Advertising copy recommendation method and its device, equipment, medium, product

Abstract

The present application discloses an advertising copy recommendation method, its device, equipment, medium, and product. The method includes: obtaining the product category and text information in the product information of the target product, where the text information includes the product description information of the target product; coarsely and multi-channel recalling the selling point copy in the selling point library according to the product category and text information to form a candidate copy set; using each selling point copy in the candidate copy set to match the text information as input, and calculating the semantic similarity score corresponding to the selling point copy and the text information by using a neural network model that has been pre-trained to convergence; finely recalling the selling point copy in the candidate copy set according to the semantic similarity score, and obtaining the finely recalled selling point copy to be pushed to the first type of users. The present application can quickly and accurately recall the selling point copy required for the recommended advertisement, facilitating users to refer to and create the selling point copy by themselves.

Description

Technical Field

[0001] The present application relates to the field of e-commerce information technology, and in particular, to an advertising copy recommendation method, a corresponding device, a computer device, a computer-readable storage medium, and a computer program product. Background Art

[0002] The advertising placement process includes three core components: product selection, targeting, and creativity. Among them, the product selection part is to select suitable products for advertising placement; targeting is mainly to determine the appropriate target audience for advertising placement; creativity determines what kind of advertising content to display to the target audience. The three complement each other and are all indispensable, jointly determining the success or failure of advertising placement. As text is a basic way of information dissemination, advertising copy has become a major part of advertising creativity. However, in real practical scenarios, there are the following problems:

[0003] 1. For different products, it is necessary to spend a lot of manpower and time to create advertising copy, which is quite inconvenient.

[0004] 2. Due to the different writing levels of the creators, the created copy is uneven, which in turn affects the overall advertising placement effect.

[0005] 3. The core of advertising copy is the product selling points. However, there are a wide variety of existing products. When the creators are not familiar with the products, it is easy to have an inaccurate grasp of the product selling points, which in turn affects the quality of the advertising copy.

[0006] In traditional technologies, there is a method for generating advertising copy based on the seq2seq model. This method extracts and encodes the product information of the product to be advertised through an encoder, and then uses a decoder to generate the advertising copy word by word for the extracted product features. The smoothness, length, and content quality of the advertising copy generated by this method are difficult to guarantee. Summary of the Invention

[0007] The primary objective of the present application is to solve at least one of the above problems, and to provide an advertising copy recommendation method, a corresponding device, a computer device, a computer-readable storage medium, and a computer program product.

[0008] To achieve the various objectives of the present application, the following technical solutions are adopted:

[0009] An advertising copy recommendation method provided to meet one of the objectives of the present application includes the following steps:

[0010] Obtain the product category and text information in the product information of the target product, where the text information includes the product description information of the target product;

[0011] Construct a candidate copy set from the selling point copy in the product selling point library by coarsely multi-channel recalling according to the product category and text information;

[0012] Using each selling point copy in the candidate copy set to match the text information as the input, calculate the semantic similarity score corresponding to the selling point copy and the text information using a neural network model that has been pre-trained to convergence;

[0013] Recall the selling point copy in the candidate copy set according to the semantic similarity score in a fine-grained manner, and push the recalled selling point copy to the first type of users.

[0014] In a further embodiment, before the step of obtaining the product category and text information in the product information of the target product, where the text information includes the product description information of the target product, the following steps are further included:

[0015] Select a part of the products corresponding to each product category on the e-commerce platform as category products according to a preset ratio;

[0016] Match the selling point copy in the advertising copy library of the e-commerce platform and / or the advertising copy library of the third-party platform according to the keyword text corresponding to the category product, where the keyword text is pre-extracted from the product information of the corresponding category product;

[0017] Construct a product selling point library for storing the mapping relationship data between the selling point copy, keyword text, and product category.

[0018] In a further embodiment, the step of constructing a candidate copy set by coarsely multi-channel recalling the selling point copy in the product selling point library according to the product category and text information includes the following steps:

[0019] Recall the selling point copy corresponding to the product category of the target product in the product selling point library through the coarse-grained first channel;

[0020] Recall the selling point copy corresponding to the keyword text that matches the text information of the target product in the product selling point library through the coarse-grained second channel;

[0021] Calculate the vector similarity between the vector corresponding to the text information of the target product and each keyword text in the product selling point library through the coarse-grained third channel, and recall the selling point copy corresponding to the keyword text in the product selling point library whose vector similarity meets the preset conditions.

[0022] In a further embodiment, the step of calculating the vector similarity between the vector corresponding to the text information of the target product and each keyword text in the product selling point library, and recalling the selling point copy corresponding to the keyword text in the product selling point library whose vector similarity meets the preset conditions includes the following steps:

[0023] Preprocess the text information of the target commodity through a preset word segmentation algorithm to obtain the corresponding segmented text;

[0024] Perform vector encoding on the segmented text and the keyword text in the commodity selling point library through a word vector algorithm to obtain the corresponding target encoding vector and key encoding vector;

[0025] Call a preset similarity function to calculate the vector similarity between the target encoding vector and the key encoding vector;

[0026] Determine the keyword text corresponding to the key encoding vector with a vector similarity that meets the preset conditions, and recall the selling point copy corresponding to this keyword text in the commodity selling point library.

[0027] In an extended embodiment, after the step of calculating the vector similarity between the vector corresponding to the text information of the target commodity and each keyword text in the commodity selling point library through a coarse-grained third channel, and recalling the selling point copy corresponding to the keyword text with a vector similarity that meets the preset conditions in the commodity selling point library, the following steps are further included:

[0028] Sort the recalled selling point copies according to the vector similarity to construct a coarsely sorted text set;

[0029] Select the selling point copies with higher rankings in the coarsely sorted text set and push them to the second type of users.

[0030] In a further embodiment, the training process of the neural network model includes the following steps:

[0031] Call a single training sample from a preset dataset to train the neural network model, and the training sample includes the text information of the preprocessed advertising commodity and its corresponding selling point copy;

[0032] Extract the deep semantic features of the text information of the training sample and its corresponding selling point copy;

[0033] Use a binary classification function to calculate the classification probability corresponding to whether the deep semantic features belong to the same sentence, as the semantic similarity score corresponding to whether the selling point copy is the next sentence of the text information;

[0034] Calculate the cross-entropy loss value corresponding to the semantic similarity score according to the supervision label corresponding to the training sample, and perform gradient update on the model based on this cross-entropy loss value until the model converges.

[0035] In an extended embodiment, after the step of finely recalling the selling point copies in the candidate copy set according to the semantic similarity score and pushing the finely recalled selling point copies to the user, the following steps are further included:

[0036] Sort the selling point copywriting in the candidate text set according to the semantic similarity score, and construct a first sorted text set;

[0037] Select the selling point copywriting with higher rankings in the first sorted text set as the selling point copywriting for fine-grained recall, and push these selling point copywriting to the first type of users.

[0038] An advertising copywriting recommendation device provided to meet one of the purposes of the present application includes: an information acquisition module, a coarse-grained recall module, a scoring calculation module, and a fine-grained recall module. Among them, the information acquisition module is used to acquire the product category and text information in the product information of the target product, and the text information includes the product description information of the target product; the coarse-grained recall module is used to coarsely recall the selling point copywriting in the product selling point library through multiple channels according to the product category and text information to form a candidate copywriting set; the scoring calculation module is used to use each selling point copywriting in the candidate copywriting set to match the text information as input, and calculate the semantic similarity score corresponding to the selling point copywriting and the text information by using a neural network model that has been pre-trained to convergence; the fine-grained recall module is used to finely recall the selling point copywriting in the candidate copywriting set according to the semantic similarity score, and obtain the selling point copywriting for fine-grained recall and push it to the first type of users.

[0039] In a further embodiment, before the information acquisition module, there is also included: a proportion selection module, which is used to select a part of the products corresponding to each product category in the e-commerce platform as category products according to a preset proportion; a matching selling point copywriting module, which is used to match the selling point copywriting in the advertising copywriting library of the e-commerce platform and / or the advertising copywriting library of the third-party platform according to the keyword text corresponding to the category product, and the keyword text is pre-extracted from the product information of the corresponding category product; a product selling point library construction module, which is used to construct a product selling point library for storing the mapping relationship data between the selling point copywriting, the keyword text, and the product category.

[0040] In a further embodiment, the coarse-grained recall module includes: a first-channel recall sub-module, which is used to recall the selling point copywriting corresponding to the product category of the target product in the product selling point library through the coarse-grained first channel; a second-channel recall sub-module, which is used to recall the selling point copywriting corresponding to the keyword text matched with the text information of the target product in the product selling point library through the coarse-grained second channel; a third-channel recall sub-module, which is used to calculate the vector similarity between the text information of the target product and the vectors corresponding to each keyword text in the product selling point library through the coarse-grained third channel, and recall the selling point copywriting corresponding to the keyword text whose vector similarity meets the preset conditions in the product selling point library.

[0041] In a further embodiment, the third-channel recall sub-module includes: a word segmentation preprocessing unit for preprocessing the text information of the target commodity through a preset word segmentation algorithm to obtain corresponding segmented text; a vector encoding unit for performing vector encoding on the segmented text and the keyword text in the commodity selling point library through a word vector algorithm to obtain corresponding target encoding vectors and key encoding vectors; a similarity calculation unit for calling a preset similarity function to calculate the vector similarity between the target encoding vector and the key encoding vector; and a selling point copywriting recall unit for determining the keyword text corresponding to the key encoding vector with a vector similarity meeting a preset condition and recalling the selling point copywriting corresponding to the keyword text in the commodity selling point library.

[0042] In an extended embodiment, after the third-channel recall sub-module, there is further included: a rough sorting sub-module for sorting the recalled selling point copywriting according to the vector similarity to construct a rough sorted text set; and a second-category user push sub-module for selecting the selling point copywriting with a higher ranking in the rough sorted text set and pushing it to second-category users.

[0043] In a further embodiment, the training process of the neural network model in the scoring calculation module includes: an implementation training sub-module for implementing training on the neural network model by calling a single training sample from a preset dataset, where the training sample includes the text information of the preprocessed advertising commodity and its corresponding selling point copywriting; a feature extraction sub-module for extracting the deep semantic features of the text information of the training sample and its corresponding selling point copywriting; a classification prediction sub-module for using a binary classification function to calculate the classification probability corresponding to whether the deep semantic features belong to the same sentence as a semantic similarity score representing whether the selling point copywriting is the next sentence of the text information; and a loss calculation sub-module for calculating the cross-entropy loss value corresponding to the semantic similarity score according to the supervision label corresponding to the training sample, and performing gradient update on the model based on the cross-entropy loss value until the model converges.

[0044] In an extended embodiment, the fine-grained recall module further includes: a first sorting module for sorting the selling point copywriting in the candidate text set according to the semantic similarity score to construct a first sorted text set; and a first-category user push module for selecting the selling point copywriting with a higher ranking in the first sorted text set as the selling point copywriting for fine-grained recall and pushing these selling point copywriting to first-category users.

[0045] The technical solution of the present application has multiple advantages, including but not limited to the following aspects:

[0046] First, recall the selling point copywriting in the product selling point library through coarse-grained multi-channel to quickly narrow down the scale of recall. Coarsely screen and lock some selling point copywriting to construct a candidate set. On this basis, further use the neural network model to finely analyze the selling point copywriting and the text information, determine the corresponding semantic similarity degree and quantify the corresponding semantic similarity score. Furthermore, the selling point copywriting in the candidate copywriting set can be recalled preferentially according to the semantic similarity score, so as to achieve fine screening and determine the final selling point copywriting. It can be seen that fine screening on the basis of coarse screening greatly improves the execution efficiency and still ensures the accuracy of recall.

[0047] Secondly, the selling point copywriting recalled with fine granularity can effectively assist users in creating the copywriting used for the products in their placed advertisements, lower the creation threshold, so that users can inspire their creation inspiration by referring to multiple selling point copywriting highly relevant to their products and create their own copywriting.

[0048] In addition, this application uses a single neural network model, takes the text information in the selling point copywriting and the product information as inputs jointly, and obtains the semantic similarity scores of both at one time, so as to conveniently select the selling point copywriting according to the semantic similarity scores. High-efficiency matching can be achieved without a complex network architecture, which is beneficial to reducing the training cost of the neural network model and improving the matching efficiency. Brief Description of the Drawings

[0049] The above and / or additional aspects and advantages of this application will become obvious and easy to understand from the following description of the embodiments in conjunction with the drawings, where:

[0050] Figure 1 is a schematic flowchart of a typical embodiment of the advertisement copywriting recommendation method of this application;

[0051] Figure 2 is a schematic flowchart of the process of constructing a product selling point library in an embodiment of this application;

[0052] Figure 3 is a schematic flowchart of the process of coarse-grained multi-channel recall in an embodiment of this application;

[0053] Figure 4 is a schematic flowchart of the process of recall according to vector similarity in an embodiment of this application;

[0054] Figure 5 is a schematic flowchart of the process of pushing to the second type of users in an embodiment of this application;

[0055] Figure 6 is a schematic flowchart of the training process of the neural network model in an embodiment of this application;

[0056] Figure 7 is a schematic flowchart of the process of pushing to the first type of users in an embodiment of this application;

[0057] Figure 8 The principle block diagram of the advertisement copy recommendation device of the present application;

[0058] Figure 9 The structural schematic diagram of a computer device adopted by the present application.

[0059] Specific implementation manners

[0060] The embodiments of the present application will be described in detail below. The examples of the embodiments are shown in the accompanying drawings, where 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 the present application and should not be construed as a limitation to the present application.

[0061] Those skilled in the art of the present technology can understand that, unless specifically stated otherwise, the singular forms "a", "an", "the", and "said" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the description of the present application means the presence of the described 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 their groups. 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 other elements, or there may also be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The phrase "and / or" used herein includes all or any unit and all combinations of one or more related listed items.

[0062] Those skilled in the art of the present technology can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as the general understanding of those of ordinary skill in the art to which the present application belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless specifically defined as here.

[0063] Those skilled in the art can understand that the "client", "terminal", and "terminal device" used herein include both devices with a wireless signal receiver that only has the ability to receive and no ability to transmit, and devices with both receiving and transmitting hardware that can perform two-way communication on a two-way communication link. Such devices may include: cellular or other communication devices such as personal computers, tablet computers, etc., which have a single-line display or a multi-line display or a cellular or other communication device without a multi-line display; PCS (Personal Communications Service), which can combine voice, data processing, fax, and / or data communication capabilities; PDA (Personal Digital Assistant), which may include a radio frequency receiver, a pager, Internet / intranet access, a web browser, a notepad, a calendar, and / or a GPS (Global Positioning System) receiver; conventional laptop and / or palm-held computers or other devices, which are conventional laptop and / or palm-held computers or other devices with and / or including a radio frequency receiver. The "client", "terminal", and "terminal device" used herein can be portable, transportable, installed in a vehicle (air, sea, and / or land), or suitable for and / or configured to run locally, and / or run in a distributed form at any other location on the earth and / or in space. The "client", "terminal", and "terminal device" used herein can also be a communication terminal, an Internet access terminal, a music / video playback terminal, such as a PDA, a MID (Mobile Internet Device), and / or a mobile phone with music / video playback function, or can also be devices such as a smart TV, a set-top box, etc.

[0064] The hardware referred to by names such as "server", "client", and "service node" in this application is essentially an electronic device with the equivalent capabilities of a personal computer, and is a hardware device with the necessary components disclosed by the von Neumann principle, including a central processing unit (including an arithmetic unit and a controller), a memory, an input device, and an output device. The computer program is stored in its memory, and the central processing unit loads the program stored in the external memory into the memory for execution, executes the instructions in the program, and interacts with the input / output devices to complete specific functions.

[0065] It should be noted that the concept of "server" as referred to in this application can similarly be extended to apply to server clusters. According to the network deployment principles understood by those skilled in the art, the various servers should be logically divided. Physically, these servers can either be independent of each other but can be invoked through interfaces, or integrated into a single physical computer or a set of computer clusters. Those skilled in the art should understand this flexibility and should not be restricted by this when implementing the network deployment method of this application.

[0066] One or several technical features of this application, unless expressly specified, can either be deployed on a server for implementation and accessed by a client remotely invoking the online service interface provided by the server, or directly deployed and run on the client for implementation and access.

[0067] The neural network models cited or possibly cited in this application, unless expressly specified, can either be deployed on a remote server and remotely invoked on the client, or deployed on a client with sufficient device capabilities for direct invocation. In some embodiments, when it runs on the client, its corresponding intelligence can be obtained through transfer learning to reduce the requirements for the client's hardware operation resources and avoid excessive occupation of the client's hardware operation resources.

[0068] All kinds of data involved in this application, unless expressly specified, can either be remotely stored on a server or stored on a local terminal device, as long as it is suitable for being invoked by the technical solution of this application.

[0069] Those skilled in the art should be aware that although the various methods of this application are described based on the same concept and thus show commonality with each other, unless otherwise specified, these methods can be executed independently. Similarly, for the various embodiments disclosed in this application, they are all proposed based on the same inventive concept. Therefore, for concepts with the same expression, as well as concepts that are only appropriately transformed for convenience although the concept expressions are different, they should be equivalently understood.

[0070] For the various embodiments to be disclosed in this application, unless expressly pointed out that there is a mutually exclusive relationship between them, otherwise, the relevant technical features involved in each embodiment can be cross-combined to flexibly construct new embodiments, as long as this combination does not deviate from the creative spirit of this application and can meet the needs in the prior art or solve certain deficiencies in the prior art. Those skilled in the art should be aware of this flexibility.

[0071] A method for recommending advertisement copy in the present application can be programmed as a computer program product and implemented by running on a server. For example, in the application scenario of the e-commerce platform in the present application, it is generally deployed and implemented on the server. Thus, by accessing the interface opened after the computer program product runs, human-computer interaction can be carried out with the process of the computer program product through a graphical user interface to execute this method.

[0072] Please refer to Figure 1 , in a typical embodiment of the advertisement copy recommendation method of the present application, it includes the following steps:

[0073] Step S1100: Obtain the product category and text information in the product information of the target product, where the text information includes the product description information of the target product;

[0074] In the application scenario of the e-commerce platform, each product can be regarded as a relatively independent single information unit, that is, each product has its corresponding product information, which is responsible for being published and maintained by the merchant users of the online stores on the e-commerce platform, and can be provided to consumer users for browsing, placing orders, etc. The above-mentioned online store can be an independent site, and the independent site independently maintains the product database of the products in its own online store.

[0075] The product category usually selects appropriate basic product characteristics as classification marks and successively summarizes them into several sub-aggregates (categories) with smaller ranges and more consistent characteristics, such as large, medium, and small categories, etc., so as to clearly distinguish and systematize the products within this range. By way of example, the product category can be hierarchically represented as the following category paths from the top layer to the bottom layer: clothes, tops, short sleeves; beverages, coffee, instant coffee.

[0076] The text information generally refers to all product description information associated with the storage of the product and suitable for being provided in text form, including but not limited to any one or more of the product title, product attribute data, product details text, product portrait label, etc. In terms of use, the product description information is generally used to describe any specific information such as the selling points, materials, usage methods, functions, models, etc. of the product.

[0077] Usually, when a merchant user of an online store needs to publish a certain target product, the corresponding product information of the target product can be entered in the corresponding product publishing page provided by the e-commerce platform, and then submitted to the background server of the e-commerce platform, and the server receives and stores it in the product database.

[0078] In one embodiment, the product database can be accessed through an interface encapsulated by the e-commerce platform for accessing the product database of an online store, and the product title and product details text in the product information of the target product can be obtained, and the two can be spliced together to form the text information. In another embodiment, the keywords in the product title and product details text of the target product can be further obtained and used as the text information, and the keywords can be text corresponding to the selling points of the target product.

[0079] Step S1200, recalling the selling point copywriting in the product selling point library in a coarse-grained multi-channel manner according to the product category and text information to form a candidate copywriting set;

[0080] The selling point copy is a textual description of the selling point of the product, which is used for advertising to attract users to place orders and purchase. It can be understood that the selling point usually refers to telling users what functions the product advertised to them has, and what psychological needs the user can meet after purchasing and using the product, and / or what practical effects it brings to them. In one embodiment, the corresponding selling point copy can be selected according to each product category of the e-commerce platform, and the selection can be achieved by manual or artificial intelligence. The selling point copy can be extracted from the advertisements historically placed by users on the e-commerce platform or from advertisements placed on a third-party platform. For example, the product category is clothing-tops-short sleeves, and its corresponding selling point copy is "This is a cute short sleeve, with pearl-decorated ice cream prints, youthful and age-reducing, slim and comfortable to wear, strong skin-friendliness, versatile, and easy to match and go out"; the product category is stationery-pen-fountain pen, and its corresponding selling point copy is "Simple style shows temperament, smooth writing, and easy to carry".

[0081] In one embodiment, the selling point copy in the commodity selling point library is recalled in a coarse-grained single-channel manner according to the commodity category of the target commodity. Specifically, the commodity category of the target commodity is matched with the commodity category corresponding to the selling point copy in the commodity selling point library, and the selling point copy corresponding to the commodity category of the target commodity is quickly recalled based on the accurate matching criterion. In addition, the selling point copy in the commodity selling point library is recalled in a coarse-grained single-channel manner according to the text information of the target commodity. Specifically, the text information is segmented using a word segmentation algorithm to obtain corresponding segmented texts, and these segmented texts are accurately and / or fuzzily matched with the selling point copy, and the selling point copy that contains all or most of these segmented texts is matched and recalled. The selling point copy recalled by the above two coarse-grained single-channels is collected to form a candidate copy set.

[0082] Step S1300, taking each selling point copy in the candidate copy set matching the text information as input, and using a neural network model pre-trained to convergence to calculate a semantic similarity score corresponding to the selling point copy and the text information;

[0083] In one embodiment, in a neural network model pre-trained to convergence, each selling point copywriting and text information are encoded to obtain corresponding high-dimensional copywriting encoding vectors and text encoding vectors. By way of example, the neural network model is Bert. In Bert, each selling point copywriting and text information are encoded through three embedding layers, namely token embedding, segment embedding, and position embedding, to obtain corresponding high-dimensional text encoding vectors and text encoding vectors. Further, in Bert, feature extraction is performed on the copywriting encoding vectors and text encoding vectors to extract corresponding copywriting semantic features and text semantic features. A preset similarity function is used to calculate the semantic similarity between the copywriting semantic features and the text semantic features as the semantic similarity score. It can be understood that the semantic similarity score represents the degree of semantic proximity between each selling point copywriting and the text information, not limited to the exact same words, but focusing on the consistency of semantic expression. The similarity function can be implemented by those skilled in the art as needed.

[0084] Step S1400: Recall the selling point copywriting in the candidate copywriting set in a fine-grained manner according to the semantic similarity score, and push the fine-grained recalled selling point copywriting to the first type of users.

[0085] It is not difficult to understand that the semantic similarity score is calculated by comparing the fine-grained semantic features corresponding to each selling point copywriting in the candidate copywriting set with the text information, so as to calculate the semantic similarity between the selling point copywriting and the text information. Accordingly, a fine-grained recall is performed based on the semantic similarity scores corresponding to the selling point copywriting in the candidate set, and the selling point copywriting with the highest semantic similarity score is recalled and pushed to the first type of users. In one embodiment, the first type of users are member users of an e-commerce platform. The server of the e-commerce platform can query the member database by obtaining the unique identification code of the user to check whether the user has recharged to become a recharged user. If there is a recharge record or corresponding recharge identifier for the user, it can be determined that the user is the first type of user.

[0086] According to the typical embodiments of the present application, it can be known that the technical solution of the present application has many advantages, including but not limited to the following aspects:

[0087] First, recall the selling point copy in the product selling point library through coarse-grained multi-channel, quickly narrow the scale of recall, roughly screen and lock some selling point copy to construct a candidate set. On this basis, then use the neural network model to finely analyze the selling point copy and the text information, determine the corresponding semantic similarity degree and quantify the corresponding semantic similarity score. Furthermore, the selling point copy in the candidate copy set can be recalled according to the semantic similarity score, and the final selling point copy can be determined through refined screening. It can be seen that refined screening based on rough screening greatly improves the execution efficiency and still ensures the accuracy of recall.

[0088] Secondly, the selling point copy recalled by fine-grained recall can effectively assist users in creating the copy used for the products in their advertised products, reduce the creation threshold, so that users can inspire their creation inspiration by referring to multiple selling point copy highly relevant to their products, and create their own copy.

[0089] In addition, this application uses a single neural network model, takes the selling point copy and the text information in the product information as inputs, and obtains the semantic similarity scores of both at one time, so as to conveniently select the selling point copy according to the semantic similarity scores, and can achieve efficient matching without a complex network architecture, which is beneficial to reducing the training cost of the neural network model and improving the matching efficiency.

[0090] Please refer to Figure 2 , in a further embodiment, before the step S1100 of obtaining the product category and text information in the product information of the target product, where the text information includes the product description information of the target product, the following steps are further included:

[0091] Step S1000: Select a part of the products corresponding to each product category in the e-commerce platform as category products according to a preset ratio;

[0092] The preset ratio can be set corresponding to the number of products corresponding to each product category in the e-commerce platform. For example, a smaller ratio is set for products with a larger number, and a larger ratio is set for products with a smaller number. Thus, the number of selected products is determined by multiplying the preset ratio by the number of products corresponding to each product category, and then the corresponding part of the products is selected as category products according to the number of selected products corresponding to each product category.

[0093] Step S1010: Match the selling point copy in the advertisement copy library of the e-commerce platform and / or the advertisement copy library of the third-party platform according to the keyword text corresponding to the category products, and the keyword text is pre-extracted from the product information of the corresponding category products;

[0094] Further, the product database can be accessed through the interface encapsulated for the product database by the e-commerce platform to obtain the text information in the product information corresponding to the category of products. The text information generally refers to all product description information associated with the storage of the product and suitable for being provided in text form, including but not limited to any one or more of the product title, product attribute data, product details text, product portrait tags, etc. of the product. Accordingly, the keyword text corresponding to the text information of the category of products can be extracted in advance by manual or artificial intelligence means. By way of demonstration, if the product title in the text information of the category of products is "Men's Cotton Blend Loose Fit Teenager Sporty Casual Ice Cream Print Short Sleeve T-shirt of Trendy Brand", the corresponding keyword text can be "loose / trendy / casual / versatile / ice cream print"; if the product title in the text information of the category of products is "Men's and Women's New Loose Couple Coats of Trendy Brand's Self-Made Spring High-Quality Crocodile Pattern PU Leather", the corresponding keyword text can be "crocodile pattern / PU leather / national trend / loose / high-quality texture / white". Of course, the text information is not limited to the product title, and more abundant text information can be combined to extract the corresponding keyword text that concisely expresses the selling points in text.

[0095] In one embodiment, the selling point copywriting in the advertising copywriting library of the e-commerce platform and / or the advertising copywriting library of the third-party platform can be accessed through the pre-encapsulated data interface, and the keyword text corresponding to the category of products is matched with the selling point copywriting, and then the selling point copywriting with a high semantic similarity to the keyword text is screened out by manual or artificial intelligence. By way of demonstration, the keyword text is "loose / trendy / casual / versatile / ice cream print", and the corresponding selling point copywriting is "This is a cute short sleeve with pearl-decorated ice cream prints, youthful and age-reducing, comfortable to wear and skin-friendly, versatile, and easy to match and go out"; the keyword text is "crocodile pattern / PU leather / national trend / loose / high-quality texture / white", and the corresponding selling point copywriting is "A must-have item for trendy outfits, a leather jacket with a super textured crocodile pattern, white neutralizes the impact of the leather jacket, sweet and cool and fashionable, more daily and versatile, and the PU leather fabric is dirt-resistant and easy to care for". The data interface can be pre-implemented by software engineers of the e-commerce platform or provided by a third-party platform.

[0096] Step S1020: Construct a product selling point library for storing the mapping relationship data among the selling point copywriting, keyword text, and product category.

[0097] Establish the mapping association relationship among the selling point copywriting, keyword text, and product category to obtain the mapping relationship data in the form of a triple, so as to facilitate obtaining the corresponding selling point copywriting according to the keyword text or product category and its corresponding mapping association relationship in the future. Further, construct a product selling point library to store the mapping relationship data.

[0098] In this embodiment, on the one hand, by selecting category products corresponding to each product category in the e-commerce platform in proportion, it not only ensures the diversity and extensiveness of the selling point copywriting in the product selling point library, but also guarantees that the selling point copywriting in the product selling point library is sufficient and not redundant. On the other hand, by refining the keyword text of the selling points corresponding to the category products, selling point copywriting that matches the semantics of the keyword text is obtained to achieve precise matching.

[0099] Please refer to Figure 3 , in the in-depth embodiment, in step S1200, the step of coarsely granular multi-channel recalling the selling point copywriting in the product selling point library according to the product category and text information to form a candidate copywriting set includes the following steps:

[0100] Step S1210: Recall the selling point copywriting corresponding to the product category of the target product in the product selling point library through the coarsely granular first channel;

[0101] The coarsely granular first channel is category recall, and its manifestation form can be a method function, an interface, etc. In one embodiment, the category recall method function is called to precisely match the product category of the target product with the product categories in the product selling point library, obtain the selling point copywriting associated with the mapped product category in the product selling point library that matches, and return this selling point copywriting as the recall result.

[0102] Step S1220: Recall the selling point copywriting corresponding to the keyword text that matches the text information of the target product in the product selling point library through the coarsely granular second channel;

[0103] The coarsely granular second channel is text recall, and its manifestation form can be a method function, an interface, etc. In one embodiment, the text recall method function is called to perform fuzzy matching on the text information of the target product and the keyword text in the product selling point library. If most of the keyword text can appear in the text information of the target product, it is regarded as a match, obtain the selling point copywriting associated with the mapped key text that matches, and return this selling point copywriting as the recall result.

[0104] Step S1230: Calculate the vector similarity between the vector corresponding to the text information of the target product and the vectors corresponding to each keyword text in the product selling point library through the coarsely granular third channel, and recall the selling point copywriting corresponding to the keyword text in the product selling point library whose vector similarity meets the preset conditions.

[0105] The coarsely granular third channel is generalization recall, and its manifestation form can be a method function, an interface, etc.

[0106] In one embodiment, the generalization recall method function is called, and the N-gram model is used to perform word segmentation preprocessing on the text information of the target commodity. Further, the word vector algorithm is used to encode the text information of the target commodity after preprocessing and each keyword text in the commodity selling point library to obtain the corresponding high-dimensional text encoding vector and each key encoding vector. The similarity function is used to calculate the vector distance corresponding between the text encoding vector and each key encoding vector as the vector similarity. The keyword text with the vector similarity meeting the preset threshold is selected, and the selling point copywriting mapped and associated with it is obtained, and the selling point copywriting is returned as the recall result. The similarity function and the word vector algorithm can be flexibly implemented by those skilled in the art according to prior knowledge or experimental data. The preset threshold can be set by those skilled in the art according to the actual business requirements.

[0107] In this embodiment, the selling point copywriting in the commodity selling point library is recalled through a coarse-grained multi-channel. Among them, the coarse-grained first channel recall adopts an exact matching method, and the coarse-grained second channel recall adopts a fuzzy matching method, which can achieve fast recall. In addition, the coarse-grained third channel recall encodes the text information and each keyword text, and then calculates the vector distance between the encoded text encoding vector and each key encoding vector as the vector similarity, and then selects the corresponding selling point copywriting accordingly. It can be seen that to a certain extent, the generalization is guaranteed, and the recall accuracy can be improved.

[0108] Please refer to Figure 4 , in a further embodiment, in step S1230, when calculating the vector similarity between the vector corresponding to the text information of the target commodity and each keyword text in the commodity selling point library and recalling the selling point copywriting corresponding to the keyword text whose vector similarity in the commodity selling point library meets the preset condition, the following steps are included:

[0109] Step S1231: Preprocess the text information of the target commodity through a preset word segmentation algorithm to obtain the corresponding segmented text;

[0110] The preset word segmentation algorithm is jieba (commonly known as "stuttering"). Jieba supports three word segmentation modes: exact mode, full mode, and search engine mode, which can be selected and implemented by those skilled in the art according to the actual business requirements.

[0111] In one embodiment, the jieba word segmentation algorithm is adopted in the search engine mode to segment the text information of the target commodity, and the corresponding segmented text is obtained. Those skilled in the art can understand that the accurate mode performs the most accurate segmentation on the text information, cuts out the corresponding segmented text, and can ensure that there is no redundant data in the segmented text. However, on the basis of the accurate mode, the search engine mode further segments the long words in the segmented text, making the granularity of the segmented text finer, which helps to improve the subsequent recall rate.

[0112] Step S1232: Perform vector encoding on the segmented text and the keyword text in the commodity selling point library through the word vector algorithm to obtain the corresponding target encoding vector and key encoding vector.

[0113] It can be understood that the keyword text is composed of corresponding individual keywords, so there is no need to segment the keyword text again.

[0114] The word vector algorithm is one-hot (one-hot encoding). The segmented text and the keyword text in the commodity selling point library are vector-encoded through one-hot. Specifically, a word register is used to perform corresponding encoding on the words in the segmented text and the keyword text. Each word has its own independent register bit, and at any time, only one bit is valid, that is, it is ensured that only 1 bit of the corresponding single word in the segmented text or the keyword text is in the state of 1, and the others are 0. Accordingly, the corresponding target encoding vector and key encoding vector are obtained.

[0115] Step S1233: Call a preset similarity function to calculate the vector similarity between the target encoding vector and the key encoding vector.

[0116] The similarity function can be the cosine similarity algorithm, Euclidean distance algorithm, Pearson correlation coefficient algorithm, Minkowski distance algorithm, Mahalanobis distance algorithm, Jaccard coefficient algorithm, etc. Those skilled in the art can choose any one to implement as long as the vector similarity can be calculated. In one embodiment, the vector distance between the target encoding vector and the key encoding vector is calculated as the vector similarity through a pre-similarity algorithm. The closer the vector similarity is to 1, the smaller the difference between the text information and the keyword text representation; the closer it is to 0, the greater the difference between the text information and the keyword text representation.

[0117] Step S1234: Determine the keyword text corresponding to the key encoding vector that meets the preset condition, and recall the selling point copy corresponding to the keyword text in the commodity selling point library.

[0118] Determine the keyword text corresponding to the key encoding vector that satisfies the preset threshold for vector similarity, and obtain the selling point copywriting associated with the mapping of this keyword text from the product selling point library as the recall result. The preset threshold can be set by those skilled in the art according to actual business requirements.

[0119] In this embodiment, by using the jieba algorithm to segment the text information of the target product, fine-grained segmented text is obtained, which helps to improve the subsequent recall rate.

[0120] Please refer to Figure 5 , in the extended embodiment, after the step S1230 of calculating the vector similarity between the vector corresponding to the text information of the target product and each keyword text in the product selling point library through the coarse-grained third channel, and recalling the selling point copywriting corresponding to the keyword text whose vector similarity satisfies the preset condition in the product selling point library, the following steps are further included:

[0121] Step S1240: Sort the recalled selling point copywriting according to the vector similarity to construct a rough sorting text set;

[0122] Sort these selling point copywriting in descending order according to the vector similarity corresponding to the recalled selling point copywriting to construct a rough sorting text set.

[0123] Step S1250: Select the selling point copywriting with a higher ranking in the rough sorting text set and push it to the second type of user.

[0124] Select the selling point copywriting with a higher ranking in the rough sorting text set, such as the top 3 selling point copywriting, and push these selling point texts to the second type of user. The second type of user is an ordinary user. The server of the e-commerce platform can query the user database by obtaining the unique identification code of the user to check whether the user is registered as a registered user. If there is an account record or corresponding registration identifier of the user, it can be determined that the registered user is the second type of user.

[0125] In this embodiment, by sorting the selling point copywriting and then selecting the selling point copywriting with a higher ranking, fast optimization can be achieved. Provide selling point copywriting with a high degree of similarity in terms of words to the text information of the target product to ordinary users, so as to provide different levels of service functions for different levels of users, which can assist users in creation and reduce the creation threshold.

[0126] Please refer to Figure 6 , in a further embodiment, the training process of the neural network model in step S1300 includes the following steps:

[0127] Step S1310: Call a single training sample from a preset dataset to train the neural network model. The training sample includes the text information of the preprocessed advertising product and its corresponding selling point copywriting.

[0128] The advertising product is a product that has been advertised. Its corresponding text information generally refers to all product description information associated with the product storage and suitable for being provided in text form, including but not limited to any one or more of the product title, product attribute data, product details text, product portrait label, etc. In terms of usage, the product description information is generally used to describe any specific information such as the selling points, materials, usage methods, functions, models, etc. of the product. In addition, the selling point copywriting is the selling point copywriting used when the advertising product is advertised. The text information of the advertising product and its corresponding selling point copywriting can be obtained through manual collection or web data extraction.

[0129] Use the corresponding CLS (first identifier) and SEP (sentence separator) of Bert to label the text information of the advertising product and its corresponding selling point copywriting to complete the preprocessing of the two. Take the preprocessed single data as the single training sample. As revealed here, each training sample in the dataset is constructed accordingly.

[0130] In addition, a corresponding supervision label can be set for the training sample, so that the neural network model performs supervised training according to this supervision label and conducts binary classification prediction. The supervision label is a supervision label corresponding to whether the text information of the advertising product included in the training sample and its corresponding selling point copywriting are semantically similar. It can be judged manually whether the semantics are similar. If they are similar, the supervision label is 1; if they are not similar, the supervision label is 0.

[0131] Step S1320: Extract the deep semantic features of the text information of the training sample and its corresponding selling point copywriting.

[0132] The neural network model is Bert. In one embodiment, the NSP module (next sentence task) of bert is used to extract the deep semantic features of the text information of the training sample and its corresponding selling point copywriting. Specifically, according to the sentence separator identifier, identify the text information and the selling point copywriting in the training sample, and accordingly extract the corresponding context semantic information of the two, that is, the information semantically related between the two, as the deep semantic features.

[0133] Step S1330: Use a binary classification function to calculate the classification probability corresponding to whether the deep semantic features belong to the same sentence, as the semantic similarity score corresponding to whether the selling point copywriting is the next sentence of the text information.

[0134] Further, input the deep semantic features into a fully connected layer for linear transformation, and map them to a binary classification space, which includes a positive class space representing that the text information of the training sample is semantically similar to the corresponding selling point copywriting, and a negative class space representing that the text information of the training sample is semantically dissimilar to the corresponding selling point copywriting. In one embodiment, the binary classification function is the sigmod function, and the sigmod function is used to calculate the probability corresponding to the prediction of the positive class space when the deep semantic features are mapped to the binary space as the classification probability corresponding to whether it belongs to the same sentence. It can be understood that the classification probability is the output result of the NSP task, which can represent whether the selling point copywriting is the next sentence of the text information, that is, whether the selling point copywriting is semantically similar to the text information. Therefore, this classification probability can be regarded as a semantic similarity score.

[0135] Step S1340: Calculate the cross-entropy loss value corresponding to the semantic similarity score according to the supervision label corresponding to the training sample, and perform gradient update on the model based on this cross-entropy loss value until the model converges.

[0136] Call a preset loss function, which can be flexibly set by those skilled in the art according to prior knowledge or experimental experience. Calculate the loss value corresponding to the cross-entropy loss of the predicted probability based on the supervision label. When this loss value reaches a preset threshold, it indicates that the model has been trained to a convergent state, and thus 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, gradient update is performed on the model according to this loss value, usually backpropagating to correct the weight parameters of each link of the model to make the model closer to convergence. Then, continue to call the next sample data in the dataset to perform iterative training on the model until the model is trained to a convergent state.

[0137] In this embodiment, the training process of calculating the semantic similarity score implemented by the NSP module of the bert model is disclosed. It can be seen that under the training of the sample data and supervision labels in the dataset, the model has the ability to quickly calculate the corresponding semantic similarity score according to the text information of the product and the selling point copywriting, thus ensuring that it can serve the calculation of the semantic similarity score corresponding to the text information of the product and the selling point copywriting in the future, greatly improving the accuracy and efficiency of the calculation.

[0138] Please refer to Figure 7 , in an extended embodiment, after step S1400: Fine-grained recall the selling point copywriting in the candidate copywriting set according to the semantic similarity score, and obtain the steps of pushing the fine-grained recalled selling point copywriting to the user, the following steps are further included:

[0139] Step S1500: Sort the selling point texts in the candidate text set according to the semantic similarity scores, and construct a first sorted text set.

[0140] Sort the selling point texts in the candidate text set according to the semantic similarity scores corresponding to them, from high to low, and construct a first sorted text set.

[0141] Step S1600: Select the selling point texts with higher rankings in the first sorted text set as the selling point texts for fine-grained recall, and push these selling point texts to the first type of users.

[0142] Select the selling point texts with higher rankings in the coarsely sorted text set, such as the top 3 selling point texts, and push these selling point texts to the first type of users. The first type of users are member users. The server of the e-commerce platform can query the member database by obtaining the unique identification code of the user to check whether the user has recharged to become a recharged user. If there is a recharge record or corresponding recharge identifier of the user, it can be determined that the recharged user is the first type of user.

[0143] In this embodiment, by sorting the selling point texts and then selecting the selling point texts with higher rankings, fast optimization can be achieved. Providing selling point texts with a high degree of semantic similarity to the text information of the target commodity to member users can assist users in creation and reduce the creation threshold.

[0144] Please refer to Figure 8 An advertising copy recommendation device provided to meet one of the purposes of this application is a functional embodiment of the advertising copy recommendation method of this application. The device includes: an information acquisition module 1100, a coarse-grained recall module 1200, a scoring calculation module 1300, and a fine-grained recall module 1400. Among them, the information acquisition module 1100 is used to acquire the commodity category and text information in the commodity information of the target commodity, and the text information includes the commodity description information of the target commodity; the coarse-grained recall module 1200 is used to coarsely recall the selling point texts in the commodity selling point library through multiple channels according to the commodity category and text information to form a candidate copy set; the scoring calculation module 1300 is used to take the matching of each selling point text in the candidate copy set with the text information as the input, and calculate the semantic similarity score corresponding to the selling point text and the text information by using a neural network model that has been pre-trained to convergence; the fine-grained recall module 1400 is used to finely recall the selling point texts in the candidate copy set according to the semantic similarity score, and obtain the finely recalled selling point texts and push them to the first type of users.

[0145] In a further embodiment, before the information acquisition module 1100, there is further included: a proportion selection module, configured to select a part of the commodities corresponding to each commodity category in the e-commerce platform as category commodities according to a preset proportion; a matching selling point copywriting module, configured to match the selling point copywriting in the advertisement copywriting library of the e-commerce platform and / or the advertisement copywriting library of a third-party platform according to the keyword text corresponding to the category commodities, where the keyword text is pre-extracted from the commodity information of the corresponding category commodities; a building commodity selling point library module, configured to build a commodity selling point library for storing the mapping relationship data among the selling point copywriting, the keyword text, and the commodity category.

[0146] In a deepened embodiment, the coarse-grained recall module 1200 includes: a first-channel recall sub-module, configured to recall the selling point copywriting corresponding to the commodity category of the target commodity in the commodity selling point library through a coarse-grained first channel; a second-channel recall sub-module, configured to recall the selling point copywriting corresponding to the keyword text that matches the text information of the target commodity in the commodity selling point library through a coarse-grained second channel; a third-channel recall sub-module, configured to calculate the vector similarity between the vector corresponding to the text information of the target commodity and each keyword text in the commodity selling point library through a coarse-grained third channel, and recall the selling point copywriting corresponding to the keyword text whose vector similarity meets a preset condition in the commodity selling point library.

[0147] In a further embodiment, the third-channel recall sub-module includes: a word segmentation preprocessing unit, configured to preprocess the text information of the target commodity through a preset word segmentation algorithm to obtain corresponding segmented text; a vector encoding unit, configured to perform vector encoding on the segmented text and the keyword text in the commodity selling point library through a word vector algorithm to obtain corresponding target encoding vectors and key encoding vectors; a similarity calculation unit, configured to call a preset similarity function to calculate the vector similarity between the target encoding vector and the key encoding vector; a recalled selling point copywriting unit, configured to determine the keyword text corresponding to the key encoding vector with a vector similarity that meets a preset condition, and recall the selling point copywriting corresponding to the keyword text in the commodity selling point library.

[0148] In an extended embodiment, after the third-channel recall sub-module, there is further included: a coarse sorting sub-module, configured to sort the recalled selling point copywriting according to the vector similarity and build a coarse sorting text set; a second-type user push sub-module, configured to select the selling point copywriting with a higher ranking in the coarse sorting text set and push it to the second type of users.

[0149] In a further embodiment, the training process of the neural network model in the scoring calculation module 1300 includes: a training implementation submodule, which is used to call a single training sample from a preset data set to implement training on the neural network model, and the training sample includes the pre-processed text information of the advertised product and its corresponding selling point copy; a feature extraction submodule, which is used to extract the deep semantic features of the text information of the training sample and its corresponding selling point copy; a classification prediction submodule, which is used to use a binary classification function to calculate the classification probability corresponding to whether the deep semantic feature belongs to the same sentence, as a semantic similarity score corresponding to the next sentence of the text information; a loss calculation submodule, which is used to calculate the cross entropy loss value corresponding to the semantic similarity score according to the supervision label corresponding to the training sample, and implement gradient update on the model according to the cross entropy loss value until the model converges.

[0150] In an extended embodiment, the fine-grained recall module 1400 also includes: a first sorting module, used to sort the selling point texts in the candidate text set according to the semantic similarity score, and construct a first sorted text set; a type of user push module, used to select the selling point texts with higher rankings in the first sorted text set as the selling point texts for fine-grained recall, and push these selling point texts to the first type of users.

[0151] In order to solve the above technical problems, the present application also provides a computer device. Figure 9 As shown, a schematic diagram of the internal structure of a computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. Among them, the computer-readable storage medium of the computer device stores an operating system, a database, and computer-readable instructions. The database may store a control information sequence. When the computer-readable instructions are executed by the processor, the processor may implement an advertising copy recommendation method. The processor of the computer device is used to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device may store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor may execute the advertising copy recommendation method of the present application. The network interface of the computer device is used to connect and communicate with a terminal. Those skilled in the art will understand that Figure 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0152] In this embodiment, the processor is used to execute Figure 8The specific functions of each module and its sub-modules therein, and the memory stores program codes and various types of data required to execute the above 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 codes and data required to execute all modules / sub-modules in the advertisement copy recommendation device of the present application, and the server can call the program codes and data of the server to execute the functions of all sub-modules.

[0153] The present 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 execute the steps of the advertisement copy recommendation method according to any embodiment of the present application.

[0154] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments of the present application can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the aforementioned storage medium can be a computer-readable storage medium such as a magnetic disk, an optical disc, a read-only memory (ROM), or a random access memory (RAM), etc.

[0155] In summary, the present application realizes fast and accurate recall by performing fine-grained recall of selling point copy on the basis of coarse-grained recall, and the selling point copy can facilitate users to refer to and create their own selling point copy.

[0156] Those skilled in the art of this technology can understand that the steps, measures, and solutions in the various operations, methods, and processes discussed in the present application can be alternated, changed, combined, or deleted. Further, the other steps, measures, and solutions in the various operations, methods, and processes discussed in the present application can also be alternated, changed, rearranged, decomposed, combined, or deleted. Further, the steps, measures, and solutions in the prior art that are the same as those disclosed in the various operations, methods, and processes in the present application can also be alternated, changed, rearranged, decomposed, combined, or deleted.

[0157] The above are only some embodiments of the present application. It should be noted that for those of ordinary skill in the art of this technology, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. An advertising copy recommendation method, characterized in that, Including the following steps: Obtain the product category and text information in the product information of the target product, where the text information includes the product description information of the target product; Coarse-grained multi-channel recall the selling point copywriting in the product selling point library according to the product category and text information to form a candidate copywriting set, including: recalling the selling point copywriting corresponding to the product category of the target product in the product selling point library through the coarse-grained first channel; recalling the selling point copywriting corresponding to the keyword text matching the text information of the target product in the product selling point library through the coarse-grained second channel; calculating the vector similarity between the vector corresponding to the text information of the target product and each keyword text in the product selling point library through the coarse-grained third channel, and recalling the selling point copywriting corresponding to the keyword text whose vector similarity meets the preset conditions in the product selling point library; Taking the matching of each selling point copywriting in the candidate copywriting set with the text information as the input, and using a neural network model pre-trained to convergence to calculate the semantic similarity score corresponding to the selling point copywriting and the text information; Fine-grained recall the selling point copywriting in the candidate copywriting set according to the semantic similarity score, and push the fine-grained recalled selling point copywriting to the first type of users.

2. The advertising copy recommendation method according to claim 1, characterized in that Before the step of obtaining the product category and text information in the product information of the target product, where the text information includes the product description information of the target product, the following steps are further included: Select a part of the products corresponding to each product category on the e-commerce platform as category products according to a preset ratio; Match the selling point copywriting in the advertisement copywriting library of the e-commerce platform and / or the advertisement copywriting library of the third-party platform according to the keyword text corresponding to the category products, where the keyword text is pre-extracted from the product information of the corresponding category products; Construct a product selling point library for storing the mapping relationship data between the selling point copywriting, keyword text, and product category.

3. The advertising copy recommendation method according to claim 1, characterized in that In the step of calculating the vector similarity between the vector corresponding to the text information of the target product and each keyword text in the product selling point library, and recalling the selling point copywriting corresponding to the keyword text whose vector similarity meets the preset conditions in the product selling point library, the following steps are included: Preprocess the text information of the target product through a preset word segmentation algorithm to obtain the corresponding segmented text; Perform vector encoding on the segmented text and the keyword text in the product selling point library through a word vector algorithm to obtain the corresponding target encoding vector and key encoding vector; Call a preset similarity function to calculate the vector similarity between the target encoding vector and the key encoding vector; Determine the keyword text corresponding to the key encoding vector with a vector similarity meeting the preset conditions, and recall the selling point copywriting corresponding to the keyword text in the product selling point library.

4. The advertising copy recommendation method according to claim 1, wherein After the step of calculating the vector similarity between the vector corresponding to the text information of the target product and each keyword text in the product selling point library, and recalling the selling point copywriting corresponding to the keyword text whose vector similarity meets the preset conditions in the product selling point library, the following steps are further included: Sort the recalled selling point copywriting according to the vector similarity accordingly to construct a coarse-sorted text set; Select the selling point copywriting with a higher ranking in the rough sorting text set and push it to the second type of users.

5. The advertising copy recommendation method according to claim 1, wherein The training process of the neural network model includes the following steps: Call a single training sample from a preset data set to train the neural network model. The training sample includes the text information of the preprocessed advertising product and its corresponding selling point copywriting. Extract the deep semantic features of the text information of the training sample and its corresponding selling point copywriting. Use a binary classification function to calculate the classification probability corresponding to whether the deep semantic features belong to the same sentence, as the semantic similarity score corresponding to whether the selling point copywriting is the next sentence of the text information. Calculate the cross-entropy loss value corresponding to the semantic similarity score according to the supervision label corresponding to the training sample, and perform gradient update on the model based on the cross-entropy loss value until the model converges.

6. The advertising copy recommendation method according to any one of claims 1 to 5, characterized in that After the step of finely granularity recalling the selling point copywriting in the candidate copywriting set according to the semantic similarity score and obtaining the finely granularity recalled selling point copywriting to be pushed to the user, the following steps are further included: Sort the selling point copywriting in the candidate copywriting set according to the semantic similarity score, and construct a first sorted text set. Select the selling point copywriting with a higher ranking in the first sorted text set as the finely granularity recalled selling point copywriting, and push these selling point copywriting to the first type of users.

7. An advertising copy recommendation device, characterized in that, Include: An information acquisition module for acquiring the product category and text information in the product information of the target product. The text information includes the product description information of the target product. A rough granularity recall module for roughly granularity multi-channel recalling the selling point copywriting in the product selling point library according to the product category and text information, including: recalling the selling point copywriting corresponding to the product category of the target product in the product selling point library through the rough granularity first channel; recalling the selling point copywriting corresponding to the keyword text matching the text information of the target product in the product selling point library through the rough granularity second channel; calculating the vector similarity between the text information of the target product and the vectors corresponding to each keyword text in the product selling point library through the rough granularity third channel, and recalling the selling point copywriting corresponding to the keyword text whose vector similarity meets the preset conditions in the product selling point library. A scoring calculation module for taking the matching of each selling point copywriting in the candidate copywriting set with the text information as the input, and using a pre-trained and converged neural network model to calculate the semantic similarity score corresponding to the selling point copywriting and the text information. A fine granularity recall module for finely granularity recalling the selling point copywriting in the candidate copywriting set according to the semantic similarity score, and obtaining the finely granularity recalled selling point copywriting to be pushed to the first type of users.

8. The advertisement copy recommendation device according to claim 7, characterized in that Before the information acquisition module, the following is further included: A proportion selection module for selecting a part of the products corresponding to each product category on the e-commerce platform as category products according to a preset proportion. A matching selling point copywriting module for matching the selling point copywriting in the advertising copywriting library of the e-commerce platform and / or the advertising copywriting library of the third-party platform according to the keyword text corresponding to the category product. The keyword text is pre-extracted from the product information of the corresponding category product. Build a product selling point library module for building a product selling point library to store the mapping relationship data among the selling point copywriting, keyword text, and product categories.

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

10. A computer-readable storage medium, characterized in that, It stores in the form of computer-readable instructions a computer program implemented according to the method described in any one of claims 1 to 6. When the computer program is called and run by a computer, it executes the steps included in the corresponding method.

Images