Multi-dimensional feature interaction influencer recommendation method based on account topic model

Abstract

The invention discloses a multi-dimensional feature interaction influencer recommendation method based on an account topic model, through the account topic model and multi-dimensional feature extraction, a more stereoscopic account feature system is constructed, recommendation precision is improved, and through a cross network and a deep network structure, recommendation efficiency is improved. The defects of a traditional method in feature interaction and nonlinear relation capture are overcome, the recommendation accuracy and individuation degree are improved, through fusion of a deep learning method and social interaction data, the technology enables the recommendation process to be more transparent, a brand can know recommendation reasons more clearly, meanwhile, good adaptability is achieved, and the recommendation effect is better. The method can cope with the continuous change of social media content, optimizes the order and correlation of the recommendation list through two loss functions of sorting loss and matching loss, helps a brand to find the most suitable cooperation influencer more quickly, and improves the efficiency and effect of marketing activities.

Description

Technical Field

[0001] The present invention belongs to the field of data processing technology, and specifically relates to a multi-dimensional feature interaction influencer recommendation method based on an account topic model. Background Art

[0002] With the rapid development of social media, social media marketing has gradually become a core means for companies to promote their brands, enhance user stickiness and improve market competitiveness. In this process, influencers become a bridge between brands and target users. With their expertise in specific fields, deep social media influence, and close interaction with the audience, they can help brands increase exposure and trust in a short period of time. How to accurately select the most suitable influencers to achieve efficient brand communication has become an important topic in influencer marketing.

[0003] Early influencer recommendation algorithms mainly focused on how to discover influencers with high potential from the massive social media users, usually relying on traditional features such as the number of fans, the influence of historical content, and the strength of relationships in social networks. These methods are based on statistical analysis of user basic data and can roughly screen out influencers with high communication potential. For example, by analyzing the frequency of interaction between influencers and their fans, the number of likes and comments, etc., the communication effectiveness of an influencer can be roughly judged. However, with the diversification of social media platform functions, this type of traditional feature-dependent method has gradually exposed many shortcomings. Simply relying on the number of fans and historical data, ignoring the diversity of content, emotional factors, and the fit with the brand, makes the accuracy and personalization of recommendations insufficient, and cannot meet more complex and sophisticated marketing needs.

[0004] In order to make up for these shortcomings, influencer recommendation algorithms have begun to introduce content relevance analysis. Specifically, by analyzing the theme, style and degree of match between the content published by influencers and brand needs, the accuracy of recommendations can be greatly improved. This type of recommendation method based on content relevance can deeply analyze the text content published by influencers through natural language processing (NLP) technology to determine the theme category of its content. At the same time, deep learning methods are gradually being applied to influencer recommendation systems, especially in the modeling of historical account activity data. Through deep neural networks (DNNs), more complex feature extraction can be performed on the historical behavior data of influencers. By jointly analyzing the influencer's past interaction records and published content, it is possible to automatically discover which behavior patterns are more attractive to the target user group of a specific brand. This method can provide more personalized recommendations and help brands more accurately identify the influencers who are most attractive to their target audiences. Although existing influencer marketing-related algorithms have made some progress, how to accurately select the most suitable influencer for a brand or activity among many influencers is still a challenging topic. When analyzing the content of influencer accounts, existing algorithms often fail to consider the diversity of account content, emotional factors, and the complex interactive relationship between influencers and brands. In addition, with the continuous development of social media content, new content forms, interaction methods and changes in social behavior require influencer recommendation algorithms to be constantly updated and optimized to adapt to these new needs and challenges.

[0005] One of the core issues of influencer recommendation is how to effectively represent and understand the characteristics of an account. In previous work, the recommendation methods constructed when implementing influencer recommendation tasks mostly rely on text information or image features to obtain content-related features of the account to measure account expression. However, in actual complex application environments, the influencer recommendation task is not just a simple content relevance matching, but a multi-dimensional and multi-level recommendation problem. The recommendation system not only needs to consider the relevance of the account content, but also should incorporate characteristics of multiple dimensions such as emotional factors, industry attributes, and the attributes of the account itself into the representation in order to fully reflect the characteristics and value of the account.

[0006] In the influencer recommendation task, in addition to solving how to effectively represent and understand account characteristics, another core issue is how to achieve accurate matching and efficient recommendation between brands and influencer accounts based on account characteristics. The core goal of influencer recommendation is to help brands find influencers that are highly compatible with product features, marketing goals, and audience groups. Compared with traditional recommendation tasks, the data characteristics of influencer recommendation tasks are abstract, heterogeneous, and multi-dimensional. Therefore, previous inventions or research works mainly focus on solving the first core problem in the influencer recommendation task. Given the complexity and particularity of the influencer recommendation task, further optimization of the influencer recommendation framework can be achieved by solving the second core problem, that is, how to achieve accurate matching and efficient recommendation between brands and influencer accounts. This problem requires more attention to the complex correlation and interaction between multi-dimensional features. Summary of the invention

[0007] In view of the above-mentioned deficiencies in the prior art, the present application provides a multi-dimensional feature interaction influencer recommendation method based on an account topic model.

[0008] In the first aspect, the present application proposes a method for recommending influencers based on a multi-dimensional feature interaction model, comprising the following steps:

[0009] Receiving and integrating multi-dimensional heterogeneous data of each target account through an input layer of an influencer recommendation framework, wherein the multi-dimensional heterogeneous data includes a brand dataset, an influencer dataset, and a partnership dataset;

[0010] Construct an account topic model according to the multi-dimensional heterogeneous data, and perform multi-dimensional feature extraction on each target account based on the account topic model to obtain account topic features, account image features, account industry probability distribution features, account sentiment features, and account attribute features;

[0011] The account theme features, account image features, account industry probability distribution features, account sentiment features and account attribute features are projected from the high-dimensional sparse space to the low-dimensional dense space by indexing the embedding matrix of the feature embedding layer to obtain a sparse feature embedding vector; the dense features are processed by the Z-Score normalization method and then spliced ​​with the sparse feature embedding vector, and the splicing result is used as the input of the feature interaction layer of the influencer recommendation framework;

[0012] Based on the deep network and the cross network, the nonlinear relationship between the features of the input data is simulated in the feature interaction layer and the deep information is extracted to achieve the deep interaction between the features and obtain the feature interaction result;

[0013] The matching prediction value between the brand and each influencer in the feature interaction result is calculated through a fully connected layer, the influencers are sorted according to the matching prediction value, a recommended sorting list is generated, and the sorting loss and matching loss loss functions are used to optimize the recommended sorting list to obtain the final recommendation result.

[0014] In some embodiments, the influencer recommendation framework includes: an input layer, a feature embedding layer, a feature interaction layer, and a ranking output layer;

[0015] The input layer is used to receive and integrate heterogeneous data inputs of different dimensions;

[0016] The feature embedding layer is used to divide the data into sparse and dense feature sets, and apply an embedding mechanism to map the sparse features to a low-dimensional space and concatenate them with the dense features;

[0017] The feature interaction layer is used to achieve deep interaction between brand features and influencer features through two sub-networks, a cross network and a deep network;

[0018] The ranking output layer is used to calculate the matching score of each potential influencer through the fully connected layer and generate a recommendation list.

[0019] In some embodiments, constructing an account topic model based on the multi-dimensional heterogeneous data includes:

[0020] Performing text cleaning, word segmentation, and hashtag extraction on the text data in the brand dataset, influencer dataset, and partnership dataset to generate a text set and a hashtag set;

[0021] Convert the text set and the image data in each data set into the same feature space to form corresponding text embedding and image embedding;

[0022] Calculating the importance weight and relevance weight of the hashtag set, constructing a hashtag weighting matrix, assigning differentiated weights to the image data according to the hashtag weighting matrix, obtaining a weighted image set, clustering and reducing the dimension of the weighted image set, and obtaining a representative clustering set;

[0023] The similarity between the text embedding and the image embedding is calculated by a cosine function, and an account topic set is determined according to the representative cluster set and the similarity.

[0024] In some embodiments, calculating the importance weight and the relevance weight of the hashtag set, constructing a hashtag weighting matrix, assigning differentiated weights to the image data according to the hashtag weighting matrix, and obtaining a weighted image set includes:

[0025] Define the text of a post published by an account as , the corresponding hashtag tag set is , then for each , represents the total number of times the tag appears in all posts of the entire account. We define the importance weight as , and introduce the smoothing factor To improve the robustness of the weight, its calculation is shown in formula (1):

[0026]

[0027] Define the relevance weight as ,in Represents the number of tags used in the post, and uses the pre-trained text model RoBERTa to calculate the text representation of a post and Characterization , and its calculation formula is shown in formula (2):

[0028]

[0029] Finally, based on the two normalized weights and , introduce the parameter and ,in Adjust the importance weight. The larger the value, the greater the weight of the post usage frequency. On the contrary, Adjust the relevance weight. The higher the value, the greater the relevance weight between the post text content and the tag. Next, define the hashtag weight The calculation formula is shown in formula (3):

[0030]

[0031] By adjusting and The value of adjusts the weight of the text over the image.

[0032] In some embodiments, clustering and reducing the dimension of the weighted image set to obtain a representative cluster set includes:

[0033] For each data point , calculate the average distance between it and other points in the same cluster and the average distance to the nearest other cluster , the calculation formulas are shown in (4) and (5) respectively:

[0034]

[0035]

[0036] in is included A cluster of points, where Does not include Clusters of points, It is a data point and The distance between

[0037] The calculation formula of the silhouette coefficient is shown in formula (6).

[0038]

[0039] Among them, the overall silhouette coefficient It is the average value of the silhouette coefficient of all data points. The closer the value is to 1, the better the clustering effect is. During the iterative optimization process, the model gradually approaches the optimal parameter combination to optimize the clustering effect, and finally outputs the optimal clustering parameters and their corresponding silhouette coefficients as a representative clustering set.

[0040] In some embodiments, the multi-dimensional feature extraction is performed on each target account based on the account topic model to obtain account topic features, account image features, account industry probability distribution features, account sentiment features and account attribute features, including:

[0041] Processing the image and text sets of each account according to the account theme set to obtain account theme words and theme distribution, and extracting account theme features from the content published by the account audience to obtain audience theme words and theme distribution;

[0042] Preprocess the image dataset of each account, input the preprocessed images into the model to obtain 2048-dimensional features, use global average pooling to compress the feature dimensions, calculate the average of all image feature vectors under each account, and obtain the account image features;

[0043] Clean and preprocess the text data of each account, extract emoticons, and calculate the VAD value of the account through dictionary-based sentiment feature extraction, sentiment feature analysis, sentiment classification model application and sentiment polarity adjustment to obtain the account sentiment features;

[0044] The number of followers of each account, the average number of comments and likes for each post are calculated to form the account attribute characteristics.

[0045] In some embodiments, the account theme features, account image features, account industry probability distribution features, account sentiment features and account attribute features are projected from a high-dimensional sparse space to a low-dimensional dense space by indexing the embedding matrix of the feature embedding layer to obtain a sparse feature embedding vector; the dense features are processed by a Z-Score standardization method and then spliced ​​with the sparse feature embedding vector, and the splicing result is used as the input of the feature interaction layer of the influencer recommendation framework, including:

[0046] The numerical features among the account topic features, account image features, account industry probability distribution features, account sentiment features, and account attribute features are classified as dense features, and the text features are classified as sparse features;

[0047] The goal of sparse feature embedding is to map the sparse high-dimensional topic word index into a low-dimensional dense vector space through the embedding matrix , the corresponding sparse features are projected from the high-dimensional sparse space to the low-dimensional dense space to achieve mapping, as shown in formulas (7)-(9):

[0048]

[0049]

[0050]

[0051] is the embedding vector of sparse features, and They represent the vocabulary size and embedding dimension respectively. For multi-valued features, the average of the embedding vectors of sparse features is used as the final vector;

[0052] The main goal of dense feature processing is to combine it with the embedding results of sparse features through preprocessing and conversion. The difference between each feature value and the mean of the feature is calculated and then divided by the standard deviation of the feature for standardization. The final output is the concatenation of all embedded vectors and normalized dense features as shown in formula (10):

[0053]

[0054] In some embodiments, the method of calculating the match prediction value between the brand and each influencer in the feature interaction result through a fully connected layer, sorting the influencers according to the match prediction value, generating a recommended sorting list, and optimizing the recommended sorting list using two loss functions, namely, sorting loss and matching loss, to obtain a final recommendation result includes:

[0055] Calculate the match prediction value between the brand and each influencer based on the results of feature interactions , the calculation formula is shown in formula (11):

[0056] (11)

[0057] in, is the weight vector for the log-odds, The expression of is shown in formula (12):

[0058] (12)

[0059] Through a custom multi-task loss function Learn the difference between the model's true value and its predicted value, The definition of is shown in formula (13):

[0060] (13)

[0061] in represents the matching loss, represents the sorting loss, is the set of all weights of the model, express The square of the L2 norm of and is the weighting parameter of the two losses, is the L2 regularization coefficient;

[0062] To reduce the impact of simple samples on model training, the implementation of focal loss is shown in formula (14):

[0063] (14)

[0064] in is the probability of the model predicting the positive class, which is calculated by the basic binary cross loss entropy. is the category balance weight, which is used to adjust the weights between different categories. is the focusing parameter, which controls the weight of difficult and easy samples;

[0065] The goal of the sorting task is to enable the model to optimize the relative order between positive and negative samples during prediction, so that the score of the positive sample is as high as possible than that of the negative sample. Positive and negative samples are randomly sampled according to a preset ratio, and the loss function calculation formula for each pair of positive and negative samples is defined as shown in formula (15):

[0066] (15)

[0067] The average pairwise ranking loss in a training batch is defined as:

[0068] (16)

[0069] in, Represents the loss function of each pair of positive and negative samples, represents the positive sample score, Represents the negative sample score.

[0070] In a second aspect, the present application proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0071] In a third aspect, the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0072] Beneficial effects of the present invention:

[0073] 1. More comprehensive and accurate account feature representation: Existing technologies mainly rely on text or image features, ignoring factors such as sentiment and industry attributes. This technology builds a more three-dimensional account feature system through the account topic model (SMATM) and multi-dimensional feature extraction, improving the recommendation accuracy.

[0074] 2. Effectively capture the complex interactive relationship between multi-dimensional features: This technology breaks down the interactive relationship between multi-dimensional features into two sub-tasks: matching and sorting. Through the cross network and deep network structure, it makes up for the shortcomings of traditional methods in capturing feature interactions and nonlinear relationships, and improves the accuracy and personalization of recommendations.

[0075] 3. Improve the explainability and adaptability of the recommendation system: Through the integration of deep learning methods and social interaction data, this technology makes the recommendation process more transparent, allowing brands to more clearly understand the reasons for the recommendation. At the same time, it has better adaptability and can cope with the constant changes in social media content.

[0076] 4. Optimize the sorting and screening of recommendation results: This technology optimizes the order and relevance of the recommendation list through two loss functions: sorting loss and matching loss, helping brands find the most suitable cooperative influencers more quickly and improving the efficiency and effectiveness of marketing activities. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 It is the overall flow chart of the present invention.

[0078] Figure 2 Schematic diagram of the recommendation framework for influencers.

[0079] Figure 3 Schematic diagram for cross-layer and depth-layer calculations. DETAILED DESCRIPTION

[0080] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein; on the contrary, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0081] On the one hand, this application proposes a multi-dimensional feature interaction influencer recommendation method based on an account topic model, such as Figure 1 As shown, the following steps are included:

[0082] S100: receiving and integrating multi-dimensional heterogeneous data of each target account through an input layer of an influencer recommendation framework, wherein the multi-dimensional heterogeneous data includes a brand dataset, an influencer dataset, and a partnership dataset;

[0083] In some embodiments, the influencer recommendation framework includes: an input layer, a feature embedding layer, a feature interaction layer, and a ranking output layer;

[0084] The input layer is used to receive and integrate heterogeneous data inputs of different dimensions;

[0085] The feature embedding layer is used to divide the data into sparse and dense feature sets, and apply an embedding mechanism to map the sparse features to a low-dimensional space and concatenate them with the dense features;

[0086] The feature interaction layer is used to achieve deep interaction between brand features and influencer features through two sub-networks, a cross network and a deep network;

[0087] The ranking output layer is used to calculate the matching score of each potential influencer through the fully connected layer and generate a recommendation list.

[0088] Among them, Figure 2As shown, the input layer receives and integrates heterogeneous data inputs of different dimensions. In order to improve the representation capability of multi-dimensional features of accounts, the present invention firstly innovatively proposes to use the account subject model SMATM of the account subject to mine account subject information. In addition, the present invention also uses different feature extraction methods based on multiple dimensions such as emotional dimensions, industry attributes, and audience attributes to obtain a more accurate representation of accounts.

[0089] The feature embedding layer divides the data into sparse and dense feature sets, and applies the embedding mechanism to map the sparse features into a low-dimensional space, which is then concatenated with the dense features for use by the feature interaction layer.

[0090] The feature interaction layer is the core part of the influencer recommendation framework (MFI-IR) to achieve feature interaction. It uses two sub-networks, the cross network and the deep network, to achieve deep interaction between brand features and influencer features. The cross network simulates the nonlinear relationship between features through multiple layers of interaction units, while the deep network extracts deep information of features through stacked hidden layers.

[0091] After the feature interaction layer, the model enters the ranking output stage. Through the fully connected layer, the model calculates the matching score of each potential influencer. These scores are then used to generate the recommendation list. In order to optimize the recommendation results, the model uses two loss functions: ranking loss and matching loss. Ranking loss is used to ensure that the order of the recommendation list is consistent with user preferences, while matching loss is used to measure the degree of match between the recommended influencer and brand needs.

[0092] S200: constructing an account topic model SMATM according to the multi-dimensional heterogeneous data, and performing multi-dimensional feature extraction on each target account based on the account topic model to obtain account topic features, account image features, account industry probability distribution features, account sentiment features, and account attribute features;

[0093] In some embodiments, constructing an account topic model based on the multi-dimensional heterogeneous data includes:

[0094] Performing text cleaning, word segmentation, and hashtag extraction on the text data in the brand dataset, influencer dataset, and partnership dataset to generate a text set and a hashtag set;

[0095] Convert the text set and the image data in each data set into the same feature space to form corresponding text embedding and image embedding;

[0096] Calculating the importance weight and relevance weight of the hashtag set, constructing a hashtag weighting matrix, assigning differentiated weights to the image data according to the hashtag weighting matrix, obtaining a weighted image set, clustering and reducing the dimension of the weighted image set, and obtaining a representative clustering set;

[0097] The similarity between the text embedding and the image embedding is calculated by a cosine function, and an account topic set is determined according to the representative cluster set and the similarity.

[0098] Among them, when constructing an account topic model, two key issues need to be solved: one is how to effectively extract and fuse the multimodal features of text and images to achieve accurate topic modeling; the other is how to deal with data sparsity to ensure that the extracted topics are both representative and explanatory. Therefore, the present invention proposes an account topic model SMATM, the core of which is to use feature extraction and clustering algorithms to achieve topic extraction for multimodal account data. The model input data includes text and images , data quality is the basis for topic extraction. The data preprocessing stage includes text cleaning, word segmentation and hashtag extraction to generate a text collection Hashtag collection Based on the BERTopic and CLIP models, text and images are converted to the same feature space to form corresponding embeddings and In order to improve the accuracy of clustering results and reduce data sparsity, a parameter learning module is introduced to adapt the optimal dimension reduction and clustering parameters for the learning image data of each account. At the same time, for hashtag data, we combine its importance with the relevance of the post body to construct a hashtag weighted matrix , giving different posts differentiated weights. Under the guidance of the optimal parameters, the weighted image set is clustered and dimensionally reduced to extract representative cluster sets To distinguish data categories. In the text embedding stage, the text embedding is calculated by the cosine function Embedding with image representation similarity to obtain the best description of the text in the post to the image data, which constitutes the topic set of each account .

[0099] In some embodiments, calculating the importance weight and the relevance weight of the hashtag set, constructing a hashtag weighting matrix, assigning differentiated weights to the image data according to the hashtag weighting matrix, and obtaining a weighted image set includes:

[0100] Define the text of a post published by an account as , the corresponding hashtag tag set is , then for each , represents the total number of times the tag appears in all posts of the entire account. We define the importance weight as , and introduce the smoothing factor To improve the robustness of the weight, its calculation is shown in formula (1):

[0101]

[0102] Define the relevance weight as ,in Represents the number of tags used in the post, and uses the pre-trained text model RoBERTa to calculate the text representation of a post and Characterization , and its calculation formula is shown in formula (2):

[0103]

[0104] Finally, based on the two normalized weights and , introduce the parameter and ,in Adjust the importance weight. The larger the value, the greater the weight of the post usage frequency. On the contrary, Adjust the relevance weight. The higher the value, the greater the relevance weight between the post text content and the tag. Next, define the hashtag weight The calculation formula is shown in formula (3):

[0105]

[0106] By adjusting and The value of adjusts the weight of the text over the image.

[0107] Among them, hashtag plays a key role in topic mining for account content. We believe that the higher the correlation between hashtag and text, the more prominent the topic of the post; the more content a user posts under a specific hashtag, the closer the posts under this hashtag are to the topic of the account. Based on this, a hashtag weighting module is introduced to learn the text weight of hashtags and provide weighting for image clustering.

[0108] In some embodiments, clustering and reducing the dimension of the weighted image set to obtain a representative cluster set includes:

[0109] For each data point , calculate the average distance between it and other points in the same cluster and the average distance to the nearest other cluster , the calculation formulas are shown in (4) and (5) respectively:

[0110]

[0111]

[0112] in is included A cluster of points, where Does not include Clusters of points, It is a data point and The distance between

[0113] The calculation formula of the silhouette coefficient is shown in formula (6).

[0114] (6)

[0115] Among them, the overall silhouette coefficient It is the average value of the silhouette coefficient of all data points. The closer the value is to 1, the better the clustering effect is. During the iterative optimization process, the model gradually approaches the optimal parameter combination to optimize the clustering effect, and finally outputs the optimal clustering parameters and their corresponding silhouette coefficients as a representative clustering set.

[0116] In order to alleviate the data sparsity problem faced by traditional multimodal topic models when facing account data, a clustering parameter learning module is used to evaluate different combinations of clustering parameters before clustering, and the parameters are selected and optimized. In this module, UMAP is used for data dimensionality reduction, and HDBSCAN is used for clustering analysis. Combining the empirical values ​​of UMAP and HDBSCAN parameters in related work and the results of sampling experiments on the scale of social media account data, the key parameters in these two processes, the number of neighboring nodes and the minimum distance parameter, are used to calculate the optimal clustering parameters. As well as the clustering cluster minimum size grouping and cluster density parameters ,During the clustering parameter learning process, the model gradually explores the ,influence of the parameters on the clustering results. After each clustering, the silhouette ,coefficient is calculated by the CEM-SS module.

[0117] S300: The account theme features, account image features, account industry probability distribution features, account sentiment features and account attribute features are projected from the high-dimensional sparse space to the low-dimensional dense space by indexing the embedding matrix of the feature embedding layer to obtain a sparse feature embedding vector; the dense features are processed by the Z-Score standardization method and then spliced ​​with the sparse feature embedding vector, and the splicing result is used as the input of the feature interaction layer of the influencer recommendation framework;

[0118] In some embodiments, the multi-dimensional feature extraction is performed on each target account based on the account topic model to obtain account topic features, account image features, account industry probability distribution features, account sentiment features and account attribute features, including:

[0119] Processing the image and text sets of each account according to the account theme set to obtain account theme words and theme distribution, and extracting account theme features from the content published by the account audience to obtain audience theme words and theme distribution;

[0120] Preprocess the image dataset of each account, input the preprocessed images into the model to obtain 2048-dimensional features, use global average pooling to compress the feature dimensions, calculate the average of all image feature vectors under each account, and obtain the account image features;

[0121] Clean and preprocess the text data of each account, extract emoticons, and calculate the VAD value of the account through dictionary-based sentiment feature extraction, sentiment feature analysis, sentiment classification model application and sentiment polarity adjustment to obtain the account sentiment features;

[0122] The number of followers of each account, the average number of comments and likes for each post are calculated to form the account attribute characteristics.

[0123] Among them, the core of the account topic feature extraction process is implemented by applying the SMATM framework proposed in the present invention. The input of SMATM is a set of images and texts composed of accounts in the data set. By calculating the representation embedding of the images and texts in the same space in these sets, and then performing image clustering based on the image embedding under the text weight, a representative image cluster is obtained. Subsequently, in the text content, the text with the highest similarity to the representative image representation is found to characterize the core keywords of the account content. In the account topic feature extraction link, the text and image collection of tweets posted by the audience account collected with the brand or influencer account as the center is subjected to account audience topic extraction, and the topic analysis results of the target account audience are obtained to enhance the dimension of the account feature. As the input of the final influencer recommendation framework, the topic features of the brand or influencer account and its audience account set will be presented in the form of account or audience keywords and topic distribution.

[0124] Image feature extraction is also one of the core tasks of image analysis. In the present invention, a model based on ResNeXt WSL is selected to extract account image features and predict industry features. As a derivative variant of ResNeXt, the ResNeXt WSL model is trained using a weakly supervised learning strategy, that is, it uses a large-scale Instagram image dataset that is not fully labeled to mine potential features and patterns, and fine-tunes on the fully labeled ImageNet dataset to adapt to task requirements and improve classification accuracy, so that it has both flexibility and efficiency when processing large-scale datasets. At the same time, during the training process of the ResNeXt WSL model, the concept of "cardinality" is introduced, that is, the number of parallel paths in the network. Specifically, the images in the dataset are first reintegrated into an account dataset according to the account. In the feature extraction process, the social media image dataset based on the account is first preprocessed. Each image has been standardized, specifically including adjusting the image size, cropping, and normalization. The preprocessed image is used as the input of the pre-trained model resnext101_32x8d_wsl to obtain the 2048-dimensional features of the social media image, which can more comprehensively reflect the main shape, texture and color features of the image. In order to obtain the image feature representation of the account, first use global average pooling to compress the feature dimension of each image, and then calculate the average value of all image feature vectors under each account to obtain a single vector representing the account image feature;

[0125] In the process of feature extraction, the present invention also adjusts the image feature extraction model based on the research on the industry information of brand accounts and the image classification method, uses it as a classification model, uses a brand image set with industry attributes to train the classification model, analyzes the image set of influencer accounts on the trained model, and provides the candidate influencer accounts with the industry feature distribution of account image information to assist in recommendation decisions, thereby enhancing the interpretability of recommendation decisions. Specifically, the calculation of industry features depends on the output of the image classification model, by classifying the images posted by the influencer accounts and generating a set of industry features for each account. In order to achieve this goal, the image classification model is adjusted to enable the model to predict the probability distribution of labels related to the industry. In this task, the image feature extraction model ResNeXt WSL is modified, the output layer of the model is adjusted, the number of neurons in the fully connected layer is increased to 12, and the softmax activation function is used to output the probability value of each industry feature. In the process of model fine-tuning training, the brand account data in the data set is used, and the brand category is used as the label, which is divided into a training set and a validation set. Subsequently, based on the fine-tuned model, all images of each account are classified, and the probability distribution of each photo under the account is averaged to obtain the image industry probability distribution of the influencer account.

[0126] In the influencer recommendation and matching task, the emotional dimension of the account is also crucial. The basic emotional theory in the field of affective computing believes that humans are born with a set of basic emotions. This study uses the three-dimensional emotion theory in emotional psychology, the VAD model, to quantify and analyze emotional expression and intensity. The VAD model advocates that emotions can be described by three dimensions: valence, arousal, and dominance. Specifically, emotional polarity reflects the "good or bad" or "positive and negative" tendency of emotions, usually represented by a continuous spectrum from positive to negative. Through the emotional polarity dimension, the emotional tendency of emotions can be identified: positive emotions are positive emotions, and negative emotions are negative emotions. Arousal measures the intensity of emotions or the intensity of emotional reactions. It describes the level of physiological arousal of emotional experience, from low to high. This dimension helps to understand the intensity or energy of emotional experience: high arousal represents intense emotions, and low arousal indicates calm emotions. Dominance assesses the sense of control or dominance in emotional experience. Low dominance emotions indicate passivity or control, while high dominance emotions convey a sense of control or control. High-dominance emotions are usually associated with positive self-perceptions, while low-dominance emotions are associated with passive and vulnerable feelings. In terms of algorithm implementation, the calculation of account sentiment features is not only based on text content, but also integrates sentiment dictionaries, the emotional impact of emoticons, and the prediction of sentiment classification models to form a multi-dimensional sentiment analysis indicator - VAD value. In order to effectively perform sentiment analysis, the account text data is first cleaned and preprocessed, and emoticons in posts are extracted. The core modules of sentiment feature calculation include dictionary-based sentiment feature extraction, sentiment feature analysis, sentiment classification model application, and sentiment polarity adjustment.

[0127] S400: Based on the deep network and the cross network, the nonlinear relationship between the features of the input data is simulated in the feature interaction layer and deep information is extracted to achieve deep interaction between the features and obtain a feature interaction result;

[0128] In some embodiments, the account theme features, account image features, account industry probability distribution features, account sentiment features and account attribute features are projected from a high-dimensional sparse space to a low-dimensional dense space by indexing the embedding matrix of the feature embedding layer to obtain a sparse feature embedding vector; the dense features are processed by a Z-Score standardization method and then spliced ​​with the sparse feature embedding vector, and the splicing result is used as the input of the feature interaction layer of the influencer recommendation framework, including:

[0129] The numerical features among the account topic features, account image features, account industry probability distribution features, account sentiment features, and account attribute features are classified as dense features, and the text features are classified as sparse features;

[0130] The goal of sparse feature embedding is to map the sparse high-dimensional topic word index into a low-dimensional dense vector space through the embedding matrix , the corresponding sparse features are projected from the high-dimensional sparse space to the low-dimensional dense space to achieve mapping, as shown in formulas (7)-(9):

[0131]

[0132]

[0133]

[0134] is the embedding vector of sparse features, and They represent the vocabulary size and embedding dimension respectively. For multi-valued features, the average of the embedding vectors of sparse features is used as the final vector;

[0135] Among them, the input layer takes the brand data in the dataset , influencer data And the data of the cooperation relationship between the two as the original input.

[0136] Taking social media accounts as the main body, based on the multi-dimensional characterization method of accounts, the topic characteristics of brand accounts and influencer accounts are calculated , extract image features , and calculate the industry probability distribution and sentiment dimension values .

[0137] In addition, we calculate the number of account followers, the average number of comments and likes per post to form account attribute features .

[0138] Subsequently, the brand and influencer feature dictionaries are initialized using account-related features as dictionary values. Feature pairs between brands and influencers are constructed by generating all possible combinations of brand and influencer IDs. Each pair of feature combinations will be labeled as a "positive sample" or a "negative sample" and the partnership will be recorded. During the feature pair generation process, the feature value of each brand and influencer will be assigned a unique ID and stored in the feature dataset. During the storage process, the model associates the feature value with the brand ID and influencer ID. After all features are calculated, stored, and labeled, the algorithm will generate the final feature pair set. , and use it as the input of the model for subsequent learning and optimization stages.

[0139] Next comes the feature embedding layer, which first divides the features of the input layer brands and influencers as shown in Table 1 into sparse features and dense features according to their characteristics.

[0140] Table 1

[0141] The main goal of dense feature processing is to combine it with the embedding results of sparse features through preprocessing and conversion. The difference between each feature value and the mean of the feature is calculated and then divided by the standard deviation of the feature for standardization. The final output is the concatenation of all embedded vectors and normalized dense features as shown in formula (10):

[0142]

[0143] The feature interaction layer is the key to learning effective feature interactions. In the influencer recommendation research framework proposed in this invention, in order to achieve meaningful interaction between brand features and influencer features, the feature interaction layer is based on the deep network and cross network modules in DCN_V2, and constructs the cross network and deep network modules in the MFI-IR recommendation model. Figure 2 The calculation process of a cross layer and a deep layer is shown respectively. When constructing the recommendation framework, for different data sets, the number and combination of cross layers and deep layers can be dynamically adjusted according to their data characteristics.

[0144] S500: Calculate the matching prediction value between the brand and each influencer in the feature interaction result through a fully connected layer, sort the influencers according to the matching prediction value, generate a recommended sorting list, and optimize the recommended sorting list using two loss functions, namely, sorting loss and matching loss, to obtain a final recommendation result.

[0145] In some embodiments, the method of calculating the match prediction value between the brand and each influencer in the feature interaction result through a fully connected layer, sorting the influencers according to the match prediction value, generating a recommended sorting list, and optimizing the recommended sorting list using two loss functions, namely, sorting loss and matching loss, to obtain a final recommendation result includes:

[0146] Calculate the match prediction value between the brand and each influencer based on the results of feature interactions , the calculation formula is shown in formula (11):

[0147] (11)

[0148] in, is the weight vector for the log-odds, The expression of is shown in formula (12):

[0149] (12)

[0150] In order to further optimize the recommendation results, it is crucial to design a suitable loss function. In combination with the influencer recommendation task, the model loss function should consider the effects of the sorting task and the matching task at the same time, that is, it should not only focus on the accuracy of the predicted value, but also optimize the relative order of the predicted results. Therefore, during model training, a custom multi-task loss function is used. Learn the difference between the model's true value and its predicted value, The definition of is shown in formula (13):

[0151] (13)

[0152] in represents the matching loss, represents the sorting loss, is the set of all weights of the model, express The square of the L2 norm of and is the weighting parameter of the two losses, is the L2 regularization coefficient;

[0153] The matching task can be regarded as a binary classification prediction problem. Considering that there are few positive samples in the sample data set and the ratio of positive and negative samples is extremely unbalanced, this part of the loss function uses focal loss to optimize the model matching effect. Focal loss is to reduce the impact of simple samples on model training by adjusting the basic binary classification cross entropy loss, thereby improving the model's learning ability for difficult samples. The implementation of focal loss is shown in formula (14):

[0154] (14)

[0155] in is the probability of the model predicting the positive class, which is calculated by the basic binary cross loss entropy. is the category balance weight, which is used to adjust the weights between different categories. is the focusing parameter, which controls the weight of difficult and easy samples;

[0156] The goal of the sorting task is to enable the model to optimize the relative order between positive and negative samples during prediction, so that the score of the positive sample is as high as possible than that of the negative sample. Positive and negative samples are randomly sampled according to a preset ratio, and the loss function calculation formula for each pair of positive and negative samples is defined as shown in formula (15):

[0157] (15)

[0158] The average pairwise ranking loss in a training batch is defined as:

[0159] (16)

[0160] in, Represents the loss function of each pair of positive and negative samples, represents the positive sample score, Represents the negative sample score. In the sorting task, the model needs to optimize the relative order between positive and negative samples so that the positive sample score is as high as possible over the negative sample score. The loss of this pair of samples is measured by calculating the value related to the difference between the positive and negative sample scores. Formula (15) is used. Margin represents the set interval value. When the difference between the positive sample score and the negative sample score is less than the interval, a loss will occur.

[0161] Furthermore, the experimental parameters and comparative experimental results of the recommended framework model are as follows:

[0162] 1. MFI-IR model parameters and settings:

[0163] The recommendation effect of the MFI-IR model is affected by multiple factors, the most critical of which are the structure of the five feature interaction layers (see Table 2 for details) and the key parameters of the model. Table 2 lists in detail the definition and description of each symbol and the parameter value range used in the experiment.

[0164] Table 2 Structure of the feature interaction layer of the recommendation algorithm and its related key parameters

[0165]

[0166] Based on the key recommendation algorithm experimental parameters contained in the table, the present invention uses a grid search method in the experiment to achieve parameter exploration and optimization. During the grid search process, the parameters are combined into a variety of different configurations and used for model training in turn. For each parameter configuration, cross-validation is implemented to evaluate the model performance to ensure the reliability of the optimization results. The model training process is based on a unified training set and validation set, and the model performance is comprehensively evaluated with reference to the five core indicators of AUC, CAUC, P@10, MRR, and MAP.

[0167] Taking into account the trade-off between global indicators and local ranking indicators, when the parameters are determined experimentally as shown in Table 3, the model has the best overall performance and achieves a good balance between feature expression ability and model generalization.

[0168] Table 3 Optimal parameter configuration of MFI-IR model

[0169]

[0170] 2. Baseline experiments and experimental data:

[0171] Baseline Model:

[0172] In order to verify the effectiveness of the influencer recommendation algorithm proposed in this paper, eight representative baseline models are selected for comparative analysis, covering three categories: traditional recommendation methods, multimodal learning methods, and conceptual representation methods:

[0173] (1) RAND: A random recommendation benchmark model that establishes a null hypothesis baseline by assigning random ranking scores to all brand-influencer pairs. This is used to verify the recommendation effectiveness of other methods.

[0174] (2) MIV: Influence evaluation model. The present invention introduces an activity factor module into its original architecture and enhances its adaptability to brand-influencer matching scenarios by dynamically adjusting the influence weight.

[0175] (3) MIR: A multimodal ranking learning model that constructs a brand-influencer relevance evaluation matrix by fusing heterogeneous data and uses a collaborative attention mechanism to achieve cross-modal feature alignment.

[0176] (4) HPTR: A multimodal historical pooled triplet model that combines pooled representation learning of historical interaction data with triplet loss optimization to capture the dynamic interaction patterns between brands and influencers through time series modeling.

[0177] (5) Wsim: A multi-task similarity ranking framework that uses a trainable inner product space to compute visual-textual bimodal similarity, jointly optimizes engagement prediction and ranking tasks, and achieves joint embedding of feature spaces.

[0178] (6) MOR: Multidimensional portrait ranking model, which builds a three-dimensional representation system of individual image, target audience, and cooperation preference based on the cross-modal joint attention mechanism, and improves the recommendation accuracy by designing a multi-dimensional fusion ranking function.

[0179] (7) CATR: Concept Alignment Triple Model, which adopts a semantic mapping method based on concept space and optimizes the triple loss under ontology constraints to strengthen the semantic consistency between brand needs and influencer characteristics.

[0180] (8) CAMERA: A conceptual influencer ranking framework that builds a brand-influencer concept graph and uses a graph attention network to propagate concept nodes, achieving deep semantic matching based on a conceptual hierarchy.

[0181] (9) MFI-IR (mmLDA): The influencer recommendation algorithm proposed in this paper replaces the account topic feature extraction method with mmLDA used by other existing influencer recommendation methods. Experimental results and analysis:

[0182] In order to verify the effectiveness of the MFI-IR model proposed in this paper, a comparison experiment of multiple recommendation system benchmark models was conducted, and the results are shown in Table 4. In the experiment, in order to ensure the comprehensiveness and fairness of the experiment, each model was trained and tested on the same data set, and the evaluation index system introduced above was used. In terms of model configuration, the optimal parameter configuration of the MFI-IR model is shown in Table 3 above.

[0183] Table 4 Comparison between the MFI-IR model and the baseline model

[0184]

[0185] Experiments show that the MFI-IR model proposed in this paper demonstrates significant performance improvement in multiple key evaluation indicators, and exhibits stronger discrimination ability and recommendation accuracy compared to traditional baseline models.

[0186] Specifically, in terms of accuracy indicators related to the recommendation matching task, the AUC value of MFI-IR is 0.9371, which is about 6% higher than the 0.884 of the MOR model. This improvement shows that MFI-IR can more effectively distinguish positive samples from negative samples and has stronger discrimination ability, thereby providing users with more accurate recommendation results. The CAUC value of MFI-IR is 0.7841, which is about 7.8% higher than the 0.727 of MOR. CAUC further examines the ranking of positive and negative samples in the same category based on AUC. The improvement of MFI-IR reflects that the model has a strong ranking ability in the same category and can more accurately judge the match between brands and influencers. The P@10 of MFI-IR reached 0.7191, which is significantly higher than the 0.019 and 0.005 of the MIV and RAND models, and the improvement is nearly 35 times. This shows that MFI-IR can accurately screen out relevant positive samples from the top 10 influencers in the recommendation list, greatly improving the practicality and accuracy of recommendations, especially in situations where users want to quickly obtain accurate recommendations.

[0187] In terms of the indicators related to recommendation ranking, the MAP value of MFI-IR is 0.9079, which is significantly higher than 0.238 of MOR, and the improvement is about 3.8 times, indicating that MFI-IR has a more accurate overall performance in multiple queries and can provide higher quality recommendations in the brand and influencer matching task. The MRR value of MFI-IR is 0.328, which is lower than 0.602 of CAMERA and 0.565 of MOR, but still better than other models. This shows that although MFI-IR still has room for improvement in this indicator, it can still ensure that relevant positive samples appear at a higher position in the recommendation list compared with other baseline models, thereby improving the user's click-through rate and satisfaction. In addition, the MedR value of MFI-IR is 3, which is not as high as 2 of the existing solutions, but is much lower than 54 of RAND and 49 of MIV, which means that the positive samples are in an earlier position in the recommendation list, and relevant influencers can be presented to users more quickly, further improving the efficiency of the recommendation system and user experience.

[0188] In addition, a comparison is made between the account topic extraction using the account topic model SMATM proposed in the present invention and the account topic extraction using the basic multimodal LDA model mentioned in the existing method. From the experimental results, the MFI-IR model using SMATM is superior to the MFI-IR (mmLDA) model using mmLDA in all indicators. The AUC value increased from 0.9113 to 0.9371, the CAUC value increased from 0.7631 to 0.7841, the P@10 increased from 0.6402 to 0.7191, and the MAP increased from 0.884 to 0.9079. This shows that the SMATM model proposed in the present invention can more effectively extract account topic features and provide more accurate input for the MFI-IR model, thereby further improving the overall performance of the model, making it perform better in recommendation matching and sorting.

[0189] Overall, compared with traditional recommendation system models, MFI-IR demonstrates higher recommendation efficiency by accurately optimizing the sorting of positive and negative samples and similar positive and negative samples. It can provide users with more personalized and accurate recommendation services and improve the overall quality of brand and influencer matching.

[0190] In a second aspect, the present application proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0191] In a third aspect, the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0192] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0193] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0194] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this disclosure.

[0195] In the embodiments provided in the present disclosure, it should be understood that the disclosed apparatus / computer equipment and methods can be implemented in other ways. For example, the apparatus / computer equipment embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. Multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection of the apparatus or unit, which may be electrical, mechanical or other forms.

[0196] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0197] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0198] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present disclosure implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. The computer program may include computer program code, and the computer program code may be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electric carrier signals and telecommunication signals.

[0199] The above are only preferred implementations of the present invention. It should be pointed out that a number of modifications and improved technical solutions made by those skilled in the art without departing from the technical solution should also be deemed to fall within the scope of protection required by the claims.

Claims

1. A multi-dimensional feature interactive influencer recommendation method based on an account topic model, characterized by: The following steps are involved: Receiving and integrating multi-dimensional heterogeneous data of each target account through an input layer of an influencer recommendation framework, wherein the multi-dimensional heterogeneous data includes a brand dataset, an influencer dataset, and a partnership dataset; Construct an account topic model according to the multi-dimensional heterogeneous data, and perform multi-dimensional feature extraction on each target account based on the account topic model to obtain account topic features, account image features, account industry probability distribution features, account sentiment features, and account attribute features; The account subject features, account image features, account industry probability distribution features, account sentiment features and account attribute features are projected from the high-dimensional sparse space to the low-dimensional dense space by indexing the embedding matrix of the feature embedding layer to obtain a sparse feature embedding vector; The dense features are processed by using a Z-Score normalization method and then concatenated with the sparse feature embedding vector, and the concatenated result is used as the input of the feature interaction layer of the influencer recommendation framework; Based on the deep network and the cross network, the nonlinear relationship between the features of the input data is simulated in the feature interaction layer and the deep information is extracted to achieve the deep interaction between the features and obtain the feature interaction result; The matching prediction value between the brand and each influencer in the feature interaction result is calculated through a fully connected layer, the influencers are sorted according to the matching prediction value, a recommended sorting list is generated, and the sorting loss and matching loss loss functions are used to optimize the recommended sorting list to obtain the final recommendation result.

2. The method according to claim 1, characterized in that: The influencer recommendation framework includes: an input layer, a feature embedding layer, a feature interaction layer, and a ranking output layer; The input layer is used to receive and integrate heterogeneous data inputs of different dimensions; The feature embedding layer is used to divide the data into sparse and dense feature sets, and apply an embedding mechanism to map the sparse features to a low-dimensional space and concatenate them with the dense features; The feature interaction layer is used to achieve deep interaction between brand features and influencer features through two sub-networks, a cross network and a deep network; The ranking output layer is used to calculate the matching score of each potential influencer through the fully connected layer and generate a recommendation list.

3. The method according to claim 2, characterized in that: The constructing of the account topic model according to the multi-dimensional heterogeneous data includes: Performing text cleaning, word segmentation, and hashtag extraction on the text data in the brand dataset, influencer dataset, and partnership dataset to generate a text set and a hashtag set; Convert the text set and the image data in each data set into the same feature space to form corresponding text embedding and image embedding; Calculating the importance weight and relevance weight of the hashtag set, constructing a hashtag weighting matrix, assigning differentiated weights to the image data according to the hashtag weighting matrix, obtaining a weighted image set, clustering and reducing the dimension of the weighted image set, and obtaining a representative clustering set; The similarity between the text embedding and the image embedding is calculated by a cosine function, and an account topic set is determined according to the representative cluster set and the similarity.

4. The method according to claim 3, characterized in that: The calculating the importance weight and the relevance weight of the hashtag set, constructing a hashtag weighting matrix, assigning differentiated weights to the image data according to the hashtag weighting matrix, and obtaining a weighted image set includes: Define the text of a post published by an account as , the corresponding hashtag tag set is , then for each , represents the total number of times the tag appears in all posts of the entire account. We define the importance weight as , and introduce the smoothing factor To improve the robustness of the weight, its calculation is shown in formula (1): Define the relevance weight as ,in Represents the number of tags used in the post, and uses the pre-trained text model RoBERTa to calculate the text representation of a post and hashtag Characterization , and its calculation formula is shown in formula (2): Finally, based on the two normalized weights and , introduce parameters and ,in Adjust the importance weight. The larger the value, the greater the weight of the post usage frequency. On the contrary, Adjust the relevance weight. The higher the value, the greater the relevance weight between the post text content and the tag. Next, define the hashtag weight The calculation formula is shown in formula (3): By adjusting and The value of adjusts the weight of the text over the image.

5. The method according to claim 4, characterized in that: The step of clustering and reducing the dimension of the weighted image set to obtain a representative cluster set includes: For each data point , calculate the average distance between it and other points in the same cluster and the average distance to the nearest other cluster , the calculation formulas are shown in (4) and (5) respectively: in is included A cluster of points, where Does not include Clusters of points, It is a data point and The distance between The calculation formula of the silhouette coefficient is shown in formula (6): Among them, the overall silhouette coefficient It is the average value of the silhouette coefficient of all data points. The closer the value is to 1, the better the clustering effect is. During the iterative optimization process, the model gradually approaches the optimal parameter combination to optimize the clustering effect, and finally outputs the optimal clustering parameters and their corresponding silhouette coefficients as a representative clustering set.

6. The method according to claim 5, characterized in that: The multi-dimensional feature extraction is performed on each target account based on the account topic model to obtain account topic features, account image features, account industry probability distribution features, account sentiment features and account attribute features, including: Processing the image and text sets of each account according to the account theme set to obtain account theme words and theme distribution, and extracting account theme features from the content published by the account audience to obtain audience theme words and theme distribution; Preprocess the image dataset of each account, input the preprocessed images into the model to obtain 2048-dimensional features, use global average pooling to compress the feature dimensions, calculate the average of all image feature vectors under each account, and obtain the account image features; Clean and preprocess the text data of each account, extract emoticons, and calculate the VAD value of the account through dictionary-based sentiment feature extraction, sentiment feature analysis, sentiment classification model application and sentiment polarity adjustment to obtain the account sentiment features; The number of followers of each account, the average number of comments and likes for each post are calculated to form the account attribute characteristics.

7. The method according to claim 6, characterized in that: The account subject features, account image features, account industry probability distribution features, account sentiment features and account attribute features are projected from a high-dimensional sparse space to a low-dimensional dense space by indexing the embedding matrix of the feature embedding layer to obtain a sparse feature embedding vector; The dense features are processed by using a Z-Score normalization method and then concatenated with the sparse feature embedding vector, and the concatenated result is used as the input of the feature interaction layer of the influencer recommendation framework, including: The numerical features among the account topic features, account image features, account industry probability distribution features, account sentiment features, and account attribute features are classified as dense features, and the text features are classified as sparse features; The goal of sparse feature embedding is to map the sparse high-dimensional topic word index into a low-dimensional dense vector space through the embedding matrix , the corresponding sparse features are projected from the high-dimensional sparse space to the low-dimensional dense space to achieve mapping, as shown in formulas (7)-(9): is the embedding vector of sparse features, and They represent the vocabulary size and embedding dimension respectively. For multi-valued features, the average of the embedding vectors of sparse features is used as the final vector; The main goal of dense feature processing is to combine it with the embedding results of sparse features through preprocessing and conversion. The difference between each feature value and the mean of the feature is calculated and then divided by the standard deviation of the feature for standardization. The final output is the concatenation of all embedded vectors and normalized dense features as shown in formula (10): 。 8. The method according to claim 7, characterized in that: The matching prediction value between the brand and each influencer in the feature interaction result is calculated through the fully connected layer, the influencers are sorted according to the matching prediction value, a recommended sorting list is generated, and the recommended sorting list is optimized by using two loss functions, namely, sorting loss and matching loss, to obtain the final recommendation result, including: Calculate the match prediction value between the brand and each influencer based on the results of feature interactions , the calculation formula is shown in formula (11): （11） in, is the weight vector for the log-odds, The expression of is shown in formula (12): （12） Through a custom multi-task loss function Learn the difference between the model's true value and its predicted value, The definition of is shown in formula (13): （13） in represents the matching loss, represents the sorting loss, is the set of all weights of the model, express The square of the L2 norm of and is the weighting parameter of the two losses, is the L2 regularization coefficient; To reduce the impact of simple samples on model training, the implementation of focal loss is shown in formula (14): （14） in is the probability of the model predicting the positive class, which is calculated by the basic binary cross loss entropy. is the category balance weight, which is used to adjust the weights between different categories. is the focusing parameter, which controls the weight of difficult and easy samples; The goal of the sorting task is to enable the model to optimize the relative order between positive and negative samples during prediction, so that the score of the positive sample is as high as possible than that of the negative sample. Positive and negative samples are randomly sampled according to a preset ratio, and the loss function calculation formula for each pair of positive and negative samples is defined as shown in formula (15): （15） The average pairwise ranking loss in a training batch is defined as: （16） in, Represents the loss function of each pair of positive and negative samples, represents the positive sample score, Represents the negative sample score.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images