Advertisement recommendation method and device based on user voiceprint clustering

Abstract

The embodiment of the invention provides an advertisement recommendation method and device based on user voiceprint clustering, and relates to the technical field of advertisement recommendation, and the method comprises the steps: obtaining voice data, and extracting acoustic features to construct a voiceprint feature model; performing voiceprint clustering on the voiceprint features through a clustering algorithm to divide user groups; according to a voiceprint clustering result, extracting common characteristics of each user group to construct a user portrait; constructing advertisement features; according to the user portraits and the advertisement features, matching a user group for each advertisement, and sorting and screening the advertisement set of each user group; the user group to which the user belongs is judged in real time based on the voiceprint characteristics of the user, and advertisements are selected from the advertisement set of the group to be recommended; according to the method, rich information contained in voice interaction of the user is fully utilized, limitation of a traditional advertisement recommendation method is broken through, and more accurate and diversified advertisement recommendation services meeting user requirements can be provided while privacy of the user is protected.

Description

Technical Field

[0001] The present application relates to the technical field of advertisement recommendation, and in particular to an advertisement recommendation method and device based on user voiceprint clustering. Background Art

[0002] With the rapid development of Internet technology, online advertising has become an important means for enterprises to conduct marketing and promotion. However, due to the diversity of user groups and the growing personalized needs, traditional advertising methods are difficult to meet the personalized needs of users, resulting in poor advertising effects. Currently, advertising recommendation technology has been widely used in various online platforms, such as search engines, e-commerce websites, social media, etc., aiming to provide users with more relevant and interesting advertising content, improve user experience and advertising effects.

[0003] Traditional advertising recommendation methods mainly rely on historical user behavior data (such as browsing history, click history, etc.) and user attribute information (such as demographic characteristics, interests, etc.) to match user profiles and advertisements. However, these methods are often limited by the availability and completeness of data, and lack in-depth mining and understanding of user personalized characteristics, resulting in unsatisfactory advertising recommendation results. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide an advertisement recommendation method and device based on user voiceprint clustering, so as to solve the problem that the existing advertisement recommendation methods are often limited by the availability and completeness of data and lack in-depth mining and understanding of user personalized characteristics, resulting in unsatisfactory advertisement recommendation effects.

[0005] In a first aspect, an embodiment of the present application provides an advertisement recommendation method based on user voiceprint clustering, comprising:

[0006] Acquire speech data and extract acoustic features of the preprocessed speech data to construct a voiceprint feature model;

[0007] The voiceprint features are clustered by clustering algorithm to divide user groups;

[0008] Based on the voiceprint clustering results, the common features of each user group are extracted to build a user portrait;

[0009] Extract features from each advertisement in the advertisement library and construct advertisement features;

[0010] Match each ad to a user group based on user portraits and ad features, and sort and filter the ad sets for each user group;

[0011] The user group to which the user belongs is determined in real time based on the user's voiceprint features, and advertisements are selected from the group's advertisement set for recommendation.

[0012] In the above implementation process, voice data is obtained, and the acoustic features of the preprocessed voice data are extracted to construct a voiceprint feature model; the voiceprint features are clustered by a clustering algorithm to divide user groups; based on the voiceprint clustering results, the common features of each user group are extracted to construct a user portrait; features are extracted for each advertisement in the advertising library to construct advertising features; based on the user portrait and advertising features, each advertisement is matched with a user group, and the advertising set of each user group is sorted and screened; the user group to which the user belongs is determined in real time based on the user's voiceprint features, and advertisements are selected from the advertising set of the group for recommendation; the rich information contained in the user's voice interaction is fully utilized, the limitations of traditional advertising recommendation methods are broken through, and more accurate, diverse, and user-friendly advertising recommendation services can be provided while protecting user privacy.

[0013] Furthermore, it also includes:

[0014] Collecting user feedback information on the recommended advertisement, wherein the feedback information includes explicit feedback and implicit feedback;

[0015] Comprehensively evaluate the recommendation effect based on feedback information and evaluation indicators;

[0016] Based on the feedback and evaluation results, the recommendation process is iteratively optimized. The optimization measures include voiceprint clustering optimization, user portrait optimization, advertising library expansion, and recommendation algorithm optimization.

[0017] In the above implementation process, through continuous iterative optimization, the performance of the recommendation process can be continuously improved to adapt to the ever-changing needs of users and maintain the advancement and competitiveness of the recommendation results.

[0018] Furthermore, the acquiring of speech data and extracting acoustic features of the preprocessed speech data to construct a voiceprint feature model includes:

[0019] Collect voice data of users in various voice interaction scenarios, including but not limited to voice search, voice assistant and voice customer service;

[0020] Preprocessing the speech data, wherein the preprocessing includes but is not limited to speech noise reduction, endpoint detection and speech segmentation;

[0021] Extracting acoustic features from the preprocessed speech data, wherein the acoustic features include but are not limited to Mel frequency cepstral coefficients, linear prediction cepstral coefficients, fundamental frequency and short-time energy;

[0022] A voiceprint feature model is constructed based on the acoustic features to obtain a voiceprint feature vector.

[0023] In the above implementation process, the collected speech data is preprocessed to improve the quality of voiceprint feature extraction; by extracting different acoustic features, the various characteristics of the speech can be comprehensively characterized; through voiceprint feature modeling, the original acoustic features can be converted into compact and effective voiceprint feature vectors, providing good feature representation for subsequent voiceprint clustering.

[0024] Furthermore, the voiceprint features are clustered by a clustering algorithm to divide the user groups, including:

[0025] Select a similarity measurement method to define the similarity measurement between voiceprint features;

[0026] According to the similarity of the voiceprint features, an adapted clustering algorithm is selected to cluster the voiceprint features so as to divide the voiceprint features into different user groups;

[0027] By calculating the clustering evaluation index, the clustering algorithm is evaluated and the optimal clustering algorithm is fed back.

[0028] In the above implementation process, selecting a suitable similarity measurement method can accurately reflect the similarity between voiceprint features and provide a reliable basis for judgment of the clustering algorithm. According to the similarity of voiceprint features, selecting a suitable clustering algorithm can accurately divide user groups and lay the foundation for subsequent user portrait construction and advertising recommendation. By calculating the clustering evaluation index, the performance of the clustering algorithm can be objectively evaluated and the optimal clustering parameters can be selected.

[0029] Furthermore, the common features of each user group are extracted according to the voiceprint clustering results to construct a user portrait, including:

[0030] According to the voiceprint clustering results, the common group features of each user group are extracted to construct a user portrait at the group level; wherein the common group features include but are not limited to demographic features, behavioral preference features, and emotional features;

[0031] Extracting personalized features of each user to characterize the personalized characteristics of each user; wherein the personalized features include but are not limited to preference strength, interest segmentation and emotional state;

[0032] The user portrait is updated and maintained in real time based on the user's latest voice interaction data, including incremental updates, regular reconstruction and attenuation mechanisms.

[0033] In the above implementation process, through group feature extraction, the common characteristics of each user group can be fully characterized and a user portrait at the group level can be constructed; through individual feature extraction, the personalized characteristics of each user can be further characterized on the basis of the group portrait, and a more refined user understanding can be achieved; through portrait updating and maintenance, it can be ensured that the user portrait is synchronized with the user's actual characteristics, thereby improving the timeliness and accuracy of advertising recommendations.

[0034] Furthermore, extracting features from each advertisement in the advertisement library to construct advertisement features includes:

[0035] Feature extraction is performed on each advertisement in the advertisement library to construct advertisement features of each advertisement; wherein the advertisement features include content features, attribute features and interaction features.

[0036] In the above implementation process, through advertisement feature extraction, the characteristics of each advertisement in the advertisement library can be fully characterized, providing a basis for subsequent advertisement and group matching.

[0037] Furthermore, matching each advertisement to a user group according to the user portrait and advertisement features, and sorting and filtering the advertisement set of each user group, includes:

[0038] Based on user portraits and ad features, calculate the similarity between each ad and each user group to obtain the matching degree between the ad and the user group; the similarity includes feature similarity, semantic similarity and effect similarity;

[0039] According to the matching degree between the advertisement and the user group, the candidate advertisement set of each group is sorted and screened; wherein the sorting includes similarity sorting, and the screening includes diversity screening and business rule screening.

[0040] In the above implementation process, by comprehensively considering the similarities of multiple dimensions, the relevance and matching between advertisements and groups can be comprehensively evaluated; through the sorting and screening of advertisements, a high-quality candidate advertisement set can be generated for each user group as the basis for subsequent personalized recommendations.

[0041] Furthermore, the method of determining in real time based on the user's voiceprint feature the user group to which the user belongs, and selecting advertisements from the advertisement set of the group for recommendation, includes:

[0042] When a user interacts with a voice, the user group to which the user belongs is determined in real time based on the user's voiceprint features, and the most suitable advertisement is selected from the candidate advertisement set of the group for recommendation; the recommended advertisement includes an explanation of the advertisement;

[0043] Receive user feedback and optimize advertising recommendation strategies;

[0044] Among them, the optimization recommendation strategies include:

[0045] Randomly divide users into several groups, apply different recommendation strategies to different groups, and select the optimal strategy combination by comparing the advertising effects of each group;

[0046] The advertising recommendation problem is modeled as a sequential decision-making process, and the recommendation strategy is dynamically optimized through the reinforcement learning algorithm to maximize the cumulative benefits;

[0047] Through transfer learning, knowledge is transferred to the current voice scenario to accelerate the optimization and convergence of recommendation strategies.

[0048] Determine in real time the user group to which the user belongs, and select advertisements from the advertisement set of the group for recommendation.

[0049] In the above implementation process, through real-time recommendation, personalized advertisements can be naturally inserted during the user's voice interaction, thereby improving user acceptance and advertising effectiveness; through the explanation of recommended advertisements, the interpretability and persuasiveness of recommendation results can be enhanced, thereby improving user acceptance and willingness to interact with recommended advertisements; by continuously adjusting and improving recommendation strategies based on user feedback behavior, the effectiveness of advertising recommendations and user experience can be continuously improved, thereby realizing the self-evolution and improvement of the recommendation system.

[0050] In a second aspect, an embodiment of the present application provides an advertisement recommendation device based on user voiceprint clustering, comprising:

[0051] A voice acquisition module is used to acquire voice data and extract acoustic features of the pre-processed voice data to build a voiceprint feature model;

[0052] A user segmentation module is used to cluster voiceprint features using a clustering algorithm to divide user groups;

[0053] The portrait construction module is used to extract the common features of each user group based on the voiceprint clustering results to construct a user portrait;

[0054] A feature construction module is used to extract features from each advertisement in the advertisement library and construct advertisement features;

[0055] Advertisement matching module, used to match each advertisement to a user group based on user portraits and advertisement features, and to sort and filter advertisement sets for each user group;

[0056] The advertisement recommendation module is used to determine the user group to which the user belongs in real time based on the user's voiceprint characteristics, and select advertisements from the advertisement set of the group for recommendation.

[0057] In a third aspect, an embodiment of the present application provides an advertisement recommendation system based on user voiceprint clustering, comprising: an advertisement recommendation device based on user voiceprint clustering and a lock mechanism as described above, wherein the advertisement recommendation device based on user voiceprint clustering is connected to the lock mechanism.

[0058] In a fourth aspect, an embodiment of the present application provides an electronic device, including:

[0059] A processor, a memory and a bus, wherein the processor is connected to the memory via the bus, and the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, they are used to implement the advertisement recommendation method based on user voiceprint clustering as described above.

[0060] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a server, the advertisement recommendation method based on user voiceprint clustering as described above is implemented.

[0061] In a sixth aspect, an embodiment of the present application provides a computer program product, which includes instructions, and when the instructions are executed by a computer, the computer implements the above-mentioned advertisement recommendation method based on user voiceprint clustering. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0063] Figure 1 A flowchart of an advertisement recommendation method based on user voiceprint clustering provided in an embodiment of the present application;

[0064] Figure 2 It is a structural schematic diagram of an advertisement recommendation device based on user voiceprint clustering provided in an embodiment of the present application;

[0065] Figure 3 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0066] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0067] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0068] The existing advertising recommendation methods mainly include the following:

[0069] 1. Collaborative filtering recommendation method based on user browsing history and click behavior.

[0070] This method analyzes the user's historical browsing and clicking records to find other users with similar interests to the target user, and recommends advertisements to the target user based on the preferences of these users. In specific implementations, commonly used algorithms include user-based collaborative filtering and item-based collaborative filtering. The former calculates the similarity between users to find the K users most similar to the target user, and recommends advertisements to the target user based on the historical behaviors of these users; the latter calculates the similarity between advertisements to recommend advertisements that are most similar to the advertisements that the target user has browsed or clicked in the past.

[0071] 2. Advertising recommendation method based on user attribute characteristics.

[0072] This method collects user demographic information (such as age, gender, region, etc.) and attribute characteristics such as interests and hobbies, divides users into different groups based on these characteristics, and recommends suitable advertisements for each group. Common implementation methods include rule-based recommendations and machine learning-based recommendations. Rule-based recommendations match advertisements with user groups according to pre-set rules (such as age group, gender, etc.); machine learning-based recommendations automatically divide users into different groups by training classifiers or clustering models, and recommend corresponding advertisements for each group.

[0073] 3. Advertising recommendation method based on contextual environment.

[0074] This method dynamically generates relevant advertising recommendations based on the user's current environment (such as time, location, device, etc.) and the user's immediate needs. For example, it recommends nearby merchant ads based on the user's geographical location, and recommends related product ads based on the user's current search query. This method usually requires real-time acquisition and analysis of the user's environmental information and behavior data, and uses a rule engine or machine learning model for real-time advertising matching and recommendation.

[0075] 4. Advertising recommendation method based on deep learning.

[0076] In recent years, with the development of deep learning technology, some studies have begun to try to apply deep learning to the field of advertising recommendation. Compared with traditional recommendation methods, deep learning can automatically learn and extract complex features and patterns in user behavior data, and build a more accurate user interest model. Commonly used deep learning models include convolutional neural networks (CNN), recurrent neural networks (RNN), attention mechanisms, etc., which can be used to learn users' historical behavior sequences, capture users' long-term and short-term interest preferences, and realize personalized advertising recommendations.

[0077] 5. Advertising recommendation method based on cross-domain recommendation.

[0078] This method uses user behavior data from multiple domains (such as different e-commerce platforms, social networks, etc.) to achieve cross-platform advertising recommendations through cross-domain learning and transfer learning. Specifically, by constructing a cross-domain user embedding representation, users in different domains can be mapped to the same semantic space, capturing the common features and preferences of users in different domains, thereby achieving cross-domain advertising recommendations. This method can effectively solve the problem of single domain data sparsity and improve the coverage and accuracy of recommendations.

[0079] However, existing advertising recommendation methods have the following problems:

[0080] 1. The contradiction between data acquisition and privacy protection.

[0081] Existing advertising recommendation methods mainly rely on users' historical behavior data (such as browsing history, click history, etc.) and explicit attribute information (such as demographic characteristics, interests, etc.). However, with the increasing awareness of user privacy protection and the introduction of relevant laws and regulations, the acquisition and use of these data have been increasingly restricted. The user's increased control over personal data has led to a decrease in the amount of data available, and the problem of data sparsity has become more prominent. In addition, users are worried about the excessive collection and use of their personal information, and may take measures such as blocking ads and clearing browsing history, further reducing the availability of data. The incompleteness and missingness of data directly affect the effectiveness of advertising recommendations.

[0082] 2. Limitations of user interest preference mining.

[0083] Existing methods mainly focus on users' explicit behaviors and attributes, such as clicks, browsing, and purchases, while ignoring the implicit preferences and interests shown by users during voice interaction. Voice interaction data contains rich user feature information, such as voice emotion, tone, and semantics, which can reflect deep-level features such as users' emotional state, cognitive level, and decision-making style. However, existing methods have not fully mined and utilized this information, resulting in limitations in the construction of user portraits and the learning of interest models, making it difficult to fully and deeply characterize users' personalized characteristics.

[0084] 3. Cold start and long tail problems.

[0085] For new users or low-activity users, existing methods often face the cold start problem, that is, due to the lack of sufficient historical behavior data, it is impossible to accurately infer the user's interest preferences, resulting in poor recommendation quality. At the same time, for some niche or non-popular advertising projects, due to less exposure and interaction, it is difficult for existing methods to obtain sufficient training data, which affects the recommendation effect for long-tail users and advertisements. How to quickly build user portraits and provide personalized recommendations when data is sparse is an urgent problem to be solved.

[0086] 4. The recommendation results lack diversity and novelty.

[0087] Existing methods tend to recommend ads that are liked by most users or similar to users' historical interests, which leads to homogenization and convergence of recommendation results. Users are easily trapped in "information cocoons", which limits their opportunities to contact new things and expand their interests. How to improve the diversity and novelty of recommendation results while ensuring the accuracy of recommendations, and provide users with a richer and more interesting advertising experience, is a problem worth exploring.

[0088] 5. Challenges of cross-domain and cross-platform recommendations.

[0089] Users may show different interests and behaviors in different platforms and domains. Existing methods mainly make recommendations based on data from a single domain, which makes it difficult to fully explore and utilize users' behavior patterns and preferences in multiple domains. How to achieve cross-domain and cross-platform information fusion and knowledge transfer, build a full-domain user portrait, and provide a consistent and personalized advertising recommendation experience is an urgent problem to be solved.

[0090] In summary, existing advertising recommendation methods have shortcomings in data acquisition, user interest mining, cold start and long tail problems, recommendation diversity, cross-domain recommendation, etc., and innovative solutions are urgently needed.

[0091] Based on this, an embodiment of the present application proposes an advertisement recommendation method based on user voiceprint clustering to solve the above problem.

[0092] Please see Figure 1 , Figure 1 A flowchart of an advertisement recommendation method based on user voiceprint clustering provided in an embodiment of the present application. The advertisement recommendation method based on user voiceprint clustering includes:

[0093] 100. Acquire speech data and extract acoustic features of the preprocessed speech data to construct a voiceprint feature model.

[0094] 110. Collect voice data of various voice interaction scenarios of users, wherein the voice interaction scenarios include but are not limited to voice search, voice assistant, and voice customer service.

[0095] 120. Preprocess the speech data, wherein the preprocessing includes but is not limited to speech noise reduction, endpoint detection, and speech segmentation.

[0096] Among them, the collected voice data is preprocessed to improve the quality of voiceprint feature extraction.

[0097] 130. Extract acoustic features from the preprocessed speech data, wherein the acoustic features include but are not limited to Mel-frequency cepstral coefficients, linear prediction cepstral coefficients, fundamental frequency, and short-time energy.

[0098] Optional, Mel frequency cepstral coefficient (MFCC): reflects the spectral characteristics of speech, consistent with the auditory characteristics of the human ear. Linear prediction cepstral coefficient (LPCC): reflects the formant characteristics of speech, characterizing the characteristics of the vocal tract. Fundamental frequency (Pitch): reflects the height of the sound, related to the pitch. Short-term energy (Energy): reflects the loudness of the sound, related to the volume.

[0099] Among them, by extracting the above-mentioned acoustic features, the timbre, pitch, loudness and other characteristics of the speech can be fully characterized.

[0100] 140. Construct a voiceprint feature model based on the acoustic features to obtain a voiceprint feature vector.

[0101] Specifically, based on the extraction of acoustic features, a voiceprint feature model is constructed. Optionally, the voiceprint feature modeling method includes: Gaussian mixture model (GMM): the acoustic feature vector is modeled as a weighted mixture of multiple Gaussian distributions, and each Gaussian component corresponds to a voiceprint feature. Deep neural network (DNN): the feature learning ability of DNN is used to automatically extract and combine acoustic features to construct a high-level voiceprint feature representation. i-vector (Identity Vector) is a feature representation method for speech and language processing, which is used to map speech to a low-dimensional voiceprint feature space. The i-vector of each speech represents its coordinates in the space. Ecapa_Tdnn is a speech recognition model based on time delay neural network (TDNN) and extended context-aware parallel attention (ECAPA), which includes an attention mechanism for dynamically weighting features at different time steps to better capture key speaker information.

[0102] Therefore, through voiceprint feature modeling, the original acoustic features can be converted into compact and effective voiceprint feature vectors, providing good feature representation for subsequent voiceprint clustering.

[0103] 200. Voiceprint features are clustered by a clustering algorithm to divide user groups.

[0104] 210. Select a similarity measurement method to define a similarity measurement between voiceprint features.

[0105] It is understandable that the purpose of voiceprint clustering is to classify similar voiceprint features into the same category, and it is necessary to first define the similarity measurement between voiceprint features. Optionally, similarity measurement methods include: Euclidean distance: measures the spatial distance between voiceprint feature vectors. Cosine similarity: measures the cosine value of the angle between voiceprint feature vectors. Probability likelihood ratio (LLR): measures the similarity between two voiceprint feature models, such as the log-likelihood ratio of GMM.

[0106] Among them, by selecting the similarity measurement method, the similarity between voiceprint features can be accurately reflected, providing a reliable judgment basis for the clustering algorithm.

[0107] 220. According to the similarity of the voiceprint features, select an adapted clustering algorithm to cluster the voiceprint features, so as to divide the voiceprint features into different user groups.

[0108] Optionally, clustering algorithms include: K-means clustering: divide the voiceprint feature vector into K clusters, each cluster is represented by its mean vector, and the sum of squared distances from the sample within the cluster to the cluster mean is minimized by iteratively optimizing the cluster mean and the cluster assignment of the sample. Hierarchical clustering: clustering is performed in a bottom-up or top-down manner by calculating the distance matrix between the voiceprint feature vectors. Common hierarchical clustering methods include AGNES (Agglomerative Nesting) and DIANA (Divisive Analysis). Spectral clustering: the voiceprint feature vector is regarded as the node of the graph, and the similarity is regarded as the weight of the edge. Clustering is achieved by spectral segmentation of the graph. Spectral clustering can discover the non-convex structure of the data and is robust to noise and outliers. Density clustering: clustering is performed according to the density distribution of the voiceprint feature vector, such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise). Density clustering can discover clusters of any shape and is insensitive to the size and density of the cluster.

[0109] Therefore, by selecting a suitable clustering algorithm based on the characteristics and distribution of voiceprint data, it is possible to accurately divide user groups and lay the foundation for subsequent user portrait construction and advertising recommendations.

[0110] 230. The clustering algorithm is evaluated by calculating the clustering evaluation index and the optimal clustering algorithm is fed back.

[0111] Specifically, in order to evaluate the quality of clustering and optimize the clustering results, clustering evaluation indicators need to be introduced. Optional, commonly used clustering evaluation indicators include: Silhouette Coefficient: Measures the similarity between the sample and its cluster and the dissimilarity with other nearest clusters. The value is between [-1, 1]. The closer to 1, the better the clustering effect. Calinski-Harabaz Index: Measures the ratio of the distance between clusters to the distance within clusters. The larger the index, the better the clustering effect. Davies-Bouldin Index: Measures the ratio of the average intra-cluster distance to the inter-cluster distance of each cluster. The smaller the index, the better the clustering effect.

[0112] It can be understood that by calculating the clustering evaluation index, the performance of the clustering algorithm can be objectively evaluated and the optimal clustering parameters can be selected (such as the number of clusters K, that is, the optimal number of clusters); at the same time, the optimal combination of clustering parameters can be automatically searched through iterative optimization methods, such as genetic algorithms, particle swarm optimization, etc., to improve the clustering effect.

[0113] 300. Based on the voiceprint clustering results, extract the common features of each user group to build a user portrait.

[0114] 310. According to the voiceprint clustering results, the common group features of each user group are extracted to construct a user portrait at the group level; wherein the common group features include but are not limited to demographic features, behavioral preference features, and emotional features.

[0115] Specifically, demographic characteristics: Based on the voiceprint characteristics of users in the group, infer the distribution of their gender, age group and other attributes. Behavioral preference characteristics: Analyze the voice interaction behavior of users in the group, such as search keywords, conversation topics, etc., to explore their common interests and needs. Emotional characteristics: Based on the emotional characteristics of the voice of users in the group, such as tone, voice, emotional tendency, etc., characterize their overall emotional state and psychological characteristics.

[0116] Therefore, through group feature extraction, the common characteristics of each user group can be comprehensively characterized and user portraits at the group level can be constructed.

[0117] 320. Extract personalized features of each user to characterize the personalized characteristics of each user; wherein the personalized features include but are not limited to preference strength, interest segmentation, and emotional state.

[0118] Among them, on the basis of group characteristics, personalized characteristics of each user are further extracted. Optionally, personalized characteristics include: Preference strength: According to the relative position and distance of the user in the group, the preference degree of the user for the group characteristics is inferred. Preference strength can be measured by indicators such as membership and confidence. Interest segmentation: Based on the common interests of the group, according to the user's voice interaction records, the individual's interest tags and preference directions are refined. Emotional state: According to the emotional characteristics of the user's voice, the current emotional state and psychological needs of the user are judged, such as positive, depressed, excited, worried, etc.

[0119] Among them, through individual feature extraction, we can further characterize the personalized characteristics of each user on the basis of group portrait, and achieve more refined user understanding.

[0120] 330. Update and maintain the user portrait in real time based on the user's latest voice interaction data, including incremental updates, regular reconstruction and attenuation mechanisms for the user portrait.

[0121] Understandably, user portraits are dynamically changing and need to be updated and maintained in real time based on the user's latest voice interaction data. Optional, specific strategies include: Incremental update: Whenever new voice interaction data is generated, it is added to the feature extraction and update process of the user portrait to achieve incremental update of the portrait. Periodic reconstruction: At regular intervals (such as every week, every month), the accumulated voice interaction data is used to re-cluster voiceprints and build user portraits to achieve regular reconstruction of the portrait. Attenuation mechanism: Introduce a time decay factor to historical voice interaction data to reduce the weight of past data and highlight the impact of the latest data, so that the portrait can reflect the user's recent changes in interest.

[0122] Therefore, through portrait updating and maintenance, it is possible to ensure that the user portrait is synchronized with the user's actual characteristics, thereby improving the timeliness and accuracy of advertising recommendations.

[0123] 400. Extract features from each advertisement in the advertisement library to construct advertisement features.

[0124] Specifically, feature extraction is performed on each advertisement in the advertisement library to construct advertisement features of each advertisement; wherein the advertisement features include content features, attribute features, and interaction features.

[0125] Optional, content features: extract the text, keywords, themes and other content features of the advertisement to describe the semantic information of the advertisement. Attribute features: extract the attribute information of the advertisement, such as the advertisement type, delivery time, target group, etc. Interaction features: count the historical click-through rate, conversion rate and other interaction data of the advertisement to reflect the effect and popularity of the advertisement.

[0126] Therefore, through advertising feature extraction, the characteristics of each advertisement in the advertising library can be fully characterized, providing a basis for subsequent advertising and group matching.

[0127] 500. Match each advertisement to a user group based on the user portrait and advertisement features, and sort and filter the advertisement sets for each user group.

[0128] Specifically, based on user portraits and advertising features, the similarity between each advertisement and each user group is calculated to obtain the matching degree between the advertisement and the user group; wherein the similarity includes feature similarity, semantic similarity and effect similarity.

[0129] Optional, Feature similarity: Compare the similarity between the ad feature vector and the group feature vector, such as cosine similarity, Euclidean distance, etc. Semantic similarity: Calculate the semantic similarity between the ad content and the group interests using techniques such as topic models and word embeddings. Effect similarity: Estimate the click-through rate, conversion rate, and other effect indicators of the ad in the group based on the historical effect data of the ad and the interaction behavior of the group.

[0130] Among them, by comprehensively considering the similarities in multiple dimensions, the relevance and matching between advertisements and groups can be comprehensively evaluated.

[0131] Specifically, the candidate advertisement set of each group is sorted and screened according to the matching degree between the advertisement and the user group; wherein the sorting includes similarity sorting, and the screening includes diversity screening and business rule screening.

[0132] Optional, similarity sorting: Sort candidate ads by ad-group similarity from high to low, and give priority to ads with higher similarity. Diversity screening: Based on similarity sorting, introduce diversity factors to deduplicate and filter top-ranked ads to ensure the diversity and novelty of recommendation results. Business rule screening: Further screen candidate ads according to the business rules of ad delivery, such as delivery time, delivery budget, bidding strategy, etc., to meet the constraints of ad delivery.

[0133] Therefore, through advertisement sorting and screening, a high-quality candidate advertisement set can be generated for each user group as the basis for subsequent personalized recommendations.

[0134] 600. Determine in real time the user group to which the user belongs based on the user's voiceprint features, and select advertisements from the group's advertisement set for recommendation.

[0135] Specifically, when a user performs voice interaction, the user group to which the user belongs is determined in real time based on the user's voiceprint characteristics, and the advertisement with the highest degree of adaptability (most relevant, most interesting) is selected from the candidate advertisement set of the group for recommendation; wherein the recommended advertisement includes an explanation of the advertisement.

[0136] Optional, the key to real-time recommendation is: Voiceprint recognition: through voiceprint recognition technology, quickly determine the identity of the user and find the corresponding user group. Personalized sorting: based on the candidate ad set, personalized sorting of ads according to the user's individual characteristics (such as preference strength, interest segmentation, etc.), recommending ads that best match the user. Contextual adaptation: combined with the user's current voice interaction content, emotional state and other contextual information, dynamically adjust the ad recommendation strategy to improve the relevance and timeliness of the ad.

[0137] Optionally, in order to enhance users' trust and understanding of recommended ads, corresponding explanations need to be generated for each recommended ad. Explanations include: Reasons for recommendation: Explain the main reason for recommending the ad, such as "Based on your preference for food, we recommend this newly opened restaurant." Evidence of relevance: List evidence that the recommended ad is related to user characteristics, such as "You have been to the gym many times recently, and this sports bracelet can help you better manage your health." Personalized highlights: Highlight the points of convergence between the recommended ad and the user's personalized needs, such as "The style of this game is very similar to the cartoon style you like."

[0138] Therefore, through real-time recommendation, personalized advertisements can be naturally inserted into the process of user voice interaction, improving user acceptance and advertising effectiveness. At the same time, through recommendation explanations, the interpretability and persuasiveness of recommendation results can be enhanced, improving users' acceptance of recommended advertisements and their willingness to interact.

[0139] Specifically, we receive user feedback and optimize the advertising recommendation strategy. Advertising recommendation is a dynamic optimization process, which requires continuous adjustment and improvement of the recommendation strategy based on user feedback.

[0140] Optional optimization recommendation strategies include:

[0141] A / B testing: Randomly divide users into several groups, apply different recommendation strategies to different groups, and select the optimal strategy combination by comparing the advertising effects of each group.

[0142] Reinforcement learning: The advertising recommendation problem is modeled as a sequential decision-making process, and the recommendation strategy is dynamically optimized through reinforcement learning algorithms (such as multi-armed bandit, contextual bandit, etc.) to maximize the cumulative benefits.

[0143] Transfer learning: Utilize the advertising recommendation knowledge from other fields or platforms and transfer the knowledge to the current voice scenario through transfer learning to accelerate the optimization and convergence of the recommendation strategy.

[0144] Therefore, by continuously adjusting and improving the recommendation strategy based on user feedback behavior, the effect of advertising recommendations and user experience can be continuously improved, and the self-evolution and improvement of the recommendation system can be achieved.

[0145] On the basis of the above embodiments, the embodiments of the present application may also be embodied as follows:

[0146] 700. Iteratively optimize the recommendation process based on user feedback on the recommended advertisements.

[0147] 710. Collect user feedback information on the recommended advertisement, where the feedback information includes explicit feedback and implicit feedback.

[0148] Optional, explicit feedback: directly ask users about their satisfaction and relevance of recommended ads through questionnaires, ratings, etc. Implicit feedback: indirectly infer users' preferences and interests based on their interactive behaviors on recommended ads, such as clicks, browsing time, conversions, etc.

[0149] At the same time, we can also collect user feedback on the explanations of recommended ads to understand the user's degree of recognition of the recommendation reasons and their opinions and suggestions; thus, by comprehensively collecting user feedback, we can evaluate the recommendation effect in multiple dimensions and discover the advantages and disadvantages of the recommendation system.

[0150] 720. Based on the feedback information and evaluation indicators, a comprehensive evaluation of the recommendation effect is conducted.

[0151] Optional evaluation indicators include: User satisfaction: Calculate the average satisfaction score based on the user's subjective evaluation of the recommended ads. Click-through rate (CTR): Count the ratio of the number of clicks on the recommended ads to the number of impressions, measuring the user's interest in the ads. Conversion rate (CVR): Count the ratio of the number of conversions to the number of clicks on the recommended ads, measuring the conversion effect of the ads. User coverage: Count the proportion of users who have received recommended ads to the total number of users, measuring the user coverage ability of the recommendation system. Recommendation diversity: Calculate the diversity indicators of the recommendation results, such as the average dissimilarity of the recommendation list, to measure the richness of the recommendation results.

[0152] Therefore, through the comprehensive evaluation of multiple indicators, the performance of the recommendation system can be comprehensively measured, providing a basis for further optimizing the recommendation strategy.

[0153] 730. Based on the feedback and evaluation results, the recommendation process is iteratively optimized. The optimization measures include voiceprint clustering optimization, user portrait optimization, advertising library expansion and recommendation algorithm optimization.

[0154] Optional optimization measures include: Voiceprint clustering optimization: Based on user feedback, optimize the voiceprint clustering algorithm and parameters to improve the accuracy and stability of user group division. User portrait optimization: Based on user feedback, improve the user portrait construction method, incorporate more feature dimensions, and improve the comprehensiveness and depth of user understanding. Advertising library expansion: Based on user feedback, supplement and expand the advertising library, introduce more diversified and personalized advertising resources, and meet the diverse needs of users. Recommendation algorithm optimization: Based on the evaluation of recommendation effects, optimize the recommendation algorithm and strategy, such as improving the similarity calculation method, introducing new sorting strategies, etc., to improve the accuracy and diversity of recommendations.

[0155] Therefore, through continuous iterative optimization, the performance of the recommendation system can be continuously improved to adapt to the ever-changing needs of users and maintain the advancement and competitiveness of the recommendation results.

[0156] The advertising recommendation method based on voiceprint clustering proposed in the embodiment of the present application realizes a new personalized advertising recommendation solution through a series of key technologies such as voiceprint feature extraction, voiceprint clustering, user portrait construction, advertising library matching, personalized recommendation, feedback evaluation, etc. This method makes full use of the rich information contained in user voice interaction, breaks through the limitations of traditional advertising recommendation methods, and can provide more accurate, diverse, and user-friendly advertising recommendation services while protecting user privacy. At the same time, this method has good interpretability, scalability, and self-optimization capabilities, providing new ideas and directions for the development of intelligent advertising recommendation technology.

[0157] More specifically, the embodiment of the present application uses voiceprint features as a representation of user personalized features, breaking through the reliance on explicit user attributes and behavior data, and providing a new way to build user portraits while protecting user privacy. Voiceprint features contain rich user attribute information (such as gender, age, etc.) and semantic information (such as emotion, tone, etc.), which can fully characterize the personalized features of users and make up for the shortcomings of existing methods in user interest mining.

[0158] Through voiceprint clustering, the embodiment of the present application can automatically classify similar users into the same group. Even for new users or users with sparse data, they can be quickly matched to the corresponding group based on their voiceprint characteristics, effectively solving the cold start problem. The construction of the group fully utilizes the similarity between users, so that long-tail users can obtain effective recommendations from similar users.

[0159] In the recommendation process, the embodiment of the present application first matches the user to the corresponding user group according to the user's voiceprint characteristics, and then makes personalized selections from the recommendation candidate set of the group. This group-based recommendation method can introduce a certain degree of randomness and diversity while ensuring the accuracy of the recommendation, thereby improving the novelty and richness of the recommendation results.

[0160] The voiceprint feature extraction and clustering method in the embodiment of the present application can be performed across platforms and domains. By mapping voiceprint data from different sources to the same feature space, cross-domain user identification and group division can be achieved, thereby providing consistent cross-platform recommendation services.

[0161] The embodiment of the present application introduces a feedback learning mechanism, which collects user feedback data on recommended advertisements, dynamically optimizes the voiceprint clustering model and recommendation strategy, and realizes self-improvement and enhancement of the recommendation system.

[0162] In summary, the embodiments of the present application effectively solve the challenges faced by existing advertising recommendation technologies, such as data acquisition, user interest mining, cold start and long tail problems, recommendation diversity, and cross-domain recommendations, by innovatively introducing voiceprint features and cluster analysis, and provide a new solution for personalized advertising recommendations.

[0163] The above steps are not to be performed in a strict order as described in the numbers, but should be understood as an overall solution.

[0164] In the second aspect, based on the above embodiments, Figure 2 A schematic diagram of the structure of an advertisement recommendation device based on user voiceprint clustering provided in an embodiment of the present application. Figure 2 The advertisement recommendation device based on user voiceprint clustering provided in this embodiment specifically includes: a voice acquisition module 201, a user classification module 202, a portrait construction module 203, a feature construction module 204, an advertisement matching module 205 and an advertisement recommendation module 206.

[0165] Among them, the voice acquisition module 201 is used to acquire voice data and extract the acoustic features of the pre-processed voice data to construct a voiceprint feature model; the user segmentation module 202 is used to cluster the voiceprint features through a clustering algorithm to divide the user groups; the portrait construction module 203 is used to extract the common features of each user group according to the voiceprint clustering results to construct a user portrait; the feature construction module 204 is used to extract features of each advertisement in the advertisement library and construct advertisement features; the advertisement matching module 205 is used to match each advertisement to a user group according to the user portrait and advertisement features, and sort and filter the advertisement set of each user group; the advertisement recommendation module 206 is used to determine the user group to which the user belongs in real time based on the user's voiceprint features, and select advertisements from the advertisement set of the group for recommendation.

[0166] As described above, the embodiment of the present application obtains voice data and extracts the acoustic features of the preprocessed voice data to construct a voiceprint feature model; voiceprint features are clustered by a clustering algorithm to divide user groups; based on the voiceprint clustering results, common features of each user group are extracted to construct a user portrait; feature extraction is performed on each advertisement in the advertisement library to construct advertisement features; based on the user portrait and advertisement features, each advertisement is matched with a user group, and the advertisement set of each user group is sorted and screened; the user group to which the user belongs is determined in real time based on the user's voiceprint features, and advertisements are selected from the advertisement set of the group for recommendation; the rich information contained in the user's voice interaction is fully utilized, the limitations of traditional advertisement recommendation methods are broken, and more accurate, diverse, and user-friendly advertisement recommendation services can be provided while protecting user privacy.

[0167] The advertisement recommendation device based on user voiceprint clustering provided in the embodiment of the present application can be used to execute the advertisement recommendation method based on user voiceprint clustering provided in the above embodiment, and has corresponding functions and beneficial effects.

[0168] In a third aspect, an embodiment of the present application further provides an electronic device, which can integrate the advertisement recommendation device based on user voiceprint clustering provided in an embodiment of the present application. Figure 3 Schematic diagram of the structure of an electronic device provided by an embodiment of the present application. Figure 3 The electronic device includes: an input device 33, an output device 34, a memory 32 and one or more processors 31; the memory 32 is used to store one or more programs; when the one or more programs are executed by the one or more processors 31, the one or more processors 31 implement the advertisement recommendation method based on user voiceprint clustering as provided in the above embodiment. The input device 33, the output device 34, the memory 32 and the processor 31 can be connected by a bus or other means. Figure 3 The example of connecting through bus is taken in the following.

[0169] The processor 31 executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory 32, that is, realizes the above-mentioned advertisement recommendation method based on user voiceprint clustering.

[0170] The electronic device provided above can be used to execute the advertisement recommendation method based on user voiceprint clustering provided in the above embodiment, and has corresponding functions and beneficial effects.

[0171] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, the computer-readable storage medium including a stored computer program; wherein, when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the advertising recommendation method based on user voiceprint clustering as described above, and can achieve the same beneficial effects.

[0172] Of course, the storage medium containing computer executable instructions provided in an embodiment of the present application is not limited to the advertising recommendation method based on user voiceprint clustering as described above, and can also execute related operations in the advertising recommendation method based on user voiceprint clustering provided in any embodiment of the present application.

[0173] In a fifth aspect, the embodiments of the present application also provide a computer program product, and the methods described in the various embodiments of the present application can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instruction is loaded and executed on a computer, the processes or functions described in the various embodiments of the present application are executed in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer grid, a grid device, a user device, a core network device, an OAM (Open Application Model) or other programmable device.

[0174] The computer program or instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer program or instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired or wireless means. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; it may also be an optical medium, such as a digital video disk; it may also be a semiconductor medium, such as a solid-state drive. The computer-readable storage medium may be a volatile or non-volatile storage medium, or may include both volatile and non-volatile types of storage media.

[0175] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0176] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0177] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for enabling an electronic device (which can be a personal computer, a server, or a grid device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0178] The above description is only an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0179] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0180] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

Claims

1. An advertisement recommendation method based on user voiceprint clustering, characterized in that: include: Acquire speech data and extract acoustic features of the preprocessed speech data to construct a voiceprint feature model; The voiceprint features are clustered by clustering algorithm to divide user groups; Based on the voiceprint clustering results, the common features of each user group are extracted to build a user portrait; Extract features from each advertisement in the advertisement library and construct advertisement features; Match each ad to a user group based on user portraits and ad features, and sort and filter the ad sets for each user group; The user group to which the user belongs is determined in real time based on the user's voiceprint features, and advertisements are selected from the group's advertisement set for recommendation.

2. The method for advertising recommendation based on user voiceprint clustering according to claim 1, characterized in that: Also includes: Collecting user feedback information on the recommended advertisement, wherein the feedback information includes explicit feedback and implicit feedback; Comprehensively evaluate the recommendation effect based on feedback information and evaluation indicators; Based on the feedback and evaluation results, the recommendation process is iteratively optimized. The optimization measures include voiceprint clustering optimization, user portrait optimization, advertising library expansion, and recommendation algorithm optimization.

3. The method for advertising recommendation based on user voiceprint clustering according to claim 1, characterized in that: The acquiring of speech data and extracting acoustic features of the pre-processed speech data to construct a voiceprint feature model includes: Collect voice data of users in various voice interaction scenarios, including but not limited to voice search, voice assistant and voice customer service; Preprocessing the speech data, wherein the preprocessing includes but is not limited to speech noise reduction, endpoint detection and speech segmentation; Extracting acoustic features from the preprocessed speech data, wherein the acoustic features include but are not limited to Mel frequency cepstral coefficients, linear prediction cepstral coefficients, fundamental frequency and short-time energy; A voiceprint feature model is constructed based on the acoustic features to obtain a voiceprint feature vector.

4. The method for advertising recommendation based on user voiceprint clustering according to claim 1, characterized in that: The voiceprint features are clustered by a clustering algorithm to divide the user groups, including: Select a similarity measurement method to define the similarity measurement between voiceprint features; According to the similarity of the voiceprint features, an adapted clustering algorithm is selected to cluster the voiceprint features so as to divide the voiceprint features into different user groups; By calculating the clustering evaluation index, the clustering algorithm is evaluated and the optimal clustering algorithm is fed back.

5. The method for advertising recommendation based on user voiceprint clustering according to claim 1, characterized in that: The method of extracting common features of each user group based on the voiceprint clustering results to construct a user portrait includes: According to the voiceprint clustering results, the common group features of each user group are extracted to construct a user portrait at the group level; wherein the common group features include but are not limited to demographic features, behavioral preference features, and emotional features; Extracting personalized features of each user to characterize the personalized characteristics of each user; wherein the personalized features include but are not limited to preference intensity, interest segmentation and emotional state; The user portrait is updated and maintained in real time based on the user's latest voice interaction data, including incremental updates, regular reconstruction and attenuation mechanisms.

6. The method for advertising recommendation based on user voiceprint clustering according to claim 1, characterized in that: The step of extracting features from each advertisement in the advertisement library and constructing advertisement features includes: Feature extraction is performed on each advertisement in the advertisement library to construct advertisement features of each advertisement; wherein the advertisement features include content features, attribute features and interaction features.

7. The method for advertising recommendation based on user voiceprint clustering according to claim 1, characterized in that: The method of matching each advertisement to a user group according to the user portrait and advertisement features, and sorting and filtering the advertisement sets of each user group, includes: Based on user portraits and ad features, calculate the similarity between each ad and each user group to obtain the matching degree between the ad and the user group; the similarity includes feature similarity, semantic similarity and effect similarity; According to the matching degree between the advertisement and the user group, the candidate advertisement set of each group is sorted and screened; wherein the sorting includes similarity sorting, and the screening includes diversity screening and business rule screening.

8. The method for advertising recommendation based on user voiceprint clustering according to claim 1, characterized in that: The method of determining in real time based on the user's voiceprint features which user group the user belongs to, and selecting advertisements from the advertisement set of the group for recommendation, includes: When a user interacts with a voice, the user group to which the user belongs is determined in real time based on the user's voiceprint features, and the most suitable advertisement is selected from the candidate advertisement set of the group for recommendation; the recommended advertisement includes an explanation of the advertisement; Receive user feedback and optimize advertising recommendation strategies; Among them, the optimization recommendation strategies include: Randomly divide users into several groups, apply different recommendation strategies to different groups, and select the optimal strategy combination by comparing the advertising effects of each group; The advertising recommendation problem is modeled as a sequential decision-making process, and the recommendation strategy is dynamically optimized through the reinforcement learning algorithm to maximize the cumulative benefits; Through transfer learning, knowledge is transferred to the current voice scenario to accelerate the optimization and convergence of recommendation strategies.

9. An advertisement recommendation device based on user voiceprint clustering, characterized in that: include: A voice acquisition module is used to acquire voice data and extract acoustic features of the pre-processed voice data to build a voiceprint feature model; A user segmentation module is used to cluster voiceprint features using a clustering algorithm to divide user groups; The portrait construction module is used to extract the common features of each user group based on the voiceprint clustering results to construct a user portrait; A feature construction module is used to extract features from each advertisement in the advertisement library and construct advertisement features; Advertisement matching module, used to match each advertisement to a user group based on user portraits and advertisement features, and to sort and filter advertisement sets for each user group; The advertisement recommendation module is used to determine the user group to which the user belongs in real time based on the user's voiceprint characteristics, and select advertisements from the advertisement set of the group for recommendation.

10. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the processor is connected to the memory via the bus, and the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, they are used to implement the advertisement recommendation method based on user voiceprint clustering as described in any one of claims 1-8.

Images