Intelligent advertisement management method and system based on AI enhanced search

Abstract

The invention provides an intelligent advertisement management method and system based on AI enhanced search, and relates to the technical field of advertisement management, and the method comprises the steps: obtaining a search request, collecting multi-dimensional information, and carrying out the vector conversion of the search request according to the multi-dimensional information, and obtaining a search vector; retrieving based on the search vector to obtain an advertisement candidate set; performing advertisement effect prediction according to the advertisement candidate set to obtain a putting value score; and performing advertisement putting on the advertisement candidate set according to the putting value score. According to the method and the device, the technical problem in the prior art that the semantic matching accuracy between the search request and the advertisement content is insufficient and the correlation and the putting effect of advertisement putting in a search scene are further influenced can be solved, and the technical goal of accurate advertisement matching and putting management based on search request semantic enhancement representation is realized; the technical effects of improving the semantic consistency of the advertisement and the real demand of the user in the search scene, and improving the advertisement putting correlation and the overall putting effect are achieved.

Description

Technical Field

[0001] This application relates to the field of advertising management technology, and in particular to intelligent advertising management methods and systems based on AI-enhanced search. Background Technology

[0002] With the continuous development of internet information services and digital business activities, search engines have gradually become the main entry point for users to obtain information and make consumption decisions. Advertising based on search scenarios has also become an important way for advertisers to achieve precise marketing. Especially when the search request is highly relevant to the advertising content, it can effectively improve advertising conversion efficiency and platform revenue. Therefore, how to achieve highly relevant filtering and precise targeting of advertisements in search scenarios has become an important technical direction for the continuous optimization of search advertising systems.

[0003] Currently, most existing search advertising management technologies still rely on keyword matching mechanisms as the core basis for ad selection and placement. They typically match keywords in user search requests with preset keywords for ads, and then use simple bidding rules or historical click data to make ad ranking and placement decisions. While this type of technology has low implementation costs, it generally suffers from insufficient semantic understanding of search requests in practical applications. Especially when users use natural language, combined grammar, or implicit intent to search, existing technologies struggle to accurately depict the true semantic meaning of search requests. As a result, the ad selection process remains at the surface matching level and fails to fully utilize the contextual information and user history information contained in the search request, leading to a low degree of semantic matching between ads and users' actual needs.

[0004] In summary, existing technologies suffer from technical problems such as insufficient accuracy in semantic matching between search requests and advertising content due to the over-reliance on keyword-level matching in the search advertising management process and the lack of deep modeling and enhancement mechanisms for the semantic information of search requests. This further affects the relevance and effectiveness of advertising in search scenarios. Summary of the Invention

[0005] The purpose of this application is to provide an intelligent advertising management method and system based on AI-enhanced search, in order to solve the technical problems in the existing technology, which are due to the excessive reliance on keyword-level matching in the search advertising management process and the lack of deep modeling and enhancement processing mechanisms for the semantic information of search requests, resulting in insufficient accuracy of semantic matching between search requests and advertising content, which further affects the relevance and effectiveness of advertising in search scenarios.

[0006] In view of the above problems, this application provides an intelligent advertising management method and system based on AI-enhanced search.

[0007] In a first aspect, this application provides an intelligent advertising management method based on AI-enhanced search, implemented through an intelligent advertising management system based on AI-enhanced search, comprising: acquiring a search request and collecting multi-dimensional information; performing vector transformation on the search request based on the multi-dimensional information to obtain a search vector; retrieving an advertising candidate set based on the search vector; predicting advertising effectiveness based on the advertising candidate set to obtain a placement value score; and placing advertisements on the advertising candidate set based on the placement value score.

[0008] Preferably, the AI-enhanced search-based intelligent advertising management method further includes: receiving a search request input by a user, wherein the search request includes keywords, natural language, and combined grammar; collecting contextual information associated with the search request, wherein the contextual information includes at least search time, terminal type, geographical region, search entry source, and session identifier; obtaining historical records corresponding to the user, wherein the historical records include historical search records, historical ad click records, conversion behavior records, and dwell time information; and using the contextual information and the historical records as multi-dimensional information.

[0009] Preferably, the AI-enhanced search-based intelligent advertising management method further includes: performing word segmentation, stop word removal, and text normalization on the search request to obtain preliminary text features; performing semantic encoding on the preliminary text features to obtain initial semantic vectors at the word level; aggregating the initial semantic vectors by weighted averaging to generate global semantic vectors at the sentence level; dynamically adjusting the global semantic vectors by combining the historical records and the context information to obtain enhanced semantic vectors; and performing principal component analysis dimensionality reduction and normalization on the enhanced semantic vectors to obtain the search vector.

[0010] Preferably, the AI-enhanced search-based intelligent advertising management method further includes: converting the historical records into historical behavior vectors; converting the context information into context feature vectors; and weighting and fusing the global semantic vector, historical behavior vector, and context feature vector based on corresponding preset weight coefficients to obtain the enhanced semantic vector.

[0011] Preferably, the AI-enhanced search-based intelligent advertising management method further includes: obtaining advertising data from an advertising resource library; cleaning and concatenating the advertising title, advertising description, and advertising landing page text in the advertising data to form advertising source text; calling a semantic coding network to extract content features from the advertising source text to obtain advertising source features; pooling the advertising source features and mapping them to a real number vector of a preset dimension to obtain an advertising semantic vector; and matching the advertising semantic vector with the search vector to obtain a matching set of advertising candidates.

[0012] Preferably, the AI-enhanced search-based intelligent advertising management method further includes: calculating the vector space distance between the search vector and the advertising semantic vector to obtain a semantic similarity score; comparing the semantic similarity score with a preset relevance threshold, removing advertisements with semantic similarity scores lower than the preset relevance threshold, and generating an advertising candidate set.

[0013] Preferably, the AI-enhanced search-based intelligent advertising management method further includes: extracting multi-dimensional feature information for each candidate advertisement in the candidate advertisement set, wherein the multi-dimensional feature information includes at least historical interaction statistics and advertising revenue features, wherein the historical interaction statistics include historical exposure, historical clicks, and historical conversions; discretizing, normalizing, and cross-combining the multi-dimensional feature information to map the processed multi-dimensional feature information into an input vector; inputting the input vector into an advertising performance prediction model to output the estimated click-through rate and estimated conversion rate of the candidate advertisement, calculating the expected revenue value, wherein the expected revenue value is the product of the estimated click-through rate, the estimated conversion rate, and the revenue per advertisement; and calculating the comprehensive score of the candidate advertisement based on the weighted average of the estimated click-through rate, the conversion rate, and the expected revenue value to generate the placement value score.

[0014] Preferably, the AI-enhanced search-based intelligent advertising management method further includes: collecting historical search advertising display logs, obtaining historical display multi-dimensional features and corresponding click-through rates and conversion rates; building a deep neural network, using the historical display multi-dimensional features as input, and the click-through rate and conversion rate as supervision targets, optimizing network parameters through a backpropagation algorithm until convergence, to obtain the advertising effect prediction model.

[0015] Preferably, the AI-enhanced search-based intelligent advertising management method further includes: obtaining the placement value score of each candidate advertisement in the candidate advertisement set; sorting the candidate advertisements from high to low according to the placement value score to generate an advertisement sequence; generating an advertising placement decision instruction according to the advertisement sequence and the layout configuration of the search results page; the advertising placement decision instruction is used to determine the number of target advertisements to be displayed on the search results page, the display position of each target advertisement, and the display order; and executing the advertising placement according to the advertising placement decision instruction.

[0016] Secondly, this application also provides an AI-enhanced search-based intelligent advertising management system for executing the AI-enhanced search-based intelligent advertising management method as described in the first aspect, comprising: a search vector acquisition module for acquiring a search request and collecting multi-dimensional information, and performing vector transformation on the search request based on the multi-dimensional information to obtain a search vector; an advertising candidate set acquisition module for retrieving and obtaining an advertising candidate set based on the search vector; a placement value score acquisition module for predicting advertising effectiveness based on the advertising candidate set to obtain a placement value score; and an advertising placement module for placing advertisements on the advertising candidate set based on the placement value score.

[0017] The technical solution provided in this application has at least the following technical effects or advantages: by achieving the technical goal of accurate matching and delivery management of advertisements based on semantically enhanced representation of search requests, it achieves the technical effects of improving the semantic consistency between advertisements and users' real needs, improving the relevance of advertisement delivery and the overall delivery effect in search scenarios.

[0018] The above description is merely an overview of the technical solution of this application. To enable a clearer understanding of the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the intelligent advertising management method based on AI-enhanced search proposed in this application.

[0021] Figure 2 This is a schematic diagram of the structure of the AI-enhanced search-based intelligent advertising management system of this application.

[0022] The attached diagrams are labeled as follows: Module 1 for obtaining search vectors, Module 2 for obtaining ad candidate sets, Module 3 for obtaining ad placement value scores, and Module 4 for ad placement. Detailed Implementation

[0023] This application provides an intelligent advertising management method and system based on AI-enhanced search, addressing the technical problem in existing technologies where the over-reliance on keyword-level matching and the lack of deep modeling and enhancement mechanisms for the semantic information of search requests lead to insufficient accuracy in semantic matching between search requests and advertising content, further impacting the relevance and effectiveness of advertising in search scenarios. It achieves the technical goal of precise advertising matching and delivery management based on semantically enhanced representations of search requests, thereby improving the semantic consistency between advertisements and users' actual needs, enhancing advertising relevance, and improving overall delivery effectiveness in search scenarios.

[0024] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0025] Example 1, please refer to the appendix. Figure 1 This application provides an intelligent advertising management method based on AI-enhanced search, which is applied to an intelligent advertising management system based on AI-enhanced search, and specifically includes the following steps: A search request is acquired and multi-dimensional information is collected. The search request is then transformed into a vector based on the multi-dimensional information to obtain a search vector.

[0026] Furthermore, this application also includes: receiving a search request input by a user, wherein the search request includes keywords, natural language, and combined grammar; collecting contextual information associated with the search request, wherein the contextual information includes at least search time, terminal type, geographical region, search entry source, and session identifier; obtaining historical records corresponding to the user, wherein the historical records include historical search records, historical ad click records, conversion behavior records, and dwell time information; and using the contextual information and the historical records as multi-dimensional information.

[0027] Furthermore, this application also includes: performing word segmentation, stop word removal, and text normalization on the search request to obtain preliminary text features; performing semantic encoding on the preliminary text features to obtain initial semantic vectors at the word level; aggregating the initial semantic vectors by weighted averaging to generate global semantic vectors at the sentence level; dynamically adjusting the global semantic vectors by combining the historical records and the context information to obtain enhanced semantic vectors; and performing principal component analysis dimensionality reduction and normalization on the enhanced semantic vectors to obtain the search vector.

[0028] Furthermore, this application also includes: converting the historical records into historical behavior vectors; converting the context information into context feature vectors; and weighting and fusing the global semantic vector, historical behavior vector, and context feature vector based on corresponding preset weight coefficients to obtain the enhanced semantic vector.

[0029] Specifically, receiving user-input search requests refers to obtaining the query content actively submitted by the user through the input interface of the search system. The search request can take the form of keywords, natural language descriptions, or a combination of various syntaxes to represent the user's query intent and needs in the current search scenario. Keywords are used to accurately indicate the target object, natural language is used to carry the user's semantic description of needs, and combined syntaxes are used to balance structured constraints and the flexibility of semantic expression, thereby providing a basic data source for subsequent semantic understanding and processing.

[0030] Further contextual information associated with the search request is collected. The contextual information describes the environmental state and interaction conditions when the search request occurs. The search time reflects the temporal characteristics of the search behavior. The terminal type distinguishes the interaction differences under different device forms. The geographical region describes the spatial attributes of the search request. The search entry source represents the triggering path of the search request. The session identifier is used to associate multiple requests from the same user in continuous operation process to achieve a temporal consistency description of the search behavior.

[0031] After collecting contextual information, the system obtains the user's historical records. These records reflect the user's behavioral preferences and interaction characteristics within a historical period. Historical search records describe the user's past search focus, historical ad click records characterize the user's interest in ad content, conversion behavior records depict the user's actual decision-making results after interacting with ads or content, and dwell time information reflects the user's level of attention and engagement with relevant content, thus providing a basis for user behavior modeling.

[0032] The process of segmenting, removing stop words, and normalizing text in search requests involves splitting the original text of the search request into several semantic units according to preset segmentation rules, removing functional words that contribute little to semantic expression, and standardizing capitalization, symbols, numbers, and synonyms in the text. This reduces text noise and expression differences, forming preliminary text features that reflect the core semantics of the search request for subsequent semantic analysis.

[0033] After obtaining the initial text features, semantic encoding of the initial text features refers to vectorizing each semantic unit through a semantic representation model, so that each word generates a numerical vector that can represent its semantic meaning, thereby obtaining the initial semantic vector at the word level. The initial semantic vector is used to characterize the relative positional relationship and semantic similarity of words in the semantic space.

[0034] Aggregating the initial semantic vector by weighted averaging means assigning corresponding weights to different words according to their importance in the search request, and then performing weighted summation on the initial semantic vectors corresponding to each word to generate a sentence-level global semantic vector that can represent the overall semantic meaning of the search request, so as to avoid the bias of a single word on the overall semantic expression.

[0035] Transforming historical records into historical behavior vectors refers to extracting features and quantifying information from users' historical search records, historical ad click records, conversion behavior records, and dwell time information. By encoding, statistically analyzing, and normalizing different behavior types, discrete historical behavior data is mapped into vector forms that can represent users' long-term interests, preferences, and behavioral habits, thus forming historical behavior vectors for calculation and modeling.

[0036] After vectorizing the historical records, the context information is transformed into a context feature vector. This means encoding the context information such as the search time, terminal type, geographical region, search entry source, and session identifier corresponding to the search request, and then transforming the context information into a context feature vector that can reflect the current search environment state and interaction conditions through numerical mapping or embedding representation, so as to characterize the real-time scene features of the search request.

[0037] After obtaining the global semantic vector, historical behavior vector, and context feature vector respectively, weighted fusion is performed based on the corresponding preset weight coefficients. This means setting corresponding weights according to the contribution of different vectors in expressing search intent, and performing linear or nonlinear combination processing on each vector, thereby generating an enhanced semantic vector that comprehensively reflects text semantics, user historical preferences, and current context features within a unified vector space.

[0038] Principal component analysis (PCA) for dimensionality reduction and normalization of enhanced semantic vectors refers to removing redundant features and retaining the main semantic components through dimensionality reduction algorithms, while simultaneously performing scale unification on the vectors to improve the stability and efficiency of vector computation, ultimately obtaining search vectors for subsequent ad retrieval and matching.

[0039] Based on the search vector, an ad candidate set is obtained.

[0040] Furthermore, this application also includes: obtaining advertising data from an advertising resource library; cleaning and concatenating the advertising title, advertising description, and advertising landing page text in the advertising data to form advertising source text; calling a semantic coding network to extract content features from the advertising source text to obtain advertising source features; pooling the advertising source features and mapping them to a real number vector of a preset dimension to obtain an advertising semantic vector; and matching the advertising semantic vector with the search vector to obtain a matching set of advertising candidates.

[0041] Furthermore, this application also includes: calculating the vector space distance between the search vector and the advertising semantic vector to obtain a semantic similarity score; comparing the semantic similarity score with a preset relevance threshold, eliminating advertisements with semantic similarity scores lower than the preset relevance threshold, and generating an advertising candidate set.

[0042] Specifically, obtaining advertising data from the advertising resource library refers to retrieving advertising content information related to the current search scenario from a pre-built and continuously maintained advertising storage system. The advertising data includes at least the advertising title, advertising description, and text content corresponding to the advertising landing page. Invalid characters, duplicate content, and format differences in the advertising data are cleaned. At the same time, the cleaned advertising title, advertising description, and advertising landing page text are concatenated according to preset rules to form advertising source text that can completely represent the semantic information of the advertisement.

[0043] After the advertising source text is generated, a semantic encoding network is invoked to extract content features from it. This involves modeling the words, phrases, and contextual relationships within the advertising source text using a semantic representation model, and mapping them into intermediate representations that reflect the semantic features of the advertising content. This yields advertising source features for subsequent vectorization processing. The semantic encoding network is a neural network model used to map textual information into vectorized semantic representations. It embeds each semantic unit in the text and combines this with contextual calculations to generate feature representations that reflect the overall semantic structure and semantic similarity relationships of the advertising content, thereby achieving automatic encoding and feature extraction of the semantic information of the advertising source text.

[0044] Pooling the advertising source features and mapping them to real-number vectors of a preset dimension means extracting representative semantic information by aggregating the advertising source features in the time or semantic dimension, and converting them into a fixed-dimensional, computable real-number vector form, thereby generating an advertising semantic vector for advertising semantic matching.

[0045] Calculating the vector space distance between the search vector and the advertising semantic vector refers to calculating the distance or similarity between the search vector representing the semantic features of the search request and the advertising semantic vector representing the semantic features of the advertising content within a unified vector representation space. The vector space distance is used to measure the degree of closeness between the two at the semantic level, and the distance value is converted into a numerical semantic similarity score to reflect the degree of semantic matching between the search request and the advertising content.

[0046] After obtaining the semantic similarity score, the semantic similarity score is compared with the preset relevance threshold. This means that the semantic similarity score corresponding to each advertisement is judged against the pre-set relevance judgment criteria. The preset relevance threshold is used to limit the minimum semantic relevance requirements that the advertisement and the search request should meet. Advertisements that do not meet the relevance conditions are filtered out, and advertisements with semantic similarity scores lower than the preset relevance threshold are removed from the candidate range. Finally, a set of candidate advertisements that meet the semantic relevance requirements is generated.

[0047] Based on the set of advertising candidates, advertising effectiveness is predicted to obtain a placement value score.

[0048] Furthermore, this application also includes: extracting multi-dimensional feature information for each candidate advertisement in the candidate advertisement set, wherein the multi-dimensional feature information includes at least historical interaction statistical features and advertising revenue features, wherein the historical interaction statistical features include historical exposure, historical clicks, and historical conversions; discretizing, normalizing, and cross-combining the multi-dimensional feature information, and mapping the processed multi-dimensional feature information into an input vector; inputting the input vector into an advertising performance prediction model, outputting the estimated click-through rate and estimated conversion rate of the candidate advertisement, and calculating the expected revenue value, wherein the expected revenue value is the product of the estimated click-through rate, the estimated conversion rate, and the revenue per advertisement; and calculating the comprehensive score of the candidate advertisement based on the weighted average of the estimated click-through rate, the conversion rate, and the expected revenue value to generate the placement value score.

[0049] Furthermore, this application also includes: collecting historical search ad display logs, obtaining historical display multidimensional features and corresponding click-through rates and conversion rates; building a deep neural network, using the historical display multidimensional features as input, and the click-through rate and conversion rate as supervision targets, optimizing network parameters through a backpropagation algorithm until convergence, to obtain the ad performance prediction model.

[0050] Specifically, extracting multidimensional feature information for each candidate ad in the candidate ad set means acquiring feature data that reflects the historical performance and revenue capability of each ad that enters the candidate range. The multidimensional feature information includes at least historical interaction statistics and ad revenue features. Historical interaction statistics are used to characterize the user response to the ad in the historical delivery process, specifically including the historical exposure corresponding to the number of times the ad was displayed, the historical click volume of users clicking on the ad, and the historical conversion volume of users completing the target behavior, thereby providing an objective data basis for ad performance evaluation.

[0051] After extracting multidimensional feature information, the multidimensional feature information is discretized, normalized and cross-combined. This means dividing continuous or features with large distribution differences into intervals or numerical mappings, unifying the scale of features with different dimensions, and constructing new features that can reflect the relationship between features through feature combination. In this way, the processed multidimensional feature information is converted into a standardized feature representation suitable for model calculation and mapped to form an input vector.

[0052] After generating the input vector, the input vector is fed into the advertising performance prediction model. This means using a pre-trained prediction model to infer the performance of candidate ads in the current search scenario and output the estimated click-through rate and estimated conversion rate of the candidate ads. The estimated click-through rate represents the probability that a user clicks on the ad, and the estimated conversion rate represents the probability that a user completes the target behavior after clicking on the ad. The expected revenue value, which reflects the overall revenue capability of the ad, is calculated by multiplying the estimated click-through rate, the estimated conversion rate, and the revenue per ad.

[0053] The advertising performance prediction model is constructed by collecting historical search ad display logs. It obtains search ad display data recorded in the historical time period from the ad delivery system. The historical search ad display logs are used to fully record the display process of ads on the search results page and user interaction. The model extracts historical display multi-dimensional features that can characterize the ad delivery environment, ad content characteristics and user behavior feedback from the display logs. At the same time, it obtains the click-through rate and conversion rate corresponding to the historical display multi-dimensional features to reflect the actual performance of the ads in the historical delivery.

[0054] After obtaining the historical display multidimensional features and the corresponding click-through rate and conversion rate, building a deep neural network refers to constructing a multi-layer neural network structure to model the mapping relationship between multidimensional features and advertising performance. The historical display multidimensional features are used as the model input, and the click-through rate and conversion rate are used as the supervision targets. The network parameters are iteratively updated and optimized through the backpropagation algorithm until the model output meets the preset convergence conditions, thereby obtaining an advertising performance prediction model that can predict advertising performance.

[0055] After obtaining the expected revenue value, the candidate ads are weighted and calculated based on the estimated click-through rate, estimated conversion rate and expected revenue value. This means that the click potential, conversion potential and revenue level of the ads are comprehensively considered according to the preset weight ratio, and the various indicators are integrated and evaluated to generate a placement value score that represents the priority and value of ad placement.

[0056] Ads are delivered to the candidate set of ads based on the ad delivery value score.

[0057] Furthermore, this application also includes: obtaining the placement value score of each candidate advertisement in the candidate advertisement set; sorting the candidate advertisements from high to low according to the placement value score to generate an advertisement sequence; generating an advertisement placement decision instruction according to the advertisement sequence and the layout configuration of the search results page; the advertisement placement decision instruction is used to determine the number of target advertisements to be displayed on the search results page, the display position of each target advertisement, and the display order; and executing the advertisement placement according to the advertisement placement decision instruction.

[0058] Specifically, obtaining the placement value score of each candidate advertisement in the candidate advertisement set refers to reading the scoring indicators used to characterize the placement priority and comprehensive value of candidate advertisements from the advertising effect evaluation results, and sorting the candidate advertisements according to the placement value score, wherein the candidate advertisements are arranged in descending order of score, thereby generating an advertisement sequence that reflects the relative placement priority of each candidate advertisement.

[0059] After generating the ad sequence, ad delivery decision instructions are generated based on the ad sequence and the layout configuration of the search results page. This means that the ad sequence is adapted by combining the structure of the search results page, the number of ad display areas, and the page display rules, so as to generate ad delivery decision instructions to guide ad display. The ad delivery decision instructions are used to determine the number of target ads to be displayed on the search results page, the display position of each target ad, and the display order of each target ad on the page.

[0060] After generating the ad delivery decision instruction, executing the ad delivery according to the ad delivery decision instruction means loading and displaying the corresponding target ad in the specified position on the search results page according to the content of the delivery decision instruction, and completing the ad presentation process in a determined display order, thereby realizing the actual delivery of the ad in the search scenario.

[0061] In summary, the AI-enhanced search-based intelligent advertising management method provided in this application has the following technical effects: by achieving the technical goal of accurate matching and delivery management of advertisements based on the semantic enhancement representation of search requests, it achieves the technical effects of improving the semantic consistency between advertisements and users' real needs, improving the relevance of advertisement delivery, and enhancing the overall delivery effect in search scenarios.

[0062] Example 2: Based on the same inventive concept as the AI-enhanced search-based intelligent advertising management method in the foregoing examples, this application also provides an AI-enhanced search-based intelligent advertising management system. Please refer to the appendix. Figure 2 The system includes: a search vector acquisition module 1, used to acquire a search request and collect multi-dimensional information, and perform vector transformation on the search request based on the multi-dimensional information to obtain a search vector; an ad candidate set acquisition module 2, used to retrieve and obtain an ad candidate set based on the search vector; a placement value score acquisition module 3, used to predict the ad performance based on the ad candidate set to obtain a placement value score; and an ad placement module 4, used to place ads on the ad candidate set based on the placement value score.

[0063] Furthermore, the AI-enhanced search-based intelligent advertising management system is also used to: receive search requests input by users, wherein the search requests include keywords, natural language, and combined syntax; collect contextual information associated with the search requests, wherein the contextual information includes at least search time, terminal type, geographical region, search entry source, and session identifier; obtain historical records corresponding to the user, wherein the historical records include historical search records, historical ad click records, conversion behavior records, and dwell time information; and use the contextual information and the historical records as multi-dimensional information.

[0064] Furthermore, the AI-enhanced search-based intelligent advertising management system is also used for: segmenting the search request into words, removing stop words, and normalizing the text to obtain preliminary text features; semantically encoding the preliminary text features to obtain initial semantic vectors at the word level; aggregating the initial semantic vectors by weighted averaging to generate global semantic vectors at the sentence level; dynamically adjusting the global semantic vectors by combining the historical records and the context information to obtain enhanced semantic vectors; and performing principal component analysis dimensionality reduction and normalization on the enhanced semantic vectors to obtain the search vector.

[0065] Furthermore, the AI-enhanced search-based intelligent advertising management system is also used to: convert the historical records into historical behavior vectors; convert the context information into context feature vectors; and weight and fuse the global semantic vector, historical behavior vector, and context feature vector based on corresponding preset weight coefficients to obtain the enhanced semantic vector.

[0066] Furthermore, the AI-enhanced search-based intelligent advertising management system is also used for: obtaining advertising data from an advertising resource library; cleaning and concatenating the advertising title, advertising description, and advertising landing page text in the advertising data to form advertising source text; calling a semantic coding network to extract content features from the advertising source text to obtain advertising source features; pooling the advertising source features and mapping them to real-number vectors of a preset dimension to obtain advertising semantic vectors; and matching the advertising semantic vectors with the search vectors to obtain a set of matching advertising candidates.

[0067] Furthermore, the AI-enhanced search-based intelligent advertising management system is also used to: calculate the vector space distance between the search vector and the advertising semantic vector to obtain a semantic similarity score; compare the semantic similarity score with a preset relevance threshold, remove advertisements with semantic similarity scores lower than the preset relevance threshold, and generate an advertising candidate set.

[0068] Furthermore, the AI-enhanced search-based intelligent advertising management system is also used for: extracting multi-dimensional feature information for each candidate advertisement in the candidate advertisement set, wherein the multi-dimensional feature information includes at least historical interaction statistical features and advertising revenue features, wherein the historical interaction statistical features include historical exposure, historical clicks, and historical conversions; discretizing, normalizing, and cross-combining the multi-dimensional feature information, mapping the processed multi-dimensional feature information into an input vector; inputting the input vector into an advertising performance prediction model, outputting the estimated click-through rate and estimated conversion rate of the candidate advertisement, calculating the expected revenue value, wherein the expected revenue value is the product of the estimated click-through rate, the estimated conversion rate, and the revenue per advertisement; and calculating the comprehensive score of the candidate advertisement based on the weighted average of the estimated click-through rate, the conversion rate, and the expected revenue value, thereby generating the placement value score.

[0069] Furthermore, the AI-enhanced search-based intelligent advertising management system is also used to: collect historical search ad display logs, obtain historical display multi-dimensional features and corresponding click-through rates and conversion rates; build a deep neural network, using the historical display multi-dimensional features as input and the click-through rate and conversion rate as supervision targets, optimize network parameters through backpropagation algorithm until convergence, and obtain the advertising effect prediction model.

[0070] Furthermore, the AI-enhanced search-based intelligent advertising management system is also used to: obtain the placement value score of each candidate advertisement in the candidate advertisement set; sort the candidate advertisements from high to low according to the placement value score to generate an advertisement sequence; generate an advertisement placement decision instruction according to the advertisement sequence and the layout configuration of the search results page; the advertisement placement decision instruction is used to determine the number of target advertisements to be displayed on the search results page, the display position of each target advertisement, and the display order; and execute the advertisement placement according to the advertisement placement decision instruction.

[0071] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The AI-enhanced search-based intelligent advertising management method and specific examples in the aforementioned Embodiment 1 are also applicable to the AI-enhanced search-based intelligent advertising management system in this embodiment. Through the foregoing detailed description of the AI-enhanced search-based intelligent advertising management method, those skilled in the art can clearly understand the AI-enhanced search-based intelligent advertising management system in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0072] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0073] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. An intelligent advertising management method based on AI-enhanced search, characterized in that, include: A search request is acquired and multi-dimensional information is collected. The search request is then transformed into a vector based on the multi-dimensional information to obtain a search vector. Based on the search vector, an ad candidate set is obtained; Based on the set of advertising candidates, advertising effectiveness is predicted to obtain a placement value score; Ads are delivered to the candidate set of ads based on the ad delivery value score.

2. The intelligent advertising management method based on AI-enhanced search as described in claim 1, characterized in that, Obtain search requests and collect multi-dimensional information, including: Receive a search request input by a user, wherein the search request includes keywords, natural language, and combined grammar; Collect contextual information associated with the search request, including at least the search time, terminal type, geographical region, search entry source, and session identifier; Obtain the user's historical records, including historical search records, historical ad click records, conversion behavior records, and dwell time information; The context information and the historical records are used as multi-dimensional information.

3. The intelligent advertising management method based on AI-enhanced search as described in claim 2, characterized in that, The search request is vectorized based on the multi-dimensional information to obtain a search vector, including: The search request is processed by word segmentation, stop word removal, and text normalization to obtain preliminary text features; Semantic encoding is performed on the preliminary text features to obtain initial semantic vectors at the word level; The initial semantic vectors are aggregated by weighted averaging to generate sentence-level global semantic vectors; By combining the historical records and the context information, the global semantic vector is dynamically adjusted to obtain an enhanced semantic vector; The enhanced semantic vector is subjected to principal component analysis for dimensionality reduction and normalization to obtain the search vector.

4. The intelligent advertising management method based on AI-enhanced search as described in claim 3, characterized in that, By combining the historical records and the contextual information, the global semantic vector is dynamically adjusted to obtain an enhanced semantic vector, including: Transform the historical records into historical behavior vectors; The context information is converted into a context feature vector; The global semantic vector, historical behavior vector, and context feature vector are weighted and fused based on corresponding preset weight coefficients to obtain the enhanced semantic vector.

5. The intelligent advertising management method based on AI-enhanced search as described in claim 1, characterized in that, Based on the search vector, an ad candidate set is obtained, including: Ad data is obtained from the ad resource library, and the ad title, ad description and ad landing page text in the ad data are cleaned and spliced ​​to form the ad source text; The semantic coding network is invoked to extract content features from the advertising source text, thereby obtaining advertising source features; The ad source features are pooled and mapped to real number vectors of a preset dimension to obtain ad semantic vectors; The search vector is matched with the ad semantic vector to obtain a set of matching ad candidates.

6. The intelligent advertising management method based on AI-enhanced search as described in claim 5, characterized in that, Based on the search vector, the ad semantic vector is matched to obtain a set of matching ad candidates, including: Calculate the vector space distance between the search vector and the ad semantic vector to obtain a semantic similarity score; The semantic similarity score is compared with a preset relevance threshold, and advertisements with semantic similarity scores lower than the preset relevance threshold are removed to generate an advertisement candidate set.

7. The intelligent advertising management method based on AI-enhanced search as described in claim 1, characterized in that, Based on the set of advertising candidates, advertising effectiveness is predicted to obtain a placement value score, including: For each candidate ad in the candidate ad set, multidimensional feature information is extracted. The multidimensional feature information includes at least historical interaction statistics and ad revenue features. The historical interaction statistics include historical exposure, historical clicks, and historical conversions. The multidimensional feature information is discretized, normalized, and cross-combined to map the processed multidimensional feature information into an input vector; The input vector is input into the advertising performance prediction model, which outputs the estimated click-through rate and estimated conversion rate of the candidate ads, and calculates the expected revenue value, which is the product of the estimated click-through rate, the estimated conversion rate and the revenue per ad. The candidate ads are weighted and calculated based on the estimated click-through rate, estimated conversion rate, and expected revenue value to generate the placement value score.

8. The intelligent advertising management method based on AI-enhanced search as described in claim 7, characterized in that, Build an advertising effectiveness prediction model, including: Collect historical search ad display logs to obtain multi-dimensional features of historical displays and corresponding click-through rates and conversion rates; A deep neural network is constructed, with the historical display multidimensional features as input and the click-through rate and conversion rate as supervision targets. The network parameters are optimized through the backpropagation algorithm until convergence, thus obtaining the advertising effect prediction model.

9. The intelligent advertising management method based on AI-enhanced search as described in claim 1, characterized in that, Based on the placement value score, ads are placed on the set of ad candidates, including: Obtain the placement value score for each candidate ad in the candidate ad set, sort the candidate ads from high to low according to the placement value score, and generate an ad sequence; Based on the ad sequence and the layout configuration of the search results page, an ad delivery decision instruction is generated. The ad delivery decision instruction is used to determine the number of target ads to be displayed on the search results page, the display position of each target ad, and the display order. The advertising is delivered according to the advertising delivery decision instruction.

10. An intelligent advertising management system based on AI-enhanced search, characterized in that, The steps for implementing the AI-enhanced search-based intelligent advertising management method according to any one of claims 1 to 9 include: The search vector acquisition module is used to obtain search requests and collect multi-dimensional information, and perform vector transformation on the search requests based on the multi-dimensional information to obtain search vectors; The ad candidate set acquisition module is used to retrieve and obtain an ad candidate set based on the search vector. The ad placement value score acquisition module is used to predict ad performance based on the set of ad candidates and obtain an ad placement value score. The advertising delivery module is used to deliver advertisements to the set of advertising candidates based on the delivery value score.

Images