Advertising placement product selection methods, devices, equipment, and media

Abstract

This application relates to a method, apparatus, and medium for product selection in advertising placement within the field of computer technology. The method includes: acquiring the title text of a product; responding to an online store's advertising product selection request and acquiring the title text of each candidate product from the online store's product database, wherein the online store operates on an independent site of an e-commerce platform; determining multiple third-party target advertising products corresponding to each candidate product from pre-collected third-party advertising products based on the descriptive similarity between the title texts; calculating the placement score of the third-party target advertising products based on pre-collected advertising placement data, and determining a reference score for each candidate product based on the placement scores of the multiple third-party target advertising products corresponding to each candidate product; and selecting multiple candidate products based on the reference scores of each candidate product and pushing them to the online store. This application can provide advertising product selection references by utilizing third-party advertising placement data and operates efficiently.

Description

Technical Field

[0001] This application relates to the field of e-commerce technology, and in particular to an advertising placement product selection method and its corresponding apparatus, computer equipment, and computer-readable storage medium. Background Technology

[0002] Product advertising aims to boost sales by introducing product information to the target audience, highlighting its features, and attracting the attention of the target audience and potential consumers. It strives for direct and immediate advertising results, leaving a positive product image in the minds of the target audience, thereby laying the foundation for increasing market share and ultimately achieving the company's goals.

[0003] Traditional techniques typically rely on historical advertising performance data of products after ad placement, such as the return on ad spend (ROI). A threshold is set, and training samples are labeled based on whether the data exceeds the threshold. This training set is then used to train a classification model that can predict the probability of a product being selected for ad placement. As can be seen, this approach heavily depends on historical advertising performance data. However, generally speaking, only a minority of merchants on e-commerce platforms possess abundant historical advertising performance data. Most merchants have limited or even no such data, and many new merchants or those who have never advertised before lack any historical performance data at all. Therefore, traditional techniques lack universal applicability and cannot meet the advertising needs of most merchants.

[0004] In order to achieve a technical solution applicable to all advertising scenarios, 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 above-mentioned problems by providing an advertising placement product selection method 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 placement product selection method provided for one of the purposes of this application includes the following steps:

[0008] In response to an online store's advertising product selection request, the title text of each candidate product in the online store's product database is obtained, wherein the online store operates on an independent site of an e-commerce platform;

[0009] From the pre-collected third-party advertising products, multiple third-party target advertising products corresponding to each candidate product are identified based on the descriptive similarity between the title texts;

[0010] The ad placement score is calculated based on the pre-collected ad placement data of third-party target ad products. The reference score for each candidate product is determined by the ad placement scores of multiple third-party target ad products corresponding to each candidate product.

[0011] Based on the reference scores of each candidate product, several candidate products are selected and pushed to the online store.

[0012] In a further embodiment, the step of determining multiple third-party target advertising products corresponding to each candidate product based on the descriptive similarity between title texts from the pre-collected third-party advertising products includes the following steps:

[0013] From the pre-collected third-party advertising products, retrieve the title texts of multiple third-party advertising products that are similar to their title text descriptions for each candidate product;

[0014] For each candidate product and its corresponding multiple third-party advertising products, a pre-trained text encoding model that has converged is used to extract the deep semantic features of each title text, thereby obtaining the candidate semantic vector corresponding to each candidate product and the third-party semantic vector corresponding to each third-party advertising product.

[0015] Calculate the similarity score between each candidate semantic vector and its corresponding third-party semantic vectors;

[0016] Based on the similarity score, for each candidate product, a portion of the third-party advertising products that meet the preset conditions are determined from among the multiple third-party advertising products corresponding to it as the third-party target advertising products.

[0017] In a further embodiment, the training process of the text encoding model includes the following steps:

[0018] Obtain a single training sample from the prepared training set. Each training sample in the training set contains the title text of a product, the title text of a third-party advertised product, and a supervision label. The supervision label of the training sample indicates whether the title text of the third-party advertised product in the training sample is similar to the title text description of the product.

[0019] After extracting text semantic features from each title text in the training samples by the text encoding model, the similarity score corresponding to the training samples is predicted by the prediction module.

[0020] The loss value of the similarity score is calculated based on the supervision label corresponding to the training sample. When the loss value of the model does not reach the preset threshold, the model weights are updated, and other training samples are called to carry out iterative training until the model converges.

[0021] In a further embodiment, the step of calculating the placement score based on the pre-collected advertising placement data of the third-party target advertising product includes the following steps:

[0022] Obtain advertising placement data pre-collected from third-party target advertising products, wherein the advertising placement data includes comment information, number of likes, number of favorites, and advertising placement duration;

[0023] A behavioral score and a feedback score are calculated based on the advertising data. The behavioral score is determined based on the number of likes, the number of favorites, and the duration of the advertising campaign. The feedback score is determined based on the comment information.

[0024] The behavioral score is normalized, and the corresponding normalization result is added to the feedback score and normalized again to obtain the placement score of the third-party target advertising product.

[0025] In a further embodiment, the steps of calculating behavioral scores and feedback scores based on advertising delivery data include the following steps:

[0026] A pre-trained, convergent sentiment analysis model was used to identify positive reviews in the advertising data.

[0027] Calculate the percentage of positive reviews out of all reviews, and use a smoothing algorithm to process this percentage to obtain a feedback score.

[0028] The sum of likes and favorites in the ad placement information is divided by the ad placement duration to obtain the corresponding behavioral score.

[0029] In a further embodiment, a smoothing algorithm is used to process the quantity proportion to obtain a feedback score, including the following steps:

[0030] Obtain the comment information of the third-party target advertising product at each preset interval within a recent preset time period, and calculate the proportion of positive comments in the comment information at each preset interval.

[0031] Based on the proportion of quantities corresponding to each preset interval duration, the preset first smoothing parameter and second smoothing parameter are obtained using the maximum likelihood algorithm;

[0032] The current quantity percentage is adjusted based on the first smoothing parameter and the second smoothing parameter, and the adjusted quantity percentage is used as a feedback score.

[0033] In a further embodiment, the reference score for each candidate product is determined by the placement scores of multiple third-party target advertising products corresponding to each candidate product. This includes: for each candidate product, the similarity score of each third-party target advertising product is multiplied by the placement score, and the average of these scores is calculated to determine the reference score for each candidate product.

[0034] On the other hand, an advertising product selection device provided to meet one of the purposes of this application includes a request response module, a similarity determination module, a score determination module, and a score optimization module, wherein: the request response module is used to respond to the advertising product selection request of an online store and obtain the title text of each candidate product in the product database of the online store, wherein the online store runs on an independent site of an e-commerce platform; the similarity determination module is used to determine multiple third-party target advertising products corresponding to each candidate product from pre-collected third-party advertising products based on the descriptive similarity between the title texts; the score determination module is used to calculate the placement score of the third-party target advertising products based on the pre-collected advertising placement data, and determine the reference score of each candidate product based on the placement scores of the multiple third-party target advertising products corresponding to each candidate product; the score optimization module is used to optimize multiple candidate products based on the reference scores of each candidate product and push them to the online store.

[0035] In a further embodiment, the similarity determination module includes: a similarity retrieval submodule, used to retrieve, from the pre-collected third-party advertising products, the title texts of multiple third-party advertising products that are similar to the title text description of each candidate product; a vectorization representation submodule, used to extract deep semantic features of each title text for each candidate product and its corresponding multiple third-party advertising products using a pre-trained and converged text encoding model, to obtain candidate semantic vectors corresponding to each candidate product and third-party semantic vectors corresponding to each third-party advertising product; a similarity retrieval submodule, used to calculate the similarity score between each candidate semantic vector and its corresponding third-party semantic vectors; and a target determination submodule, used to determine, based on the similarity score, a portion of the third-party advertising products that meet preset conditions from the multiple third-party advertising products corresponding to each candidate product as third-party target advertising products.

[0036] In a further embodiment, the training process of the text encoding model in the vectorization representation submodule includes the following steps: a training sample acquisition module, used to acquire a single training sample in a prepared training set, wherein each training sample in the training set contains the title text of a product, the title text of a third-party advertised product, and a supervision label, wherein the supervision label of the training sample indicates whether the title text of the third-party advertised product in the training sample is similar to the title text description of the product; a rating prediction module, used to extract text semantic features from each title text in the training sample by the text encoding model, and then predict the similarity score corresponding to the training sample by the prediction module; and an iterative training module, used to calculate the loss value of the similarity score according to the supervision label corresponding to the training sample, and when the model loss value does not reach a preset threshold, to update the weights of the model and continue to call other training samples to perform iterative training until the model converges.

[0037] In a further embodiment, the scoring determination module includes: an advertising data acquisition submodule, used to acquire pre-collected advertising data of a third-party target advertising product, the advertising data including comment information, number of likes, number of favorites, and advertising duration; a feedback score and behavior score determination submodule, used to calculate a behavior score and a feedback score based on the advertising data, the behavior score being determined based on the number of likes, number of favorites, and advertising duration, and the feedback score being determined based on the comment information; and an advertising score acquisition submodule, used to normalize the behavior score, add the corresponding normalization result to the feedback score, and perform normalization to obtain the advertising score of the third-party target advertising product.

[0038] In a further embodiment, the feedback scoring and behavior scoring determination submodule includes: a sentiment analysis unit, used to determine the positive review information in the advertising data using a pre-trained and converged sentiment analysis model; a feedback scoring calculation and processing unit, used to calculate the proportion of positive review information to all review information, and use a smoothing algorithm to process the proportion to obtain a feedback score; and a behavior scoring calculation unit, used to calculate the sum of the number of likes and favorites in the advertising information divided by the advertising duration to obtain the corresponding behavior score.

[0039] In a further embodiment, the feedback score calculation and processing unit includes: a historical quantity proportion calculation subunit, used to obtain the comment information of the third-party target advertising product at each preset interval within a recent preset time period, and calculate the quantity proportion of positive comments in the comment information at each preset interval; a smoothing parameter acquisition subunit, used to obtain preset first smoothing parameters and second smoothing parameters based on the quantity proportion corresponding to each preset interval using a maximum likelihood algorithm; and a quantity proportion adjustment subunit, used to adjust the current quantity proportion based on the first smoothing parameters and the second smoothing parameters, and obtain the adjusted quantity proportion as the feedback score.

[0040] In a further embodiment, the scoring determination module includes: a reference scoring determination submodule, used to determine the reference score for each candidate product by multiplying the similarity score of each third-party target advertising product by the placement score, adding them together, and averaging the results.

[0041] On another front, 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 placement and product selection method described in this application.

[0042] 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 placement and product selection method, which, when invoked by a computer, performs the steps included in the method.

[0043] The technical solution of this application has many advantages, including but not limited to the following aspects:

[0044] This application overcomes the drawback of sparse historical advertising data on e-commerce platforms. It can determine a reference score that can be used as a decision-making reference for alternative products by utilizing the advertising data of third-party products similar to the alternative products in the online store, without relying on the online store's own historical advertising data. Thus, alternative products in the online store can be selected based on the reference score and recommended to the online store to assist in product selection for advertising.

[0045] On the other hand, the technical solution of this application is convenient, efficient and low-cost to implement, solves the existing technical pain points of product selection in e-commerce advertising and is applicable to general advertising scenarios. Attached Figure Description

[0046] 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:

[0047] Figure 1 This is a flowchart illustrating a typical embodiment of the product selection method for advertising placement according to this application;

[0048] Figure 2 This is a schematic diagram illustrating the process of determining third-party target advertising products based on the descriptive similarity between title texts in an embodiment of this application;

[0049] Figure 3 This is a schematic diagram illustrating the training process of the text encoding model in an embodiment of this application;

[0050] Figure 4 This is a schematic diagram illustrating the process of obtaining a placement score for a third-party target advertising product in an embodiment of this application;

[0051] Figure 5 This is a schematic diagram illustrating the process of obtaining feedback scores and behavioral scores in an embodiment of this application;

[0052] Figure 6 This is a schematic diagram of the process for obtaining feedback scores in one embodiment of this application;

[0053] Figure 7 This is a schematic diagram of the advertising placement and product selection device of this application;

[0054] Figure 8 This is a schematic diagram of the structure of a computer device used in this application. Detailed Implementation

[0055] 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.

[0056] 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.

[0057] Those skilled in the art will understand 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.

[0058] 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; PDAs (Personal Digital Assistants) that may include radio frequency receivers, pagers, internet / intranet access, web browsers, notebooks, calendars, and / or GPS (Global Positioning System) receivers; 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.

[0059] 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.

[0060] 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.

[0061] 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.

[0062] 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.

[0063] 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.

[0064] 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.

[0065] 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.

[0066] The advertising product selection method of this application can be programmed into a computer program product and deployed on a client or server. For example, in an exemplary application scenario of this application, it can be deployed on the server of an e-commerce platform. In this way, the method can be executed by 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 run.

[0067] Please see Figure 1 The product selection method for advertising placement in this application, in its typical embodiment, includes the following steps:

[0068] Step S1100: Respond to the online store's advertising product selection request and obtain the title text of each candidate product in the online store's product database. The online store operates on an independent site of the e-commerce platform.

[0069] 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.

[0070] The e-commerce platform's server cluster centrally maintains the backend operational information of all online stores under the platform, providing these online stores with various standardized services, including but not limited to the advertising placement and product selection service described in this application. When providing the advertising placement and product selection service, each online store is allowed to enable the service by launching an interface encapsulated for the advertising placement and product selection service within the e-commerce platform. Thus, when a merchant user of an online store initiates an advertising product selection request, the service can obtain a list of recommended advertising products for that online store and push and display this list to the online store, providing a reference for advertising product selection.

[0071] Online stores can maintain their own product database to store product information, including but not limited to product images, title text, and details text, so that e-commerce platforms can display corresponding products on the page based on the product information.

[0072] Online store merchants can operate on the pages related to advertising and product selection services provided by the e-commerce platform displayed on their client. For example, they can touch the relevant controls on the page to trigger the generation of an advertising product selection request on the client. Then, the client submits the advertising product selection request to the e-commerce service platform's server to obtain the corresponding advertising product recommendation list.

[0073] The server receives the advertising product selection request, responds to it, obtains the title text of each product in the product database of the online store, and regards each product as a candidate product for advertising product selection.

[0074] Step S1200: From the pre-collected third-party advertising products, based on the descriptive similarity between the title texts, determine multiple third-party target advertising products corresponding to each candidate product;

[0075] It can collect sufficient advertising information of third-party advertised products from various advertising platforms, including independent e-commerce sites, social media, and search engines. The advertising information includes advertising data and title text.

[0076] Regarding advertising data, consumers can access the display page of third-party advertised products through advertisements published by those products. On the page, consumers can like, favorite, and other actions related to the third-party advertised products. Consumers who purchase third-party advertised products can also leave comments. These likes, favorites, and positive reviews reflect a certain degree of consumer liking for the third-party advertised products, indicating the effectiveness of the advertising campaign. Based on this, advertising data for the third-party advertised products can be collected, including comments, number of likes, number of favorites, and advertising duration. Subsequently, a corresponding advertising score can be calculated based on this data to characterize the advertising effectiveness of the third-party advertised products.

[0077] Each third-party advertising product is associated with its advertising data and title text and stored for later retrieval.

[0078] Generally, title text typically includes descriptive information such as product brand, product name, basic attributes (material / function / features) or specifications (model / color / size / specification / use / item number). Functionally, the corresponding product can be identified based on the description in the title text. Therefore, based on the descriptive similarity between the title text of a third-party advertised product and the title text of candidate products (i.e., the degree of descriptive similarity), it can be determined whether the third-party advertised product and the candidate products are similar. Third-party advertised products similar to the candidate products are then designated as target third-party advertised products. Thus, from the pre-collected multiple third-party advertised products, multiple corresponding target third-party advertised products are determined for each candidate product.

[0079] A deep learning model can be used to vectorize the deep semantic features of the title text of the third-party advertised product and the title text of the alternative products. This vectorization represents the semantic features of the title text description using quantified numerical values. Therefore, the similarity between the vectorization representations of the two products can characterize the descriptive similarity between the third-party advertised product and the alternative products. The deep learning model can be a deep semantic learning-based network model suitable for extracting text semantic features in the field of Natural Language Processing (NLP). Specifically, the open-source framework Sentence Transformers is used, which provides a large number of pre-trained and converged Transformer models, such as BERT, RoBERTa, XLM-RoBERTa, and MPNet.

[0080] Step S1300: Calculate the placement score based on the pre-collected advertising placement data of the third-party target advertising products, and determine the reference score of each candidate product based on the placement scores of multiple third-party target advertising products corresponding to each candidate product.

[0081] We obtain the advertising data for multiple third-party target advertising products corresponding to each candidate product, and calculate the behavioral score and feedback score for each third-party target advertising product based on the advertising data. For ease of discussion, the following focuses on a single third-party target advertising product:

[0082] The behavioral score is determined based on the number of likes, favorites, and ad duration in the ad delivery data. In one embodiment, the number of likes and favorites per unit time can be quantified, i.e., the sum of the number of likes and favorites is divided by the ad delivery duration, and the quantified result is used as the behavioral score. In a preferred embodiment, the number of likes and favorites per unit time can be quantified, i.e., the number of likes is divided by the ad duration, and the number of favorites is divided by the ad duration, respectively. Then, the quantified results are matched with corresponding preset weights, weighted, and then summed. For example, the weight matched with the number of likes per unit time is 0.4, and the weight matched with the number of favorites per unit time is 0.6. The sum is used as the behavioral score.

[0083] The feedback score is determined based on comment information in the advertising data. In one embodiment, the percentage of positive reviews out of the total number of comments is used as the feedback score. In a preferred embodiment, based on the determined percentage of positive reviews, a Bayesian smoothing algorithm is used to determine an adjustment parameter based on the comment information of the corresponding third-party target advertising product during its historical advertising campaigns. This adjustment parameter is then used to smooth the percentage, resulting in a smoothed percentage as the feedback score. The determination of the adjustment parameter will be further disclosed in subsequent embodiments, and will not be discussed in this step.

[0084] The behavioral score and feedback score are formatted to unify them into the same value range. The formatting process can use a normalization algorithm to unify them into the same value range of [0-1]. It can be understood that the value range of the feedback score itself is [0-1], so it does not need to be normalized. Only the behavioral score needs to be normalized. The normalization algorithm can be implemented by domain experts as needed, using any one of the following: truncation, binning, logarithmic transformation, min-max normalization, center normalization, etc.

[0085] Based on the above, we can derive corresponding behavioral and feedback scores from the advertising data of multiple third-party target advertising products corresponding to each candidate product. These scores are then summed and normalized to obtain the final advertising score. Furthermore, the similarity score between a candidate product and its corresponding third-party target advertising product represents the degree of referenceability of that third-party target advertising product for the candidate product. Therefore, for each candidate product, we calculate the advertising scores of its multiple corresponding third-party target advertising products, multiply them by the average of the corresponding similarity scores, and calculate the reference score for each candidate product. An illustrative formula for calculating the reference score for a single candidate product is as follows:

[0086]

[0087] in: For reference rating of candidate products, For the placement score of the third-party target advertisement products corresponding to the candidate products, The similarity between the candidate product and its corresponding third-party target advertising product is denoted by N, where N is the number of third-party target advertising products corresponding to the candidate product.

[0088] Step S1400: Select multiple candidate products based on the reference scores of each candidate product and push them to the online store.

[0089] Based on the reference ratings of each candidate product, they are sorted from highest to lowest. The top-ranked products are then selected, and an advertising product recommendation list is built using this ranking. This list is then pushed to the client-side of the online store corresponding to the advertising product selection request, where it is displayed and the request is responded to. At this point, online store merchants can refer to the advertising product recommendation list to select products for advertising campaigns.

[0090] 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:

[0091] This application overcomes the drawback of sparse historical advertising data on e-commerce platforms. It can determine a reference score that can be used as a decision-making reference for alternative products by utilizing the advertising data of third-party products similar to the alternative products in the online store, without relying on the online store's own historical advertising data. Thus, alternative products in the online store can be selected based on the reference score and recommended to the online store to assist in product selection for advertising.

[0092] On the other hand, the technical solution of this application is convenient, efficient and low-cost to implement, solves the existing technical pain points of product selection in e-commerce advertising and is applicable to general advertising scenarios.

[0093] Please see Figure 2 In a further embodiment, step S1200, determining multiple third-party target advertising products corresponding to each candidate product based on the descriptive similarity between title texts from the pre-collected third-party advertising products, includes the following steps:

[0094] Step S1210: From the pre-collected third-party advertising products, retrieve the title texts of multiple third-party advertising products that are similar to their title text descriptions for each candidate product;

[0095] An existing open-source Elasticsearch-based search engine can be used to retrieve the title texts of multiple third-party advertising products that are roughly similar to the description of each candidate product from a pre-collected list of third-party advertising products. Generally, the search engine uses the N-Gram segmentation algorithm to segment the title text of each candidate product and all the title texts of the third-party advertising products to obtain the corresponding segmented texts. Further, the BM25 algorithm is used to calculate the similarity between the segmented text of each candidate product and the segmented texts of each third-party advertising product, thereby recalling multiple third-party advertising product title texts with high similarity for each candidate product. The "multiple" can be set as needed by those skilled in the art; 20 is recommended. It is evident that the above implementation can quickly and coarsely filter out a small number of third-party advertising product title texts that are roughly similar to the descriptions of the corresponding candidate products from a large number of third-party advertising product title texts.

[0096] Step S1220: For the title text of each candidate product and its corresponding multiple third-party advertising products, use a pre-trained text encoding model that has converged to extract the deep semantic features of each title text, and obtain the candidate semantic vector corresponding to each candidate product and the third-party semantic vector corresponding to each third-party advertising product.

[0097] Furthermore, from each candidate product, the title texts of multiple third-party advertised products that are basically similar to their title text descriptions are retrieved. A pre-trained neural network model that has reached convergence is used to perform fine screening based on the semantic features of the title text descriptions of each candidate product and its corresponding multiple third-party advertised products.

[0098] The text encoding model is a dual-tower BERT model, where the two branches of the BERT model have the same structure and weights. For each candidate product and its corresponding multiple third-party advertised products' title text, a data pair is formed by combining the title text of each candidate product with the title text of each of its corresponding third-party advertised products. For example, the title text of a candidate product and the title texts of 20 corresponding third-party advertised products constitute 20 data pairs. Each data pair contains the title text of the candidate product and the title text of one of the 20 corresponding third-party advertised products.

[0099] The title texts of the candidate products and the title texts of the third-party advertised products in each data pair are input into the two branches of the dual-tower Bert model. The title texts are segmented to obtain the corresponding segmented texts. The segmented texts are encoded, and the semantic features that represent the descriptions of each segmented word in the segmented texts are extracted. These features are then concatenated to form the deep semantic features of the corresponding title texts. The corresponding candidate semantic vectors for each candidate product and the third-party semantic vectors for each third-party advertised product are obtained by vectorizing the deep semantic features.

[0100] Step S1230: Calculate the similarity score between each candidate semantic vector and its corresponding third-party semantic vectors;

[0101] Any vector distance algorithm can be used, including but not limited to cosine similarity algorithm, Euclidean distance algorithm, Pearson correlation coefficient algorithm, Jaccard coefficient algorithm, etc., to calculate the vector distance between each candidate semantic vector and its corresponding third-party semantic vectors as the similarity score.

[0102] Step S1240: Based on the similarity score, for each candidate product, select a portion of the third-party advertising products that meet the preset conditions from among the multiple third-party advertising products corresponding to it as the third-party target advertising products.

[0103] Based on the similarity scores from high to low, the third-party semantic vectors corresponding to each candidate semantic vector are sorted to obtain the third-party advertising products corresponding to the top-ranked third-party semantic vectors. The recommended number of third-party advertising products is three, but those skilled in the art can set it as needed. For each candidate semantic vector, the third-party advertising products corresponding to each candidate product can be determined as the third-party target advertising products.

[0104] Regarding steps S1230-1240 above, in one embodiment, each candidate semantic vector can be mapped and associated with its corresponding third-party semantic vectors to construct a vector set, which is then stored in the Faiss library. Further, after preprocessing and encapsulating this vector set into an index file for storage, each candidate semantic vector can be used as input to Faiss. Based on the similarity between each candidate semantic vector and its corresponding third-party semantic vectors, multiple third-party semantic vectors with high similarity to each candidate semantic vector are recalled. It is recommended that these multiple third-party semantic vectors be three, but those skilled in the art can set them as needed. Thus, for each candidate product, the third-party advertising products corresponding to multiple third-party semantic vectors with high similarity to its corresponding candidate semantic vector can be determined as the third-party target advertising products. Faiss (Facebook AI Similarity Search) provides an efficient and reliable similarity clustering and retrieval method for massive amounts of data (dense vectors) in high-dimensional space, supporting searches of billions of vectors, and is an approximate nearest neighbor search library.

[0105] In this embodiment, for each candidate product, a search engine is used to quickly and roughly screen out the title texts of multiple third-party advertising products that are roughly similar to the title text description of the corresponding candidate product. Then, based on the semantic features of each title text description using a pre-trained and converged text encoding model, the similarity score between each candidate product and its corresponding third-party target advertisement is calculated. Finally, a portion of the third-party advertising products are finely screened out from the multiple third-party advertising products as third-party target advertising products.

[0106] Please see Figure 3 In a further embodiment, step S1220, the training process of the text encoding model, includes the following steps:

[0107] Step S1221: Obtain a single training sample from the prepared training set. Each training sample in the training set includes the title text of the product, the title text of the third-party advertised product, and a supervision label. The supervision label of the training sample indicates whether the title text of the third-party advertised product of the training sample is similar to the title text description of the product.

[0108] 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.

[0109] Sufficient product title texts corresponding to each category in the product category system can be obtained from the product databases of one or more online stores. The specific number can be set as needed by those skilled in the art. Following step S1210, from the pre-collected third-party advertising products, multiple third-party advertising product title texts that are roughly similar to the title text description of each product are retrieved. Further, manual annotation is used to determine whether the title text of each third-party advertising product is similar to the title text description of its corresponding product, and corresponding supervision labels are assigned accordingly. For example, a supervision label of 1 is assigned for similar descriptions, and a supervision label of 0 is assigned for dissimilar descriptions. Each third-party advertising product title text is associated with its corresponding product title text and supervision labels to construct training samples. All training samples are used to construct a training set.

[0110] Step S1222: After extracting text semantic features from each title text in the training samples by the text encoding model, the similarity score corresponding to the training samples is predicted by the prediction module.

[0111] The text encoding model is a dual-tower BERT model, in which the two branches of the BERT model have the same structure and share weights.

[0112] The title texts of the products in the training samples and the title texts of the third-party advertised products are respectively input into the two branches of the dual-tower BERT model. The title texts are segmented into words to obtain the corresponding word segments. The word segments are encoded, and the semantic features that represent the descriptions of each word in the word segments are extracted. These features are then concatenated to form the deep semantic features of the corresponding title text. The corresponding product semantic vectors and third-party semantic vectors of the third-party advertised products are obtained by vectorizing the deep semantic features.

[0113] Any vector distance algorithm can be used, including but not limited to cosine similarity algorithm, Euclidean distance algorithm, Pearson correlation coefficient algorithm, Jaccard coefficient algorithm, etc., to calculate the vector distance between each product semantic vector and its corresponding third-party semantic vectors as the similarity score.

[0114] Step S1223: Calculate the loss value of the similarity score based on the supervision label corresponding to the training sample. If the model loss value does not reach the preset threshold, update the weights of the model and continue to call other training samples to carry out iterative training until the model converges.

[0115] The 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 cross-entropy loss value is calculated based on the supervision label corresponding to the training sample. When the loss value reaches a preset threshold, it indicates that the 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. Usually, the weight parameters of each part of the model are corrected through backpropagation to make the model closer to convergence. Then, the next sample data in the training set is called to iteratively train the model until the model is trained to a convergent state.

[0116] In this embodiment, the text encoding model is trained with training samples that are precisely labeled with supervisory tags. After the model is trained to convergence, it can make predictions based on the deep semantic features corresponding to the title texts input to its two branches, and accurately predict the similarity between the two title texts.

[0117] Please see Figure 4 In a further embodiment, step S1300, calculating the placement score based on the pre-collected advertising placement data of the third-party target advertising product, includes the following steps:

[0118] Step S1310: Obtain pre-collected advertising data of the target advertising product from a third party. The advertising data includes comment information, number of likes, number of favorites, and advertising duration.

[0119] For each candidate product and its corresponding multiple third-party target advertising products, pre-collected advertising data for each third-party target advertising product is obtained. For ease of discussion, the following focuses on a single third-party target advertising product:

[0120] Specifically, consumers can access the display page of third-party advertised products through advertisements published by those products. On the page, consumers can like, favorite, and other actions related to the advertised products. Consumers who purchase the advertised products can also leave comments. The likes, favorites, and positive reviews reflect a certain degree of consumer liking for the advertised products, indicating the effectiveness of the advertising campaign. Based on this, advertising data for the third-party advertised products can be collected, including comments, number of likes, number of favorites, and advertising duration. Subsequently, a corresponding advertising score can be calculated based on the advertising data to characterize the effectiveness of the third-party advertised products' advertising campaign.

[0121] Step S1320: Calculate the behavior score and feedback score based on the advertising data. The behavior score is determined based on the number of likes, the number of favorites, and the duration of the advertising campaign. The feedback score is determined based on the comment information.

[0122] The behavioral score is determined based on the number of likes, favorites, and ad duration in the ad delivery data. In one embodiment, the number of likes and favorites per unit time can be quantified, i.e., the sum of the number of likes and favorites is divided by the ad delivery duration, and the quantified result is used as the behavioral score. In a preferred embodiment, the number of likes and favorites per unit time can be quantified, i.e., the number of likes is divided by the ad duration, and the number of favorites is divided by the ad duration, respectively. Then, the quantified results are matched with corresponding preset weights, weighted, and then summed. For example, the weight matched with the number of likes per unit time is 0.4, and the weight matched with the number of favorites per unit time is 0.6. The sum is used as the behavioral score.

[0123] The feedback score is determined based on comment information in the advertising data. In one embodiment, the percentage of positive reviews in the total number of comments is used as the feedback score. In a preferred embodiment, based on the determined percentage of positive reviews, a Bayesian smoothing algorithm is used to determine an adjustment parameter based on the comment information of the corresponding third-party target advertising product during its historical advertising campaigns. This adjustment parameter is then used to smooth the percentage, resulting in a smoothed percentage as the feedback score. The determination of the adjustment parameter will be further disclosed in subsequent embodiments, and will not be discussed in this step.

[0124] Step S1330: Normalize the behavior score, add the corresponding normalization result to the feedback score and normalize again to obtain the placement score of the third-party target advertising product.

[0125] The behavior score and feedback score are formatted to unify them into the same value range of [0-1]. It can be understood that the value range of the feedback score itself is [0-1], so it does not need to be normalized. Only the behavior score needs to be normalized using a normalization algorithm. The normalization algorithm can be implemented by domain experts as needed, using any one of the following: truncation, binning, logarithmic transformation, min-max normalization, center normalization, etc.

[0126] Furthermore, the normalized behavioral score is added to the feedback score, and normalization is performed again to obtain the placement score of the third-party target advertising product.

[0127] Based on the above steps S1310-S1330, the corresponding placement score can be calculated according to the advertising placement data of multiple third-party target advertising products corresponding to each candidate product.

[0128] In this embodiment, it is possible to easily and efficiently obtain the corresponding quantitative values ​​as the advertising score, which accurately reflects the advertising effectiveness of the third-party target advertising products.

[0129] Please see Figure 5 In a further embodiment, step S1320, calculating the behavioral score and feedback score based on the advertising data, includes the following steps:

[0130] Step S1321: Use a pre-trained, convergent sentiment analysis model to determine the positive review information in the advertising data;

[0131] Generally, after purchasing products listed on an e-commerce platform, users can provide feedback by leaving comments. These comments are text-based and are typically tagged with corresponding categories, including but not limited to positive and negative reviews. Therefore, sufficient positive and negative reviews of listed products on the e-commerce platform can be collected and labeled with corresponding supervisory tags. For example, a supervisory tag of 1 is assigned to positive reviews, and a supervisory tag of 0 is assigned to negative reviews. Then, each comment, associated with its corresponding supervisory tag, is used as a training sample to construct a training set.

[0132] The FastText model can be used as the sentiment analysis model. A single training sample from the training set is input into the FastText model to extract the classification features of the training sample. These classification features represent the type of comment information in the training sample, resulting in a vectorized classification encoding vector. A classification function is used to linearly transform the classification encoding vector to a preset classification space, obtaining the classification probability of the mapped classification space. The classification function can be Sigmoid, Softmax, etc., and those skilled in the art can choose and implement it as needed. The preset classification space includes a positive class space and a negative class space. The positive class space represents positive reviews, and the negative class space represents negative reviews. Further, the loss value of the classification probability is calculated based on the supervision label corresponding to the training sample. If the model loss value does not reach a preset threshold, the model weights are updated, and iterative training continues using other training samples until the model converges. It can be understood that after the model is trained to convergence, it can learn the ability to determine the type of the input title text.

[0133] Based on the above, a sentiment analysis model that has been trained to convergence can be prepared in advance.

[0134] A pre-trained, convergent sentiment analysis model is input into the advertising data containing all comments. For each individual comment, its corresponding classification features are extracted, resulting in a vectorized classification encoding vector. This vector is then linearly transformed using a classification function and mapped to a predefined classification space. The mapped classification space is used as the category to which the comment belongs. In this way, the category of each comment can be determined, and based on the category of each comment, comments belonging to the "positive" category can be identified from all comments.

[0135] Step S1322: Calculate the proportion of positive reviews to all reviews, and use a smoothing algorithm to process this proportion to obtain a feedback score.

[0136] The number of positive reviews in the advertising data is counted, along with the total number of all reviews. The number of positive reviews is divided by the total number of all reviews to obtain the percentage. A Bayesian smoothing algorithm is used to determine the appropriate adjustment parameters based on the historical review information of the third-party target advertising product corresponding to the review information. The percentage is then smoothed using these adjustment parameters to obtain the smoothed percentage as the feedback score.

[0137] Step S1323: Calculate the sum of the number of likes and favorites in the ad placement information and divide it by the ad placement duration to obtain the corresponding behavior score.

[0138] The exemplary formula is as follows:

[0139]

[0140] in: Scoring behavior For clicks, For the number of collections, This refers to the duration of the advertisement.

[0141] By referring to the above behavioral scoring formula, the behavioral score of the corresponding third-party target advertising product can be calculated.

[0142] In this embodiment, based on the advertising data of the third-party target advertising products, the corresponding behavioral scores and feedback scores are accurately quantified, so that the advertising effectiveness of the third-party target advertising can be accurately represented by quantitative values.

[0143] Please see Figure 6 In a further embodiment, step S1322, processing the quantity ratio using a smoothing algorithm to obtain a feedback score, includes the following steps:

[0144] Step S13221: Obtain the comment information of the third-party target advertising product for each preset interval within a recent preset time period, and calculate the proportion of positive comments in the comment information for each preset interval.

[0145] The current timestamp is taken as the current timestamp of the review information of the third-party target advertising product. Furthermore, the time from the current timestamp is preset as the recent preset time, such as 3 months. The time interval corresponding to each interval within the preset time is preset as the preset interval time, such as one week. The specific time corresponding to the recent preset time and the preset interval time can be set as needed by those skilled in the art.

[0146] For the past three months, review information of the third-party target advertising products has been obtained once a week. A pre-trained and converged sentiment analysis model has been called to determine the number of positive reviews in the weekly review information. Based on this, the number of positive reviews is divided by the total number of reviews for each week to obtain the percentage of positive reviews in the weekly review information.

[0147] Step S13222: Based on the proportion of quantities corresponding to each preset interval duration, the preset first smoothing parameter and second smoothing parameter are obtained using the maximum likelihood algorithm;

[0148] Those skilled in the art should know that, in order to make the current quantity proportion closer to the normal quantity proportion of the corresponding third-party target advertising product, a Bayesian smoothing algorithm can be used. This involves using multiple past quantity proportions of the corresponding third-party target advertising product, i.e., the quantity proportion corresponding to each preset interval, and employing a maximum likelihood algorithm to obtain the preset quantity proportions required by the Bayesian algorithm. That is, the first adjustment parameter, That is, the second adjustment parameter. Then, the obtained first adjustment parameter and the second adjustment parameter can be used to smoothly adjust the current quantity ratio, so that the current quantity ratio is closer to the true quantity ratio, thereby improving confidence and accuracy.

[0149] By using the weekly volume percentage of third-party targeted advertising products, the maximum likelihood algorithm can be used to fit the preset parameters required by the Bayesian algorithm. That is, the first adjustment parameter, That is, the second adjustment parameter.

[0150] Step S13223: Adjust the current quantity ratio according to the first smoothing parameter and the second smoothing parameter, and obtain the adjusted quantity ratio as a feedback score.

[0151] Example formulas are shown below:

[0152]

[0153] in: For feedback rating, For the first adjustment parameter, C is the second adjustment parameter, C is the number of positive reviews in the review information, and L is the total number of reviews.

[0154] Referring to the above feedback scoring formula, it is possible to fit the historical quantity proportion. , The current quantity ratio of the corresponding third-party target advertising products is smoothly adjusted, and the adjusted quantity ratio is calculated as the feedback score.

[0155] In this embodiment, by applying a Bayesian algorithm to smooth the current proportion of third-party target advertising products using historical advertising data, the adjusted proportion is used as the feedback score for the corresponding third-party target advertising products. It is evident that the Bayesian smoothed proportion more closely approximates objective reality, improving confidence and accuracy.

[0156] Please see Figure 7 This application provides an advertising product selection device to meet one of its objectives. It is a functional embodiment of the advertising product selection method of this application. The device includes a request response module 1100, a similarity determination module 1200, a rating determination module 1300, and a rating optimization module 1400. Specifically: the request response module 1100 responds to advertising product selection requests from online stores and obtains the title text of each candidate product from the online store's product database. The online store operates on an independent site of an e-commerce platform. The similarity determination module 1200 determines multiple third-party target advertising products corresponding to each candidate product based on the descriptive similarity between the title texts from pre-collected third-party advertising products. The rating determination module 1300 calculates the advertising score of the third-party target advertising products based on pre-collected advertising data, and determines a reference score for each candidate product based on the advertising scores of the multiple third-party target advertising products corresponding to each candidate product. The rating optimization module 1400 optimizes multiple candidate products based on their reference scores and pushes them to the online store.

[0157] In a further embodiment, the similarity determination module 1200 includes: a similarity retrieval submodule, used to retrieve the title texts of multiple third-party advertising products that are similar to the title text description of each candidate product from the pre-collected third-party advertising products; a vectorization representation submodule, used to extract the deep semantic features of each title text of each candidate product and its corresponding multiple third-party advertising products using a pre-trained and converged text encoding model to obtain the candidate semantic vector corresponding to each candidate product and the third-party semantic vector corresponding to each third-party advertising product; a similarity retrieval submodule, used to calculate the similarity score between each candidate semantic vector and its corresponding third-party semantic vectors; and a target determination submodule, used to determine, based on the similarity score, a portion of the third-party advertising products that meet preset conditions from the multiple third-party advertising products corresponding to each candidate product as third-party target advertising products.

[0158] In a further embodiment, the training process of the text encoding model in the vectorization representation submodule includes the following steps: a training sample acquisition module, used to acquire a single training sample in a prepared training set, wherein each training sample in the training set contains the title text of a product, the title text of a third-party advertised product, and a supervision label, wherein the supervision label of the training sample indicates whether the title text of the third-party advertised product in the training sample is similar to the title text description of the product; a rating prediction module, used to extract text semantic features from each title text in the training sample by the text encoding model, and then predict the similarity score corresponding to the training sample by the prediction module; and an iterative training module, used to calculate the loss value of the similarity score according to the supervision label corresponding to the training sample, and when the model loss value does not reach a preset threshold, to update the weights of the model and continue to call other training samples to perform iterative training until the model converges.

[0159] In a further embodiment, the scoring determination module 1300 includes: an advertising delivery data acquisition submodule, used to acquire pre-collected advertising delivery data of a third-party target advertising product, the advertising delivery data including comment information, number of likes, number of favorites, and advertising delivery duration; a feedback score and behavior score determination submodule, used to calculate a behavior score and a feedback score based on the advertising delivery data, the behavior score being determined based on the number of likes, number of favorites, and advertising delivery duration, and the feedback score being determined based on the comment information; and a delivery score acquisition submodule, used to normalize the behavior score, add the corresponding normalization result to the feedback score, and perform normalization to obtain the delivery score of the third-party target advertising product.

[0160] In a further embodiment, the feedback scoring and behavior scoring determination submodule includes: a sentiment analysis unit, used to determine the positive review information in the advertising data using a pre-trained and converged sentiment analysis model; a feedback scoring calculation and processing unit, used to calculate the proportion of positive review information to all review information, and use a smoothing algorithm to process the proportion to obtain a feedback score; and a behavior scoring calculation unit, used to calculate the sum of the number of likes and favorites in the advertising information divided by the advertising duration to obtain the corresponding behavior score.

[0161] In a further embodiment, the feedback score calculation and processing unit includes: a historical quantity proportion calculation subunit, used to obtain the comment information of the third-party target advertising product at each preset interval within a recent preset time period, and calculate the quantity proportion of positive comments in the comment information at each preset interval; a smoothing parameter acquisition subunit, used to obtain preset first smoothing parameters and second smoothing parameters based on the quantity proportion corresponding to each preset interval using a maximum likelihood algorithm; and a quantity proportion adjustment subunit, used to adjust the current quantity proportion based on the first smoothing parameters and the second smoothing parameters, and obtain the adjusted quantity proportion as the feedback score.

[0162] In a further embodiment, the scoring determination module 1300 includes: a reference scoring determination submodule, used to determine the reference score of each candidate product by multiplying the similarity score of each third-party target advertising product by the placement score, adding them together, and averaging the results for each candidate product.

[0163] To address the aforementioned technical problems, embodiments of this application also provide computer equipment. For example... Figure 8 The 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, the processor can implement an advertising placement and product selection method. The processor of the computer device provides computing and control capabilities to support the operation of the entire computer device. The memory of the computer device may store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor can execute the advertising placement and product selection 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 8The 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.

[0164] In this embodiment, the processor is used to execute... Figure 7 The system contains the specific functions of each module and its sub-modules. The memory stores the program code and various 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 placement and product selection device of this application. The server can call the server's program code and data to execute the functions of all sub-modules.

[0165] 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 placement and product selection method of any embodiment of this application.

[0166] 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.

[0167] In summary, this application does not rely on the online store's own advertising data throughout the entire process. Instead, it uses matching third-party advertising data to evaluate the reference scores for each candidate product of the online store. It has universal applicability and can meet the advertising needs of most merchants.

[0168] 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.

[0169] 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. A method for selecting products for advertising placement, characterized in that, Includes the following steps: In response to an online store's advertising product selection request, the title text of each candidate product in the online store's product database is obtained, wherein the online store operates on an independent site of an e-commerce platform; From the pre-collected third-party advertising products, multiple third-party target advertising products corresponding to each candidate product are identified based on the descriptive similarity between the title texts; The ad placement score is calculated based on the pre-collected ad placement data of third-party target ad products. The reference score for each candidate product is determined by the ad placement scores of multiple third-party target ad products corresponding to each candidate product. Based on the reference scores of each candidate product, several candidate products are selected and pushed to the online store.

2. The advertising product selection method according to claim 1, characterized in that, The step of identifying multiple target third-party advertising products corresponding to each candidate product from pre-collected third-party advertising products based on the descriptive similarity between title texts includes the following steps: From the pre-collected third-party advertising products, retrieve the title texts of multiple third-party advertising products that are similar to their title text descriptions for each candidate product; For each candidate product and its corresponding multiple third-party advertising products, a pre-trained text encoding model that has converged is used to extract the deep semantic features of each title text, thereby obtaining the candidate semantic vector corresponding to each candidate product and the third-party semantic vector corresponding to each third-party advertising product. Calculate the similarity score between each candidate semantic vector and its corresponding third-party semantic vectors; Based on the similarity score, for each candidate product, a portion of the third-party advertising products that meet the preset conditions are determined from among the multiple third-party advertising products corresponding to it as the third-party target advertising products.

3. The advertising product selection method according to claim 2, characterized in that, The training process of the text encoding model includes the following steps: Obtain a single training sample from the prepared training set. Each training sample in the training set includes the title text of the product, the title text of the third-party advertised product, and a supervision label. The supervision label of the training sample indicates whether the title text of the third-party advertised product of the training sample is similar to the title text description of the product. After extracting text semantic features from each title text in the training samples by the text encoding model, the similarity score corresponding to the training samples is predicted by the prediction module. The loss value of the similarity score is calculated based on the supervision label corresponding to the training sample. When the loss value of the model does not reach the preset threshold, the model weights are updated, and other training samples are called to carry out iterative training until the model converges.

4. The advertising product selection method according to claim 1, characterized in that, The steps for calculating the ad placement score based on the pre-collected ad placement data of the target product from a third party include the following: Obtain advertising placement data pre-collected from third-party target advertising products, wherein the advertising placement data includes comment information, number of likes, number of favorites, and advertising placement duration; A behavioral score and a feedback score are calculated based on the advertising data. The behavioral score is determined based on the number of likes, the number of favorites, and the duration of the advertising campaign. The feedback score is determined based on the comment information. The behavioral score is normalized, and the corresponding normalization result is added to the feedback score and normalized again to obtain the placement score of the third-party target advertising product.

5. The advertising product selection method according to claim 4, characterized in that, The steps for calculating behavioral scores and feedback scores based on advertising data include the following: A pre-trained, convergent sentiment analysis model was used to identify positive reviews in the advertising data. Calculate the percentage of positive reviews out of all reviews, and use a smoothing algorithm to process this percentage to obtain a feedback score. The sum of likes and favorites in the ad placement information is divided by the ad placement duration to obtain the corresponding behavioral score.

6. The advertising product selection method according to claim 5, characterized in that, The quantity percentage is processed using a smoothing algorithm to obtain feedback scores, including the following steps: Obtain the comment information of the third-party target advertising product at each preset interval within a recent preset time period, and calculate the proportion of positive comments in the comment information at each preset interval. Based on the proportion of quantities corresponding to each preset interval duration, the preset first smoothing parameter and second smoothing parameter are obtained using the maximum likelihood algorithm; The current quantity percentage is adjusted based on the first smoothing parameter and the second smoothing parameter, and the adjusted quantity percentage is used as a feedback score.

7. The advertising product selection method according to claim 1, characterized in that, The reference score for each candidate product is determined by the placement scores of multiple third-party target advertising products corresponding to each candidate product. This includes multiplying the similarity score of each third-party target advertising product by its placement score, adding them together and averaging the results to determine the reference score for each candidate product.

8. An advertising placement and product selection device, characterized in that, include: The request and response module is used to respond to the advertising product selection request of the online store and obtain the title text of each candidate product in the product database of the online store, wherein the online store runs on an independent site of the e-commerce platform. The similarity determination module is used to determine multiple third-party target advertising products corresponding to each candidate product from the pre-collected third-party advertising products based on the descriptive similarity between the title texts; The scoring determination module is used to calculate the placement score of the target advertising products based on the pre-collected advertising placement data of the third-party target advertising products. The reference score of each candidate product is determined by the placement scores of multiple third-party target advertising products corresponding to each candidate product. The rating and selection module is used to select multiple candidate products based on the reference ratings of each candidate product and push them to the online store.

9. A computer device comprising a central processing unit and a memory, characterized in that, The central processing unit is used to invoke and run a computer program stored in the memory to perform the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores, in the form of computer-readable instructions, a computer program implemented according to any one of claims 1 to 7, which, when invoked by a computer, executes the steps included in the corresponding method.

Images