the implementation of the network deployment method in this application.

[0032] One or more of the technical features of this application, unless explicitly specified herein, can be deployed on a server and accessed by a client remotely calling the online service interface provided by the server, or can be directly deployed and run on a client for access.

[0033] Unless otherwise specified, the neural network models referenced or potentially referenced in this application may be deployed on a remote server and invoked remotely on the client, or deployed on a client with the capability to invoke directly. In some embodiments, when running on the client, the corresponding intelligence may be acquired through transfer learning in order to reduce the requirements on the client's hardware resources and avoid excessive consumption of the client's hardware resources.

[0034] Unless otherwise specified, all data involved in this application may be stored remotely on a server or on a local terminal device, as long as it is suitable for use by the technical solution of this application.

[0035] Those skilled in the art will understand that although the various methods in this application are described based on the same concept and thus present commonality among them, they can be performed independently unless otherwise specified. Similarly, the various embodiments disclosed in this application are all based on the same inventive concept; therefore, concepts expressed in the same way, as well as concepts that are appropriately changed for convenience but are expressed differently, should be understood equivalently.

[0036] Unless otherwise expressly stated, the various embodiments disclosed in this application can be combined in a cross-cutting manner to flexibly construct new embodiments, as long as such combination does not depart from the inventive spirit of this application and can meet the needs of the prior art or solve a certain deficiency in the prior art. Those skilled in the art should be aware of such modifications.

[0037] The advertising text generation method of this application can be programmed into a computer program product and deployed on a client or server for execution. For example, in an exemplary application scenario of this application, it can be deployed on the server of an e-commerce platform, thereby allowing human-computer interaction with the process of the computer program product through a graphical user interface by accessing the interface opened after the computer program product is running.

[0038] Please see Figure 1 The advertising text generation method of this application, in its typical embodiment, includes the following steps:

[0039] Step S1100: Obtain the product title of the advertised product, and determine multiple selling point keywords that match the product title from the selling point database of the product category to which the advertised product belongs;

[0040] The e-commerce platform in this application scenario primarily refers to a cross-border e-commerce platform, which allows each online store to operate on a corresponding independent site, providing services to the public. Public users can access the online store's pages through various channels, including but not limited to the online store's built-in pages and social media information display pages. Public users can directly access the corresponding pages of the online store through links on these channels.

[0041] When an online store user needs to generate advertising text for a product to be advertised, in one embodiment, this can be done through an ad generation page where the user can specify the product corresponding to the advertising text. Typically, the advertising product can be a product from the online store's product database, or it can be a customized product provided by the user with specific product information. For ease of understanding, it is assumed that all product-related information is stored in the product database, and the user can specify the advertising product through the ad generation page.

[0042] When an online store user needs to specify an advertised product, they can typically retrieve the product information using the product's unique identifier, i.e., the product ID. This product information includes, but is not limited to, different types of data such as images and text. The text data within the product information includes, but is not limited to, any one or more of the following: product title, product details text, and product attribute data. However, the product title is usually a concise description of the product's selling points, encompassing the selling points needed to generate the advertising text. Therefore, using the product title as input data for generating the corresponding advertising text helps extract deep semantic features related to the product's selling points from the input product title when using a generative approach to generate the advertising text describing the product's selling points. Accordingly, after an online store user specifies an advertised product, they can obtain the product title based on the product's unique identifier.

[0043] Furthermore, in order to more accurately describe the selling points of the product in the subsequent generated advertising text, selling point keywords that match the product title of the advertised product can be identified. The product title and its matching selling point keywords are used as input data for generating advertising text. The selling point keywords are mainly words that describe the selling points of the product and play the role of describing the selling point information of the product.

[0044] A product category selling point database can be pre-built. This database stores selling point keywords for the corresponding product category. These keywords can be extracted from the advertising text used in historical advertising campaigns for the corresponding product category and / or from the product's text information set. For example, the selling point keywords could be "variable speed," "adjustable speed," "gear shifter," "chain puller," etc. The product text information set includes, but is not limited to, any one or more of the product title, product details text, and product attribute data. The construction of this product category selling point database can be flexibly implemented by those skilled in the art based on the disclosure herein.

[0045] E-commerce platforms typically use a product category system to categorize and organize their vast array of goods, ensuring each product falls under a specific category. This category system can be multi-layered, containing multiple classification levels, each with several specific product categories. The e-commerce platform can provide a standardized template for this system, which merchants can then modify and define themselves.

[0046] Multiple selling point keywords matching the product title are determined from the selling point database of the product category to which the advertised product belongs. This matching can be based on semantic similarity. A deep learning model is used to extract the deep semantic features corresponding to each selling point keyword in the selling point database, as well as the deep semantic features of the product title of the advertised product. Vectorized representations of each selling point keyword and the product title are obtained. The vector distance between the vectorized representation of the product title and the vectorized representations of each selling point keyword is calculated as the semantic similarity. It can be understood that the semantic similarity characterizes the degree of similarity in semantic expression between the product title and each selling point keyword. Therefore, based on the semantic similarity corresponding to each selling point keyword, multiple selling point keywords with semantic similarity exceeding a preset threshold can be determined from the selling point database as selling point keywords matching the product title. The preset threshold can be set as needed by those skilled in the art. The deep learning model mentioned can be a deep semantic learning-based network model in the field of NLP (Natural Language Processing) suitable for extracting semantic features from text. Specifically, it adopts the open-source framework Sentence Transformers, which provides a large number of pre-trained and converged Transformer models, such as BERT, RoBERTa, XLM-RoBERTa, and MPNet. When calculating the vector distance, it can be implemented using any of the large-scale vector retrieval engines such as Faiss, ElasticSearch, and Milvus, or any existing algorithm such as cosine similarity, dot product, Manhattan distance, or Euclidean distance can be used.

[0047] Step S1200: Construct the first sentence pair by combining the product title with each selling point keyword; input each first sentence pair into a preset first half sentence generation model to obtain the corresponding first half sentence text, and construct a first half sentence set.

[0048] The first half-sentence generation model employs an Encoder-Decoder structure. The encoder is suitable for learning vector representation and multi-layer semantic space transformation of the input text and can be selected from various known models, including but not limited to RNN, BiLSTM, BiGRU, RoBERTa, ALBert, ERNIE, and BERT-WWM. The decoder is suitable for learning and adjusting the state information of the hidden layer at the current time step using the encoded vector sequence of the preceding text at the current time step and the state information of the last hidden layer of the encoder, through multi-layer self-attention and cross-self-attention mechanisms. Selectable autoregressive pre-trained models include, but are not limited to, GPT2, XLNet, and DistilGPT2, and classic models such as RNN, LSTM, and GRU can also be used. In one embodiment, a Transformer model can be used, providing both an encoder and a decoder, as the selection of the first half-sentence generation model.

[0049] It is understood that the first half of the sentence generation model can be pre-trained until it converges, and can learn to generate the corresponding first half of the sentence from the input text, so that it can be put into use in this step.

[0050] The product title is concatenated with each selling point keyword to construct multiple first sentence pairs. Each first sentence pair is then input into a pre-trained first half sentence generation model that has converged.

[0051] For a single first sentence pair, the encoder of the first half sentence generation model encodes the first sentence pair, extracts the deep semantic features corresponding to the first sentence pair, and obtains the corresponding first encoding vector sequence. Each element in the first encoding vector sequence is the vectorized representation of the corresponding word in the first sentence pair. The word can be determined by segmenting the first sentence pair.

[0052] Furthermore, the decoder of the first half-sentence generation model performs autoregressive decoding using the first encoded vector sequence as prefix information. After passing through a preset vocabulary classifier, it predicts the first text vector corresponding to each prefix information step by step. Then, it updates the prefix information with the words determined based on the first text vector to continue predicting the first text vector of the next time step, thus realizing autoregressive decoding. After the autoregressive decoding is completed, the words corresponding to each time step can be determined, thereby constructing the first half-sentence text.

[0053] The vocabulary classifier maps the hidden layer vectors obtained from the decoder to a reference vocabulary referenced by the first-sentence generation model, thereby obtaining a probability distribution mapped to each word in the reference vocabulary as a text vector. The vocabulary classifier can be constructed using the Softmax function, which is connected after the last hidden layer of the decoder. It predicts the word corresponding to the current time step and calculates the probability distribution of the corresponding hidden layer vector mapped to each word in the entire reference vocabulary, thus obtaining a text vector.

[0054] Based on the above, it can be understood that by using a pre-trained, converged first-sentence generation model with each first sentence pair as input, the first-sentence text corresponding to each first sentence pair can be obtained. It is recommended to select multiple first-sentence texts with significant differences from the first-sentence texts corresponding to each first sentence pair to construct a first-sentence set. This effectively reduces the number of first-sentence texts while ensuring the diversity of first-sentence texts remains unchanged, thereby improving the efficiency of subsequent operations. Subsequently, diverse second-sentence texts can be generated based on the diverse first-sentence texts in this first-sentence set.

[0055] Specifically, in one embodiment, the filtering can employ an algorithm based on character-level similarity calculation, including but not limited to: Levenshtein distance, Jaccard distance, BLEU, and ROUGE series (ROUGE-N, ROUGE-L, ROUGE-W, ROUGE-S), to calculate the similarity between the corresponding first half of the text for each first sentence pair, thereby filtering out multiple first half texts with similarity below a preset threshold, i.e., those with significant differences. The preset threshold can be set as needed by those skilled in the art.

[0056] In another embodiment, the deep learning model can be used to extract deep semantic features of the first half of the text corresponding to each first sentence pair based on the semantic level, obtain the corresponding vectorized representation, and calculate the vector distance between the vectorized representations corresponding to each first half of the text as the similarity, thereby filtering out multiple first half of the text with similarity below a preset threshold, i.e., those with large differences. The preset threshold can be set as needed by those skilled in the art.

[0057] Step S1300: After concatenating the product title and each first half of the text in the first half of the sentence set, construct a second sentence pair with each of the selling point keywords, input each second sentence pair into the preset second half of the sentence generation model to obtain the corresponding second half of the sentence text, and construct a second half of the sentence set.

[0058] The second half of the sentence generation model also adopts an Encoder-Decoder structure. It can directly use the same first half of the sentence generation model that has not been trained, or select the corresponding encoder and decoder as the second half of the sentence generation model as revealed in step S1200. Then, the second half of the sentence generation model is trained in advance with the corresponding training set until it converges, so that it can learn the ability to generate the corresponding second half of the sentence text from the input text, and can be put into use in this step.

[0059] After concatenating the product title and each first half of the ...

[0060] For a single second sentence pair, the encoder of the second half sentence generation model encodes the second sentence pair, extracts the deep semantic features corresponding to the second sentence pair, and obtains the corresponding second encoding vector sequence. Each element in the second encoding vector sequence is the vectorized representation of the corresponding word in the second sentence pair. The word can be determined by segmenting the second sentence pair.

[0061] Furthermore, the decoder of the second half sentence generation model performs autoregressive decoding using the second encoded vector sequence as prefix information. After passing through a preset vocabulary classifier, it predicts the second text vector corresponding to each prefix information step by step. Then, it updates the prefix information with the words determined by the second text vector to continue predicting the second text vector of the next time step, thus realizing autoregressive decoding. After the autoregressive decoding is completed, the words corresponding to each time step can be determined, thereby constructing the second half sentence text.

[0062] The vocabulary classifier maps the hidden layer vectors obtained from the decoder to a reference vocabulary referenced by the second-sentence generation model, thereby obtaining a probability distribution mapped to each word in the reference vocabulary as a text vector. The vocabulary classifier can be constructed using the Softmax function, which is connected after the last hidden layer of the decoder. It predicts the word at the current time step and calculates the probability distribution of the corresponding hidden layer vector mapped to each word in the entire reference vocabulary, thus obtaining a text vector.

[0063] Based on the above, it can be understood that by using a pre-trained, converged second-sentence generation model as input, the second-sentence text corresponding to each second-sentence pair can be obtained, thus constructing a set of second-sentences.

[0064] Step S1400: Concatenate the first half of the text in the first half of the sentence set with the second half of the text in the second half of the sentence set.

[0065] Each first half of the text in the set of first half sentences is concatenated with the corresponding second half of the text in the set of second half sentences, which are used as partial inputs to the second half sentence generation model, to obtain multiple complete advertising texts. Further, referring to step 1200, several advertising texts with significant differences are selected from these complete texts and pushed to users of the online store specifying the advertised product. This effectively reduces the number of advertising texts while maintaining their diversity, making it easier for users to choose the advertising texts they need.

[0066] As can be seen from the typical embodiments of this application, the technical solution of this application has many advantages, including but not limited to the following aspects:

[0067] This application implements a two-stage generation scheme. First, multiple first-sentence pairs are constructed using the product title and its matching selling point keywords. A first-sentence generation model is used to generate corresponding first-sentence texts. Then, the product title and selling point keywords are combined with each first-sentence text to construct a second-sentence pair. A second-sentence generation model is used to obtain corresponding second-sentence texts. Finally, each first-sentence text is concatenated with its corresponding second-sentence text to obtain complete advertising texts. On the one hand, the selling point features of the advertised product are introduced in both stages of this two-stage generation process, ensuring that both the generated first-sentence and second-sentence texts maintain a strong correlation with the selling point features and accurately describe the selling points of the advertised product. Thus, the description of the complete advertising text can fully focus on the selling points of the advertised product. On the other hand, the two-segment generation scheme achieves intelligent generation. The generation of the second half of the text depends on the first half of the text as input, and the two still maintain semantic coherence, which can ensure that the expression of the resulting advertising text is fluent. Moreover, it can generate diverse first half of the text based on multiple selling points and product titles related to the advertised product, and then generate diverse second half of the text accordingly, so that the final advertising text is rich enough to provide sufficient reference and effectively assist in the creation of advertising text.

[0068] Please see Figure 2 In a further embodiment, before step S1100, the step of obtaining the product title of the advertised product, the following steps are also included:

[0069] Step S1001: Obtain the historical advertising texts corresponding to multiple historical advertising products according to product categories, and statistically select the benchmark keywords carrying the selling point information of the corresponding products in the historical advertising texts for each product category.

[0070] From the product category systems built by various online stores on e-commerce platforms, different product categories are identified. Multiple historical advertising products corresponding to each product category are obtained from the advertising system provided by the e-commerce platform. Historical advertising texts used during their historical advertising campaigns are then selected based on relevant advertising performance data, filtering out historical advertising texts that meet preset conditions for optimal advertising performance. Furthermore, for each product category, through statistical analysis of a large number of these historical advertising texts, selling point keywords corresponding to the selling points of the historically advertised products are determined. The advertising performance data includes, but is not limited to, click-through rate (CTR), conversion rate, and ROAS (Return on Ad Spend). Those skilled in the art can preset the conditions for optimal advertising performance based on any one or more of these data, such as CTR and conversion rate exceeding 85%, ROAS greater than 4, etc.

[0071] In one implementation, for each product category, the entire set of historical advertising texts selected from the database is first segmented into words, resulting in various word segmentation sets. Then, word frequency statistics are performed on each word in all word segmentation sets, counting the frequency of each word appearing across all word segmentation sets. Generally, a higher word frequency indicates that the corresponding word is used more frequently to describe the product's selling points, and thus has a greater chance of becoming a selling point keyword. Based on this principle, after determining the word frequency corresponding to each word, a subset of words can be selected as selling point keywords according to a preset threshold. These selling point keywords are then used as benchmark keywords to subsequently expand and find more selling point keywords.

[0072] Step S1002: Extract extended keywords with similar semantics from the product text information set corresponding to the corresponding product category based on the benchmark keywords. The product text information set contains text information corresponding to multiple historical advertising products under the corresponding product category. The text information includes any one or more of the following: product title, product details text, and product attribute data.

[0073] The number of selling point keywords obtained from multiple historical advertising texts based on word frequency statistics is relatively small. In order to make the number of selling point keywords corresponding to the selling point information of historical advertising products sufficiently rich, we can use the selling point keywords determined from multiple historical advertising texts as benchmark keywords, and search for semantically similar selling point keywords in a given product text information set.

[0074] The product text information set can be extracted from the product database of online stores on e-commerce platforms. The selection of corresponding text information can be any one or more of the product title, product details text, product attribute data, product category, etc., so that the product text information set contains a sufficient amount of text information corresponding to multiple products.

[0075] To facilitate vocabulary expansion, each text message in the product text information set can be segmented first. Then, based on the semantic similarity between each segment and each benchmark keyword, a selling point keyword semantically similar to the benchmark keyword is determined as the extended keyword for the corresponding product category according to a preset threshold. The specific implementation of the semantic similarity can be referred to step S1100. Example: The benchmark keyword can be: transmission, speed, brake, auxiliary system, foldable, driving range, applicable terrain, battery, motor, etc. Extended keywords semantically similar to the benchmark keyword, taking "transmission" as an example, can be: transmission, adjustable speed, gearbox, chain puller, etc.

[0076] Step S1003: Construct a selling point database corresponding to each product category, which stores a set of selling point keywords corresponding to each benchmark keyword under the corresponding product category. Each set of selling point keywords includes its corresponding benchmark keyword and extended keywords obtained based on the benchmark keyword.

[0077] After identifying multiple extended keywords, a database of selling points corresponding to each product category can be constructed. This database stores the baseline keywords extracted from historical advertising texts, as well as extended keywords that are semantically similar to the corresponding baseline keywords. This creates a set of selling point keywords. The baseline keywords in each set of selling point keywords serve as references, allowing the matching of product titles in subsequent advertisements to first match the baseline keywords in each set of selling point keywords. After identifying the matching baseline keywords, the database then matches them with the extended keywords associated with those baseline keywords, thereby determining the selling point keywords that match the product title. Since this process only matches extended keywords that are semantically similar to the baseline keywords that match the product title, matching efficiency can be effectively improved.

[0078] This embodiment reveals that in the process of constructing the selling point database under each product category, multiple basic keywords are first determined by referring to the word frequency statistics of historical advertising texts. Then, each basic keyword is expanded by referring to the product text information set to obtain corresponding extended keywords. Finally, each basic keyword is associated with its extended keywords to construct a corresponding set of selling point keywords. It can be understood that the entire technology can be implemented automatically without the introduction of human factors, with low implementation costs and good results.

[0079] Please see Figure 3 In a further embodiment, before step S1100, the step of obtaining the product title of the advertised product, the following steps are also included:

[0080] Step S1010: Obtain the product title and its historical advertising text corresponding to the historical advertising product; determine the first selling point keyword in the historical advertising text based on the selling point keywords in the selling point database corresponding to the product category of the historical advertising product.

[0081] Obtain sufficient product titles corresponding to historical advertising products, as well as the historical advertising text used by each historical advertising product during its historical advertising campaigns. Furthermore, based on the selling point keywords stored in the selling point database corresponding to the product category to which each historical advertising product belongs, use a direct text matching method to determine the first matching selling point keyword in the historical advertising text of each historical advertising product as the first selling point keyword.

[0082] Step S1020: Divide the historical advertising text into the first half of the text before the first selling point keyword and the second half of the text after the first selling point keyword;

[0083] After determining the first selling point keyword corresponding to each historical advertisement text, in one embodiment, the historical advertisement text is positionally encoded based on characters to obtain the starting position code and ending position code corresponding to the first selling point keyword for each historical advertisement text. Thus, based on the starting position code, the first half of the text before the first selling point keyword can be segmented for the corresponding historical advertisement text, and based on the ending position code, the second half of the text after the first selling point keyword can be segmented for the corresponding historical advertisement text. For example, a historical advertisement text might be: "Swift and quick gear change, stable gear shifting, one key for precise and easy shifting," with the corresponding first selling point keyword being "gear change." If the position coding starts from 0, then the starting position code corresponding to the first selling point keyword would be 16, and the ending position code would be 26. It is evident that position coding eliminates the need for manual segmentation, effectively achieving automated and accurate segmentation of the first and second halves of the historical advertisement text.

[0084] Step S1030: Take the product title and its first selling point keyword from the historical advertising text as input to train the first half sentence generation model to generate the first half sentence text;

[0085] For each of the aforementioned historical advertising texts, the product title is concatenated with its first selling point keyword to form a first sample sentence pair, which serves as a single first training sample. This concatenation can use special characters such as [SEP]. For example, "Product Title [SEP] First Selling Point Keyword". The special character [SEP] helps the first-sentence generation model distinguish between product information preceding the special character and selling point control information following it. It can be understood that the selling point control information guides the model to generate the selling point semantics required for the first-sentence text. Thus, a training set is constructed from each of these first training samples to train the first-sentence generation model. After convergence, the model learns the ability to generate the corresponding first-sentence text by concatenating the product title with the corresponding selling point keyword as input. The specific training process will be further revealed in subsequent embodiments; this step will not be discussed further here.

[0086] Step S1040: Take the product title of the historical advertising text, its first selling point keyword, and the second sample sentence pair consisting of the first half of the sentence text as input, and train the second half sentence generation model to generate the second half sentence text.

[0087] For each of the aforementioned historical advertising texts, the product title is concatenated with the corresponding first half of the text generated by the first half of the sentence generation model, which has been trained to convergence. Then, the first selling point keyword is concatenated to form a second sample sentence pair, which serves as a single second training sample. This concatenation can use special characters such as [SEP]. For example, "Product Title [SEP] First Selling Point Keyword (First Half of Sentence)". The special character [SEP] helps the second half of the sentence generation model distinguish that the part before the special character is product information, and the part after is contextual information and selling point control information. It can be understood that the selling point control information guides the model to generate the selling point semantics required for the second half of the sentence. Thus, a training set is constructed using each of the second training samples to train the second half of the sentence generation model. After training to convergence, the second half of the sentence generation model learns the ability to generate the corresponding second half of the text by concatenating the product title with the corresponding first half of the text and then concatenating the corresponding selling point keyword as input. The specific training process will be further revealed in subsequent embodiments; this step will not be discussed further here.

[0088] This embodiment reveals the construction of training samples corresponding to the first half sentence generation model and the second half sentence generation model. In this process, a positional encoding method is adopted, which eliminates the need for human intervention, enables automated implementation, reduces implementation costs, and achieves better results.

[0089] Please see Figure 4In a further embodiment, step S1010, obtaining the product title and its historical advertising text corresponding to the historical advertised product, and determining the first selling point keyword in the historical advertising text based on the selling point keywords in the selling point database corresponding to the product category of the historical advertised product, includes the following steps:

[0090] Step S1011: Obtain the historical advertising text and advertising performance data corresponding to multiple historical advertising products by product category;

[0091] From the product category system built by each online store on the e-commerce platform, we determine the different product categories. From the advertising system provided by the e-commerce platform, we obtain multiple historical advertising products corresponding to each product category, the historical advertising text used when advertising was placed in the past, and the advertising performance data.

[0092] Step S1012: Select the historical advertising text corresponding to each advertising performance data that meets the preset conditions as the target advertising text, and obtain the product title and product category of the corresponding historical advertising product.

[0093] Furthermore, based on the advertising performance data, historical advertising texts that meet the preset conditions for better advertising performance are selected as target advertising texts. Further, the product titles and product categories of the corresponding historical advertising products are obtained. The advertising performance data includes, but is not limited to, click-through rate (CTR), conversion rate, and ROAS (Return on Ad Spend). Those skilled in the art can preset the preset conditions for better advertising performance based on any one or more of the advertising performance data, such as CTR and conversion rate exceeding 85%, ROAS greater than 4, etc.

[0094] Step S1013: For each target advertisement text, determine the first selling point keyword that matches it from the selling point database of the product category to which its historical advertisement products belong.

[0095] For specific implementation details, please refer to step S1010. This step will not be elaborated upon here.

[0096] In this embodiment, historical advertising texts with better advertising performance are selected as target advertising texts by using the advertising performance data corresponding to historical advertising texts. It can be understood that the selling point information of the historical advertising products described in the target advertising texts is more accurate and is well recognized by the actual market. This makes the determined matching selling point keywords more practical, which helps to ensure the accuracy of the description of the product's selling points by the training samples corresponding to the first half sentence generation model and the second half sentence generation model prepared with the target advertising text and its selling point keywords.

[0097] Please see Figure 5 In a further embodiment, step S1030, taking the product title of the historical advertising text and its first selling point keyword as input, and training the first half-sentence generation model to generate the first half-sentence text, includes the following steps:

[0098] Step S1031: Obtain a single first training sample from the prepared training set. The first training sample includes the first sample sentence pair of a single pre-selected historical advertising text.

[0099] The implementation of the pre-preparation of the training set can be referred to step S1030, which will not be described in detail here.

[0100] Step S1032: Input the first training sample into the first half sentence generation model, and predict each word in the corresponding first half sentence text by time-step autoregression.

[0101] The first half-sentence generation model uses the first training sample as input to segment the first training sample into words, obtaining a segmented text sequence containing each segmented word element in the first sample sentence pair. Referring to the reference vocabulary used by the first half-sentence generation model, each segmented word element in the segmented text sequence is converted into the encoding feature of the corresponding word element in the reference vocabulary. These encoding features are constructed into a corresponding first encoding vector, which is input to the encoder of the first half-sentence generation model for inference. In one embodiment, the positional encoding of each segmented word element can be further combined with the encoding features of each segmented word element in the first sample sentence pair to enrich the reference information of the encoding vector.

[0102] The encoder of the first half sentence generation model, as described in the previous embodiments, can be implemented using a bidirectional feature representation encoding model. Therefore, it is suitable for extracting deep semantic information by referring to the context information of the input first encoding vector to obtain the first encoding vector sequence. The context information includes the encoding features of the product itself information, i.e., the product title, and the encoding features of the selling point control information, i.e., the selling point keywords, in the first sample sentence pair. Those skilled in the art should know that when extracting deep semantic information from the context information, the extraction of some deep semantic information can refer to the encoding features of the selling point control information and focus on the corresponding selling point semantics to extract it.

[0103] Furthermore, the decoder of the first half-sentence generation model takes the prefix information formed by the first encoded vector sequence as initial input. Based on the multi-head attention mechanism, it performs serial decoding, decoding step by step. The decoded results are then classified and mapped by a vocabulary classifier to obtain the probability distribution of each word in the reference vocabulary, forming a text vector. Based on the classification probability in the text vector, the word corresponding to the current time step can be selected. The encoded feature of this word in the reference vocabulary is then appended to the prefix information to achieve autoregression. Then, the decoding continues in the next time step based on the updated prefix information. This process continues until the end symbol is encountered, completing the prediction of all text vectors, thus completing the prediction of each word in the first half-sentence text and obtaining the corresponding first half-sentence text. The selection is usually based on choosing the word with the highest probability in the text vector. Of course, those skilled in the art can also use other decoding strategies to determine the word selection for each time step, such as the Top_K algorithm, the cluster search algorithm, the greedy search algorithm, etc.

[0104] Step S1033: Using the first half of the historical advertising text included in the first training sample as the first supervision label, determine the loss value of the first half of the text predicted by the first half of the sentence generation model, update the weights of the first half of the sentence generation model when the loss value does not reach the preset threshold, and continue to call other training samples to perform iterative training until the model converges.

[0105] In one embodiment, a preset cross-entropy loss function is invoked. This function can be flexibly set by those skilled in the art based on prior knowledge or experimental experience. The first half of the historical advertising text included in the first training sample is used as the first supervision label. The cross-entropy loss value of the first half of the text is calculated. When the loss value reaches a preset threshold, it indicates that the first half of the sentence generation model has been trained to a convergent state, and the model training can be terminated. When the loss value does not reach the preset threshold, it indicates that the model has not converged. Therefore, the model is updated with gradients based on the loss value. Typically, the weight parameters of each part of the model are corrected through backpropagation to make the model closer to convergence. Then, the next training sample in the training set is called to iteratively train the model until the model is trained to a convergent state.

[0106] In another embodiment, a preset cross-entropy loss function is invoked. This function can be flexibly set by those skilled in the art based on prior knowledge or experimental experience. The first half of the historical advertising text included in the first training sample is used as the reversed version of the first half of the text as the first supervision label. The cross-entropy loss value of the first half of the text is calculated. When the loss value reaches a preset threshold, it indicates that the first half of the sentence generation model has been trained to a convergent state, and the model training can be terminated. If the loss value does not reach the preset threshold, it indicates that the model has not converged. Therefore, gradient updates are performed on the model based on the loss value. Typically, backpropagation is used to correct the weight parameters of each stage of the model to make the model further approach convergence. Then, the next training sample in the training set is used to iteratively train the model until it is trained to a convergent state. It can be understood that the generated first half of the text is a reversed version after such supervised training of the first half of the sentence generation model to convergence. Therefore, converting the generated reversed version of the first half of the text into a forward version can be used subsequently. This embodiment uses the reversed version of the first half of the sentence as the first supervisory label to train the first half of the sentence generation model, which to some extent makes the first half of the sentence generated by the model more fluent and reasonable.

[0107] In this embodiment, the first half of the sentence generation model is trained under supervision until convergence, so that the first half of the sentence generation model learns to generate the first half of the sentence text that accurately describes the corresponding selling points of the product by taking the product title and corresponding selling point keywords as input.

[0108] Please see Figure 6 In a further embodiment, step S1040, using the product title of the historical advertising text, its first selling point keyword, and the first half of the text to form a second sample sentence pair as input, trains the second half of the sentence generation model to generate the second half of the sentence text, including the following steps:

[0109] Step S1041: Obtain a single second training sample from the prepared training set, the second training sample comprising a single pre-selected second sample sentence pair of historical advertising text;

[0110] The implementation of the pre-preparation of the training set can be referred to step S1040, which will not be described in detail here.

[0111] Step S1042: Input the second training sample into the second half sentence generation model, and predict each word in the corresponding second half sentence text by time step autoregression;

[0112] The second training sample is used as input to the second sentence generation model. The second training sample is segmented into words to obtain a segmented text sequence, which contains each segmented word element in the second sample sentence pair. The second sentence generation model can refer to the reference vocabulary to convert each segmented word element in the segmented text sequence into the encoding feature of the corresponding word element in the reference vocabulary. These encoding features are constructed into a corresponding second encoding vector, which is input to the encoder of the second sentence generation model for inference. In one embodiment, the position encoding of each segmented word element can be further combined with the encoding features of each segmented word element in the second sample sentence pair to enrich the reference information of the encoding vector.

[0113] The encoder of the second half sentence generation model, as described in the previous embodiments, can be implemented using a bidirectional feature representation encoding model. Therefore, it is suitable for extracting deep semantic information by referring to the context information of the input second encoding vector to obtain the second encoding vector sequence. The context information includes the encoding features of the product itself in the second sample sentence pair, i.e., the product title, the encoding features of the preceding sentence text, and the encoding features of the selling point control information, i.e., the selling point keywords. Those skilled in the art should know that when extracting deep semantic information from the context information, the extraction of some deep semantic information can refer to the encoding features of the selling point control information, focusing on the corresponding selling point semantics to extract it.

[0114] Furthermore, the decoder of the second half-sentence generation model takes the prefix information formed by the second encoded vector sequence as initial input. Based on the multi-head attention mechanism, it performs serial decoding, decoding step by step. The decoded results are then classified and mapped by a vocabulary classifier to obtain the probability distribution of each word in the reference vocabulary, forming a text vector. Based on the classification probability in the text vector, the word corresponding to the current time step can be selected. The encoded feature of this word in the reference vocabulary is then appended to the prefix information to achieve autoregression. Then, the decoding continues in the next time step based on the updated prefix information. This process is repeated until the end symbol is encountered, completing the prediction of all text vectors, thus completing the prediction of each word in the second half-sentence text and obtaining the corresponding second half-sentence text. The selection is usually based on choosing the word with the highest probability in the text vector. Of course, those skilled in the art can also use other decoding strategies to determine the word selection for each time step, such as the Top_K algorithm, the cluster search algorithm, the greedy search algorithm, etc.

[0115] Step S1043: Using the second half of the historical advertising text included in the second training sample as the second supervision label, determine the loss value of the second half of the text predicted by the second half of the sentence generation model. When the loss value does not reach the preset threshold, perform weight update on the second half of the sentence generation model, and continue to call other training samples to perform iterative training until the model converges.

[0116] In one embodiment, a preset cross-entropy loss function is invoked. This function can be flexibly set by those skilled in the art based on prior knowledge or experimental experience. The second half of the historical advertising text included in the second training sample is used as the second supervision label to calculate the cross-entropy loss value of the second half of the text. When the loss value reaches a preset threshold, it indicates that the second half of the sentence generation model has been trained to a convergent state, and the model training can be terminated. When the loss value does not reach the preset threshold, it indicates that the model has not converged. Therefore, the model is updated with gradients based on the loss value. Typically, the weight parameters of each part of the model are corrected through backpropagation to make the model further approach convergence. Then, the next training sample in the training set is called to iteratively train the model until the model is trained to a convergent state.

[0117] In this embodiment, the second half sentence generation model is trained under supervision until convergence, enabling the model to learn to generate accurate second half sentence text describing the corresponding product selling points by taking the product title, the first half of the text, and the corresponding selling point keywords as input.

[0118] Please see Figure 7 This application provides an advertising text generation device, which is a functional embodiment of the advertising text generation method of this application. The device includes a data acquisition module 1100, a first-sentence generation module 1200, a second-sentence generation module 1300, and an advertising text generation module 1400. The data acquisition module 1100 is used to acquire the product title of the advertised product and determine multiple selling point keywords matching the product title from a selling point database of the product category to which the advertised product belongs. The first-sentence generation module 1200 is used to associate the product title with each selling point keyword. The keyword construction forms a first sentence pair. Each first sentence pair is input into a preset first-sentence generation model to obtain the corresponding first-sentence text, thus constructing a first-sentence set. The second-sentence generation module 1300 is used to concatenate the product title and each first-sentence text in the first-sentence set, and then construct a second sentence pair with each of the selling point keywords. Each second sentence pair is input into a preset second-sentence generation model to obtain the corresponding second-sentence text, thus constructing a second-sentence set. The advertising text generation module 1400 is used to concatenate the first-sentence text in the first-sentence set with the second-sentence text in the second-sentence set.

[0119] In a further embodiment, before the data acquisition module 1100, the system further includes: a benchmark keyword selection module, used to acquire historical advertising texts corresponding to multiple historical advertising products according to product categories, and to statistically select benchmark keywords carrying the selling point information of the corresponding products in the historical advertising texts for each product category; an extended keyword selection module, used to extract extended keywords with semantically similar meanings from the product text information set corresponding to the benchmark keywords from the product text information set corresponding to the corresponding product category, wherein the product text information set contains text information corresponding to multiple historical advertising products under the corresponding product category, and the text information includes any one or more of the product title, product details text, and product attribute data; and a selling point database construction module, used to construct a selling point database corresponding to each product category, wherein the database stores a set of selling point keywords corresponding to each benchmark keyword under the corresponding product category, and each set of selling point keywords contains its corresponding benchmark keyword and extended keywords obtained based on the benchmark keyword.

[0120] In a further embodiment, before the data acquisition module 1100, the system further includes: a selling point keyword determination module, used to acquire the product title corresponding to the historical advertised product and its historical advertising text, and determine the first selling point keyword in the historical advertising text based on the selling point keywords in the selling point database corresponding to the product category of the historical advertised product; a text segmentation module, used to segment the historical advertising text into a first half of the text before the first selling point keyword and a second half of the text after the first selling point keyword; a first half of the sentence generation training module, used as input a first sample sentence pair consisting of the product title and the first selling point keyword of the historical advertising text, and trained the first half of the sentence generation model to generate the first half of the sentence text; and a second half of the sentence generation training module, used as input a second sample sentence pair consisting of the product title, the first selling point keyword, and the first half of the sentence text of the historical advertising text, and trained the second half of the sentence generation model to generate the second half of the sentence text.

[0121] In a further embodiment, the text acquisition module includes: a product information acquisition submodule, used to acquire historical advertising texts and their advertising performance data corresponding to multiple historical advertising products according to product categories; a text filtering submodule, used to filter out the historical advertising texts corresponding to each advertising performance data that meet preset conditions as target advertising texts, and acquire the product titles and product categories of the corresponding historical advertising products; and a selling point keyword determination submodule, used to determine the first selling point keyword that matches each target advertising text from the selling point database of the product category to which its historical advertising products belong.

[0122] In a further embodiment, the first half-sentence generation training module includes: a first training sample acquisition submodule, used to acquire a single first training sample from a prepared training set, the first training sample including a single pre-selected first sample sentence pair of historical advertising text; a first autoregressive prediction submodule, used to input the first training sample into the first half-sentence generation model, and autoregressively predict each word in the corresponding first half-sentence text step by step; and a first iterative training submodule, used to use the first half-sentence text of the historical advertising text included in the first training sample as a first supervision label, determine the loss value of the first half-sentence text predicted by the first half-sentence generation model, update the weights of the first half-sentence generation model when the loss value does not reach a preset threshold, and continue to call other training samples to perform iterative training until the model converges.

[0123] In a further embodiment, the second half-sentence generation training module includes: a second training sample acquisition submodule, used to acquire a single second training sample from a prepared training set, the second training sample including a single pre-selected second sample sentence pair of historical advertising text; the second training sample acquisition submodule is used to input the second training sample into the second half-sentence generation model, and autoregressively predict each word in the corresponding second half-sentence text step by step; a second iterative training submodule is used to use the second half-sentence text of the historical advertising text included in the second training sample as a second supervision label, determine the loss value of the second half-sentence text predicted by the second half-sentence generation model, update the weights of the second half-sentence generation model when the loss value does not reach a preset threshold, and continue to call other training samples to perform iterative training until the model converges.

[0124] In a further embodiment, the first iterative training submodule includes: the first supervision label is the reverse version of the first half of the historical advertising text included in the first training sample.

[0125] To address the aforementioned technical problems, embodiments of this application also provide computer equipment. For example... Figure 8The diagram shows 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. The computer-readable storage medium stores an operating system, a database, and computer-readable instructions. The database may store a sequence of control information. When the computer-readable instructions are executed by the processor, they enable the processor to implement an advertising text generation method. The processor of the computer device provides computing and control capabilities, supporting the operation of the entire computer device. The memory of the computer device may store computer-readable instructions, which, when executed by the processor, enable the processor to execute the advertising text generation method of this application. The network interface of the computer device is used for communication with a terminal. Those skilled in the art will understand that… Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0126] In this embodiment, the processor is used to execute... Figure 7 The specific functions of each module and its sub-modules are defined within the device. The memory stores the program code and various types of data required to execute these modules or sub-modules. The network interface is used for data transmission between the user terminal and the server. In this embodiment, the memory stores the program code and data required to execute all modules / sub-modules in the advertising text generation device of this application. The server can call the server's program code and data to execute the functions of all sub-modules.

[0127] This application also provides a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the advertising text generation method of any embodiment of this application.

[0128] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. This computer program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0129] In summary, this application, on the one hand, adopts a two-stage generation scheme. In both stages, the selling points of the advertised product are incorporated, ensuring that both the first and second halves of the generated text accurately describe the product's selling points. This results in a complete advertising text that fully focuses on the product's key selling points. On the other hand, the two-stage generation scheme achieves intelligent generation. Based on multiple selling point features related to the advertised product, this scheme generates rich advertising text, providing ample reference and effectively assisting in the creation of advertising text.

[0130] Those skilled in the art will understand that the steps, measures, and solutions in the various operations, methods, and processes discussed in this application can be alternated, modified, combined, or deleted. Furthermore, other steps, measures, and solutions in the various operations, methods, and processes discussed in this application can also be alternated, modified, rearranged, decomposed, combined, or deleted. Furthermore, steps, measures, and solutions in the prior art that are similar to those disclosed in this application can also be alternated, modified, rearranged, decomposed, combined, or deleted.

[0131] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. An advertisement text generation method characterized by comprising: The method comprises the following steps: obtaining a product title of an advertised product, and determining a plurality of selling point keywords matched with the product title from a selling point database of a product category to which the advertised product belongs; constructing the product title and each selling point keyword into a first sentence pair, inputting each first sentence pair into a preset first half sentence generation model to obtain a corresponding first half sentence text, and constructing a first half sentence set; splicing the product title and each first half sentence text in the first half sentence set, and constructing a second sentence pair with each selling point keyword, inputting each second sentence pair into a preset second half sentence generation model to obtain a corresponding second half sentence text, and constructing a second half sentence set; splicing the first half sentence text in the first half sentence set and the second half sentence text in the second half sentence set; Before obtaining the product title of the advertised product, the method comprises the following steps: obtaining a product title of an advertised product, and determining a plurality of selling point keywords matched with the product title from a selling point database of a product category to which the advertised product belongs; determining a first selling point keyword in a historical advertisement text according to a selling point keyword in a selling point database corresponding to a historical advertised product; segmenting the historical advertisement text into a first half sentence text before the first selling point keyword and a second half sentence text after the first selling point keyword, comprising: obtaining a single second training sample from a prepared training set, wherein the second training sample comprises a second sample sentence pair of a single preselected historical advertisement text; inputting the second training sample into the second half sentence generation model to predict each word element in the corresponding second half sentence text by time step autoregression; using the second half sentence text of the historical advertisement text included in the second training sample as a second supervision label to determine a loss value of the second half sentence text predicted by the second half sentence generation model, and updating the weights of the second half sentence generation model when the loss value does not reach a preset threshold, and continuing to call other training samples for iterative training until the model converges; inputting a first sample sentence pair formed by the product title of the historical advertisement text and the first selling point keyword into the first half sentence generation model to generate the first half sentence text; 2. The advertisement text generating method according to claim 1, characterized by, inputting a second sample sentence pair formed by the product title of the historical advertisement text, the first selling point keyword and the first half sentence text into the second half sentence generation model to generate the second half sentence text. Before the step of obtaining the product title of the advertised product, the method further comprises the following steps: obtaining a plurality of historical advertisement texts corresponding to historical advertised products according to product categories, and selecting a reference keyword carrying selling point information of a corresponding product for each product category; extracting an extended keyword similar in semantic to each reference keyword from a product text information set corresponding to the corresponding product category, wherein the product text information set comprises text information corresponding to a plurality of historical advertised products in the corresponding product category, and the text information comprises any one or more of a product title, a product detail text and product attribute data; constructing a database of selling points corresponding to each commodity category, wherein a set of selling point keywords corresponding to each reference keyword under the corresponding commodity category is stored, and each set of selling point keywords contains the corresponding reference keyword and the extended keywords obtained according to the reference keyword.

3. The advertisement text generating method according to claim 1, characterized by, In the step of obtaining the first selling point keyword in the historical advertisement text according to the selling point keywords in the selling point database corresponding to the commodity category of the historical advertisement commodity, the following steps are included: obtaining the historical advertisement texts and the advertisement launching effectiveness data of a plurality of historical advertisement commodities according to the commodity categories; screening the historical advertisement texts corresponding to each advertisement launching effectiveness data satisfying the preset condition as the target advertisement texts, obtaining the commodity titles of the corresponding historical advertisement commodities and the commodity categories to which the commodity titles belong; for each target advertisement text, determining the first matching selling point keyword from the selling point database of the commodity category to which the historical advertisement commodity belongs.

4. The advertisement text generating method according to claim 1, characterized by, training the first half sentence generation model to generate the first half sentence text by taking the first sentence pair composed of the commodity title of the historical advertisement text and the first selling point keyword as input, including the following steps: obtaining a single first training sample from the prepared training set, wherein the first training sample includes the first sample sentence pair of a single preselected historical advertisement text; inputting the first training sample into the first half sentence generation model to predict each word element in the corresponding first half sentence text by time step autoregression; using the first half sentence text of the historical advertisement text included in the first training sample as the first supervision label to determine the loss value of the first half sentence text predicted by the first half sentence generation model, and when the loss value does not reach the preset threshold, updating the weight of the first half sentence generation model, and continuing to call other training samples for iterative training until the model converges.

5. The advertisement text generating method according to claim 4, characterized by, using the first half sentence text of the historical advertisement text included in the first training sample as the first supervision label, including: the first supervision label is the reverse version of the first half sentence text of the historical advertisement text included in the first training sample.

6. An advertisement text generating apparatus characterized by comprising: including: a data acquisition module configured to obtain a commodity title of an advertisement commodity, and determine a plurality of selling point keywords matching the commodity title from a selling point database of a commodity category to which the advertisement commodity belongs; a first half sentence generation module configured to construct a first sentence pair by taking the commodity title and each selling point keyword, input each first sentence pair into a preset first half sentence generation model to obtain a corresponding first half sentence text, and construct a first half sentence set; a second half sentence generation module configured to splice the commodity title and each first half sentence text in the first half sentence set to construct a second sentence pair with each selling point keyword, input each second sentence pair into a preset second half sentence generation model to obtain a corresponding second half sentence text, and construct a second half sentence set; an advertisement text generation module configured to splice the first half sentence text in the first half sentence set and the second half sentence text in the second half sentence set; wherein, before obtaining the commodity title of the advertisement commodity, the following steps are included: The data acquisition module further comprises: a selling point keyword determination module, configured to acquire a product title corresponding to a historical advertising product and a historical advertising text of the historical advertising product, and determine a first selling point keyword in the historical advertising text according to a selling point keyword in a selling point database corresponding to a product category of the historical advertising product; and a text segmentation module, configured to segment the historical advertising text into a first half sentence text before the first selling point keyword and a second half sentence text after the first selling point keyword. The data acquisition module further comprises: a selling point keyword determination module, configured to acquire a product title corresponding to a historical advertising product and a historical advertising text of the historical advertising product, and determine a first selling point keyword in the historical advertising text according to a selling point keyword in a selling point database corresponding to a product category of the historical advertising product; and a text segmentation module, configured to segment the historical advertising text into a first half sentence text before the first selling point keyword and a second half sentence text after the first selling point keyword. The data acquisition module further comprises: a selling point keyword determination module, configured to acquire a product title corresponding to a historical advertising product and a historical advertising text of the historical advertising product, and determine a first selling point keyword in the historical advertising text according to a selling point keyword in a selling point database corresponding to a product category of the historical advertising product; and a text segmentation module, configured to segment the historical advertising text into a first half sentence text before the first selling point keyword and a second half sentence text after the first selling point keyword. The data acquisition module further comprises: a selling point keyword determination module, configured to acquire a product title corresponding to a historical advertising product and a historical advertising text of the historical advertising product, and determine a first selling point keyword in the historical advertising text according to a selling point keyword in a selling point database corresponding to a product category of the historical advertising product; and a text segmentation module, configured to segment the historical advertising text into a first half sentence text before the first selling point keyword and a second half sentence text after the first selling point keyword.

7. The advertisement text generating apparatus according to claim 6, wherein The selling point keyword determination module comprises: a product information acquisition submodule, configured to acquire historical advertising texts and advertising delivery effectiveness data of a plurality of historical advertising products according to product categories; a text screening submodule, configured to screen a historical advertising text corresponding to each advertising delivery effectiveness data satisfying a preset condition as a target advertising text, acquire a product title of a historical advertising product corresponding to the target advertising text and a product category to which the historical advertising product belongs; and a selling point keyword determination submodule, configured to determine a first selling point keyword matched with each target advertising text from a selling point database of a product category to which a historical advertising product corresponding to the target advertising text belongs.

8. The advertisement text generating apparatus according to claim 6, wherein ​ 9. A computer device comprising a central processing unit and a memory, characterized in that The central processing unit is configured to invoke a computer program stored in the memory to execute the steps of the method according to any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that, The computer program is stored in the form of computer readable instructions and is configured to execute the steps of the method according to any one of claims 1 to 5 when the computer program is invoked by a computer.

Images