A digital media content intelligent distribution system and method based on user profiles

Abstract

This application provides an intelligent digital media content distribution system and method based on user profiles. It acquires complete user interaction records of historical digital media content and constructs a user profile model. When new digital media content enters the distribution system, it extracts feature description vectors from the metadata of the new digital media content. These feature description vectors are then compared with explicit interest tags and implicit interest preferences in the user profile model using similarity matching and potential association analysis to obtain the user's explicit interest matching degree and potential interest association degree. Based on the explicit interest matching degree and potential interest association degree, the user's preference feedback value for the new digital media content is identified. Based on the preference feedback value, interested user groups are selected from the user database, and the new digital media content is distributed to these selected groups. The technical solution provided by this application can accurately locate interested users during the distribution of new digital media content.

Description

Technical Field

[0001] This application relates to the field of content distribution technology, and more specifically, to a digital media content intelligent distribution system and method based on user profiles. Background Technology

[0002] With the diversification of internet content formats and the continuous expansion of user scale, high-definition, high-bandwidth content such as short videos and live streaming has experienced explosive growth. Traditional content distribution faces bottlenecks such as low efficiency, high latency, and insufficient accuracy. The popularization of 5G technology provides a high-speed, low-latency network foundation, while the CDN distributed architecture effectively shortens the content transmission distance and alleviates the pressure on the origin server through global edge node caching. The integration of edge computing and AI algorithms has become a core breakthrough. At the same time, technological upgrades such as high-concurrency processing and multi-modal content adaptation further support the access of massive devices and the needs of complex scenarios, driving content distribution towards efficiency, intelligence, and personalization.

[0003] In existing content distribution, the process begins with feature extraction from massive amounts of content and user profiling. Then, complex recommendation algorithms calculate relevance scores between content and users. Finally, based on scores, business objectives, and diversity, the system personalizes and pushes content most likely to attract users to their feeds. However, in digital media content distribution, traditional methods over-rely on explicit user behavior data (such as clicks, ratings, and browsing history) for content matching. This results in repeatedly pushing content highly similar to users' historical preferences, failing to effectively uncover users' potential interests and adapt to their exploration of new content. Consequently, digital media content cannot be accurately distributed to users with specific interests. Therefore, accurately targeting users with specific interests in the new digital media content distribution landscape has become a challenge for the industry. Summary of the Invention

[0004] This application provides a digital media content intelligent distribution system and method based on user profiles, which can accurately locate interested users in the new digital media content distribution.

[0005] Firstly, this application provides a method for intelligent distribution of digital media content based on user profiles, comprising the following steps:

[0006] Obtain complete user interaction records of historical digital media content, and construct a user profile model containing explicit interest tags and implicit interest preferences based on the complete interaction records;

[0007] When new digital media content enters the distribution system, feature description vectors of the digital media content are extracted from the metadata of the new digital media content.

[0008] The feature description vector is matched with the explicit interest tags in the user profile model to obtain the explicit interest matching degree of the user. The feature description vector is also matched with the implicit interest preferences in the user profile model to obtain the potential interest correlation degree of the user.

[0009] The user's preference feedback value for the new digital media content is identified by performing dual preference feedback on the new digital media content based on the explicit interest matching degree and the potential interest correlation degree.

[0010] Based on the preference feedback values, select interested user groups from the user database and distribute new digital media content to the selected interested user groups.

[0011] In some embodiments, constructing a user profile model containing explicit interest tags and implicit interest preferences based on the complete interaction record specifically includes:

[0012] The complete interaction record is standardized to obtain structured user interaction data;

[0013] Based on the structured user interaction data, explicit interaction features of users with digital media content are extracted to generate explicit interest tags for users.

[0014] Based on the structured user interaction data, latent association features not explicitly expressed by users are mined to obtain users' implicit interest preferences;

[0015] The user's explicit interest tags and implicit interest preferences are combined to form a user profile model.

[0016] In some embodiments, when new digital media content enters the distribution system, extracting the feature description vector of the digital media content from its metadata specifically includes:

[0017] When new digital media content enters the distribution system, the metadata of the new digital media content is processed to standardize the format and obtain a standardized metadata dataset.

[0018] The standardized metadata dataset is split into textual, categorical, and numerical feature subsets;

[0019] The textual feature subset is semantically encoded using a pre-trained language model, the categorical feature subset is one-hot encoded, and the numerical feature subset is normalized to obtain the coded features of each type.

[0020] The feature description vector of digital media content is determined based on the coding features of each type.

[0021] In some embodiments, performing similarity matching between the feature description vector and the explicit interest tags in the user profile model to obtain the user's explicit interest matching degree specifically includes:

[0022] The explicit interest tags in the user profile model are processed by vector mapping to obtain explicit interest tag vectors;

[0023] Perform dimension normalization and feature space alignment on the feature description vector and explicit interest label vector to generate a set of aligned vectors of the same dimension;

[0024] The initial explicit similarity value of the user is calculated based on the same-dimensional alignment vector group;

[0025] The initial explicit similarity value is normalized to obtain the user's explicit interest matching degree.

[0026] In some embodiments, performing a potential association analysis between the feature description vector and the implicit interest preferences in the user profile model to obtain the user's potential interest association degree specifically includes:

[0027] The implicit interest preferences in the user profile model are processed by feature vector encoding to obtain the implicit interest preference vector;

[0028] The feature description vector and the latent interest preference vector are normalized in dimension and aligned with the feature space to generate a set of same-dimensional adaptive vectors.

[0029] The user's initial potential matching value is calculated based on the same-dimensional adaptation vector group;

[0030] The initial potential matching values ​​are normalized to obtain the user's potential interest correlation.

[0031] In some embodiments, filtering out interested user groups from the user database based on the preference feedback value and distributing new digital media content to the filtered interested user groups specifically includes:

[0032] The preference feedback values ​​of each user are associated and bound to the user to generate a user-distribution index mapping set;

[0033] Users whose preference feedback values ​​are greater than the feedback threshold are selected from the user-distribution index mapping set to form a candidate interest user group;

[0034] The candidate interest user groups are distributed and prioritized to generate interest user groups;

[0035] Distribute new digital media content to the aforementioned user groups with interests.

[0036] In some embodiments, a complete record of user interactions with historical digital media content is obtained through server logs.

[0037] Secondly, this application provides a digital media content intelligent distribution system based on user profiles, used to execute a digital media content intelligent distribution method based on user profiles. The system includes:

[0038] The acquisition module is used to acquire the complete interaction records of users with historical digital media content, and to construct a user profile model containing explicit interest tags and implicit interest preferences based on the complete interaction records.

[0039] The processing module is used to extract the feature description vector of the digital media content from the metadata of the new digital media content when the new digital media content enters the distribution system.

[0040] The processing module is further configured to perform similarity matching between the feature description vector and the explicit interest tags in the user profile model to obtain the explicit interest matching degree of the user, and to perform potential association analysis between the feature description vector and the implicit interest preferences in the user profile model to obtain the potential interest association degree of the user.

[0041] The processing module is also used to provide dual preference feedback on the new digital media content based on the explicit interest matching degree and the potential interest correlation degree, thereby identifying the user's preference feedback value for the new digital media content;

[0042] The execution module is used to filter out interested user groups from the user database based on the preference feedback value and distribute new digital media content to the filtered interested user groups.

[0043] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described intelligent distribution method for digital media content based on user profiles.

[0044] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described intelligent distribution method for digital media content based on user profiles.

[0045] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0046] The intelligent digital media content distribution system and method based on user profiles provided in this application first obtains complete interaction records of users with historical digital media content, and constructs a user profile model containing explicit interest tags and implicit interest preferences based on the complete interaction records. Second, when new digital media content enters the distribution system, feature description vectors of the digital media content are extracted from the metadata of the new digital media content. Further, the feature description vectors are matched with the explicit interest tags in the user profile model to obtain the user's explicit interest matching degree, and the feature description vectors are analyzed for potential correlation with the implicit interest preferences in the user profile model to obtain the user's potential interest correlation degree. Then, based on the explicit interest matching degree and the potential interest correlation degree, dual preference feedback is applied to the new digital media content to identify the user's preference feedback value for the new digital media content. Finally, based on the preference feedback value, interest user groups are selected from the user database, and the new digital media content is distributed to the selected interest user groups.

[0047] Therefore, this application demonstrates its ability to accurately target users with specific interests within the context of new digital media content distribution. Firstly, by acquiring complete user interaction records and constructing a user profile model containing explicit interest tags and implicit interest preferences, it can capture explicit user preferences through explicit interest tags and uncover unexpressed latent needs through implicit interest preferences, achieving a comprehensive and multi-dimensional portrayal of user interests and avoiding matching biases caused by single-interest dimension portrayals. Secondly, by extracting feature description vectors from the metadata of new digital media content, it transforms the diverse attributes of content into standardized and computable quantitative carriers, solving the problem of different types of content having varying attribute forms and being unable to be directly correlated, thus providing unified data support for the accurate matching of digital media content and user interests. Furthermore, by calculating the explicit interest matching degree and the latent interest correlation degree through dual dimensions, it ensures that the data... This method precisely matches digital media content with users' known preferences and uncovers potential points of interest, overcoming the limitations of single-dimensional matching and enhancing the comprehensiveness and depth of interest matching. It avoids the problem of traditional digital media content distribution relying excessively on explicit user behavior data, which leads to the repeated pushing of content highly similar to users' historical preferences without effectively uncovering their potential interests. Then, it identifies users' preference feedback values ​​based on explicit interest matching and potential interest correlation, providing an intuitive reference for accurately targeting users with interests. Finally, it filters and distributes content to interest-based user groups based on preference feedback values, accurately targeting users with interests, avoiding resource waste from indiscriminate distribution, and improving the accuracy and user acceptance rate of content distribution. In summary, the technical solution provided in this application can accurately target users with interests in the new digital media content distribution landscape. Attached Figure Description

[0048] Figure 1This is an exemplary flowchart of a digital media content intelligent distribution method based on user profiles, according to some embodiments of this application;

[0049] Figure 2 This is an exemplary flowchart illustrating the determination of a feature description vector according to some embodiments of this application;

[0050] Figure 3 This is a schematic diagram of the structure of a user profile-based intelligent digital media content distribution system according to some embodiments of this application;

[0051] Figure 4 This is a schematic diagram of the structure of a computer device that implements a user profile-based intelligent distribution method for digital media content, according to some embodiments of this application. Detailed Implementation

[0052] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0053] refer to Figure 1 This figure is an exemplary flowchart of a user profile-based intelligent distribution method for digital media content according to some embodiments of this application. The figure mainly includes the following steps:

[0054] In step S101, a complete record of the user's interaction with historical digital media content is obtained, and a user profile model containing explicit interest tags and implicit interest preferences is constructed based on the complete record of interaction.

[0055] In specific implementation, complete user interaction records of historical digital media content can be obtained through server logs. This involves capturing user clicks, scrolling, and form submissions on historical digital media using front-end tracking technologies (such as JavaScript code), obtaining in-depth interaction data (such as video playback progress and pause / play frequency) within historical digital media using a Software Development Kit (SDK), supplementing this with third-party data interfaces (such as social media application programming interfaces) to support user cross-platform behavior, and employing structured data storage (such as a four-tuple pattern of timestamp + user ID + content ID + behavior type) to obtain a complete record of user interaction with historical digital media content. In this embodiment, server logs refer to structured text data related to various accesses and operations automatically and in real-time recorded by the server during operation, serving as the data source for collecting basic user access behaviors to digital media content.

[0056] It should be noted that, in this application, a complete interaction record refers to a collection of user behavior data throughout the entire process of their interaction with historical digital media content on a digital media platform, including but not limited to access time, browsing path, content clicks, and dwell time. Obtaining a complete interaction record can provide core data for building a user profile model.

[0057] In some embodiments, constructing a user profile model containing explicit interest tags and implicit interest preferences based on the complete interaction record is achieved through the following steps:

[0058] The complete interaction record is standardized to obtain structured user interaction data;

[0059] Based on the structured user interaction data, explicit interaction features of users with digital media content are extracted to generate explicit interest tags for users.

[0060] Based on the structured user interaction data, latent association features not explicitly expressed by users are mined to obtain users' implicit interest preferences;

[0061] The user's explicit interest tags and implicit interest preferences are combined to form a user profile model.

[0062] In specific implementation, firstly, the complete interaction records generated by users on digital media platforms are standardized. This involves converting interaction data from different sources (e.g., apps, web pages) into a consistent format through a format unification operation (i.e., unifying timestamps of different formats to UTC time format and unifying content identifiers to globally unique string encodings). Then, interaction data of different magnitudes in the complete interaction records are mapped to the [0,1] interval through a linear transformation, resulting in structured user interaction data. This structured user interaction data refers to a standardized dataset after being organized according to preset data fields. Secondly, statistical analysis methods are used to quantify and statistically analyze the interaction behavior of each user in different digital media content within the structured user interaction data, in order to calculate the weighted value of the user's interaction intensity under different digital media content (i.e., for different digital media content, a set number of...). Interactive behaviors in digital media content are categorized, weighted, and their basic interaction values ​​are calculated as follows: Interactive behaviors are divided into 5 categories, with the following weights: Favorite (weight W1=0.3), Comment (weight W2=0.25), Complete View (weight W3=0.2), Repeated View (weight W4=0.15), and Click to Enter (weight W5=0.1). The sum of all weights is 1. The basic interaction value is set as follows: If the favorite behavior occurs, the basic interaction value V1=1 (if it does not occur, V1=0). For the comment behavior, V2 is calculated based on the number of characters in the comment: V2=number of characters in the comment / 100 (the upper limit is 1.5, that is, if the number of characters exceeds 150, it is still counted as 1.5; if there is no comment, V2=0). For the complete view behavior, V3=actual view time / total content time (if the ratio is ≥1, it is counted as 1; if there is no view or the view time is less than 10% of the total content time, V3=0). For the repeated view behavior, V4=number of repeated views × 0.3 (maximum value is 1, i.e., repeating 3 times or more counts as 1, no repeated browsing counts as 0), click-to-enter behavior V5=1 (one click counts as 1, multiple clicks still count as 1, no click counts as 0), interaction intensity weighted value S=V1×W1+V2×W2+V3×W3+V4×W4+V5×W5), then filter out the top N weighted values ​​(N is preset to 5-10 according to the actual application scenario, not limited here) of digital media content as the user's explicit interaction features, and then generate the user's explicit interest tags. The explicit interest tags refer to specific identifiers that can directly reflect the user's clear preferences, specifically presented in the form of keywords or phrases, such as "financial news", "food making videos", "outdoor adventure live broadcast"; then, based on the structured user interaction data, mine the potential association features that the user has not directly expressed to obtain the user's implicit interest preferences, that is: based on the structured user interaction data, construct a user-content interaction matrix, user-content interaction matrix The matrix represents users, columns represent digital media content, and matrix elements are the weighted values ​​of the interaction intensity between the corresponding user and the corresponding digital media content (0 is recorded if there is no interaction). Based on the user's interaction intensity vector (i.e., the vector composed of interaction intensity weights), the Pearson correlation coefficient formula is used to calculate the similarity between the target user and all other users. Users with similarity values ​​greater than a similarity threshold (the specific similarity threshold can be set according to actual needs and is not limited here) are selected as the target user's neighbor user group. The top N (i.e., the number consistent with the aforementioned explicit interaction characteristics) digital media content in the neighbor user group that the target user has not interacted with are collected. These digital media contents are extracted as the user's implicit interest preferences, which refer to the user's preference for digital media content that has not yet been discovered. Finally, the user's explicit interest tags and implicit interest preferences are combined to form a user profile model that can comprehensively characterize the user's interest features.

[0063] It should be noted that the user profile model in this application refers to a structured data set that integrates users' explicit and implicit preferences. The determination of the user profile model lies in transforming the scattered and unstructured interactive behaviors generated by users on the platform into structured and quantifiable interest representations, providing core data support for the accurate matching of content and users. Specifically, by integrating users' explicit interest tags and implicit interest preferences, the model can comprehensively and three-dimensionally depict users' interest characteristics. Explicit interest tags directly reflect the core preferences shown by users through explicit interactive behaviors, providing an intuitive basis for content matching. Implicit interest preferences, on the other hand, uncover the interest tendencies that users have not directly expressed but are potentially concerned about, making up for the limitations of relying solely on explicit behaviors. The combination of the two makes the interest characterization both consistent with users' known preferences and has a certain degree of scalability.

[0064] In step S102, when new digital media content enters the distribution system, the feature description vector of the digital media content is extracted from the metadata of the new digital media content.

[0065] In some embodiments, reference Figure 2 As shown, this figure is an exemplary flowchart of determining feature description vectors according to some embodiments of this application. In this embodiment, when new digital media content enters the distribution system, the extraction of feature description vectors of the digital media content from the metadata of the new digital media content can be achieved by the following steps:

[0066] In step S1021, when new digital media content enters the distribution system, the metadata of the new digital media content is processed to standardize the format to obtain a standardized metadata dataset.

[0067] In step S1022, the standardized metadata dataset is split into textual, categorical, and numerical feature subsets;

[0068] In step S1023, the text-type feature subset is semantically encoded using a pre-trained language model, the categorical feature subset is one-hot encoded, and the numerical feature subset is normalized to obtain the coded features of each type.

[0069] In step S1024, the feature description vector of the digital media content is determined based on the coding features of each type.

[0070] In specific implementation, firstly, when new digital media content enters the distribution system, the metadata of the new digital media content undergoes format standardization processing. This metadata includes, but is not limited to, various data such as content title, summary, category identifier, publication time, duration, and creator information. The metadata fields of the new digital media content are unified to a preset standardized format (i.e., text fields are uniformly encoded in UTF-8, time fields are uniformly converted to standard timestamp format, and numerical fields are uniformly represented in decimal floating-point format), thereby obtaining a standardized metadata dataset. This standardized metadata dataset refers to a set of standardized metadata for new digital media content with uniform field formats. Secondly... The standardized metadata dataset is split into textual, categorical, and numerical feature subsets using the data processing tool Python. The textual feature subset consists of descriptive textual information, such as content titles and summaries, conveying the semantic meaning of the content. The categorical feature subset consists of discrete and mutually exclusive category identifiers, such as entertainment, finance, and education category tags, and creator domain tags. The numerical feature subset consists of directly quantifiable numerical information, such as content duration, publication timestamp, number of favorites, and number of comments. Then, existing methods are applied to the textual feature subsets. A pre-trained language model (such as a Transformer-based bidirectional encoder representation model) is used for semantic encoding. First, each piece of text data in the text feature subset is split into a sequence of words according to the pre-trained language model's preset word segmentation rules. Then, this word sequence is mapped to a sequence of word vectors that the pre-trained language model can recognize and input into the pre-trained language model. The multi-layer Transformer encoder built into the pre-trained language model extracts and integrates the text semantics layer by layer, finally outputting a fixed-dimensional semantic vector as the encoding result of the text feature subset. One-hot encoding is performed on the categorized feature subset, specifically for each... All possible values ​​of a categorical feature are assigned a unique binary vector with a length equal to the total number of values ​​for that categorical feature. Only the vector position corresponding to the current value of the categorical feature is set to 1, and the rest are set to 0. This transforms the discrete categorical features into computable numerical vectors. The subset of numerical features is then processed using the min-max normalization method to obtain normalized numerical vectors, which in turn yields the coded features of each type. The coded features of each type refer to a set of text-type, categorical, and numerical features with a unified numerical format. The coded features of each type include text-type coded features, categorical coded features, and numerical feature coded features.Finally, the dimensions of each type of encoded feature are aligned. First, a unified target feature dimension is determined (i.e., the highest dimension among all types of encoded features). For encoded features with dimensions lower than the target feature dimension, zero-padding is used to pad the vector to the target feature dimension, ensuring that the dimensions of different types of encoded features are consistent. Then, the encoded features of each type are concatenated into a continuous high-dimensional numerical vector according to a preset order (e.g., text-type encoded features, categorical encoded features, and numerical encoded features), resulting in the feature description vector of the digital media content.

[0071] It should be noted that the feature description vector in this application refers to the feature vector used to describe digital media content. The determination of the feature description vector is to transform the scattered and diverse attribute information of new digital media content into a standardized numerical carrier with a unified dimension that can be directly mathematically calculated. This solves the problem that different types of content attributes (text, category, numerical) have different forms and cannot be directly used for correlation analysis. At the same time, through systematic encoding and integration, this vector fully preserves the core features of the content, ensuring the comprehensiveness and consistency of the content features. It enables it to be in the same computable feature space as the explicit interest tag vector and implicit interest preference vector in the user profile model. This provides a unified data foundation for subsequent key steps such as similarity matching and potential correlation analysis, avoiding matching deviations caused by inconsistent feature forms. Ultimately, it provides a core quantitative basis for accurately screening interest user groups and optimizing content distribution efficiency.

[0072] In step S103, the feature description vector is matched with the explicit interest tags in the user profile model to obtain the explicit interest matching degree of the user, and the feature description vector is analyzed with the implicit interest preferences in the user profile model to obtain the potential interest correlation degree of the user.

[0073] In some embodiments, the similarity matching between the feature description vector and the explicit interest tags in the user profile model to obtain the user's explicit interest matching degree is achieved through the following steps:

[0074] The explicit interest tags in the user profile model are processed by vector mapping to obtain explicit interest tag vectors;

[0075] Perform dimension normalization and feature space alignment on the feature description vector and explicit interest label vector to generate a set of aligned vectors of the same dimension;

[0076] The initial explicit similarity value of the user is calculated based on the same-dimensional alignment vector group;

[0077] The initial explicit similarity value is normalized to obtain the user's explicit interest matching degree.

[0078] In specific implementation, firstly, the explicit interest tags in the user profile model undergo vector mapping processing. These explicit interest tags are keywords or phrases that directly reflect a user's explicit preferences. During processing, a word vector tool trained on existing Chinese corpora (such as a Chinese word vector model trained based on the Word2Vec algorithm) is used to match the new digital media content corresponding to the explicit interest tags with the built-in Chinese vocabulary of the word vector tool, obtaining the basic word vectors corresponding to the explicit interest tags. If a single explicit interest tag contains multiple words, the average of the basic word vectors of these words is calculated as the explicit interest tag vector. The explicit interest tag vector refers to transforming discrete explicit interest tags into computable high-dimensional numerical vectors that can quantify the semantic information of the tags. Secondly, dimensionality normalization and feature space alignment are performed on the feature description vector and the explicit interest tag vector. Dimensional normalization uses the L2 normalization method, which calculates the sum of squares of each element of each vector and takes the square root to obtain the L2 norm of the vector. Then, each element of the vector is divided by this L2 norm to unify the vector magnitude to 1, eliminating the impact of vector length differences. The feature space alignment is achieved by verifying the feature dimensions and semantic mapping rules of the feature description vector and the explicit interest label vector. If there is a dimensional difference, the lower-dimensional vector is zero-padded (adding 0 values ​​to the end of the vector). If there is a semantic space deviation, the explicit interest label vector is mapped to the feature space of the feature description vector through linear projection, ultimately generating a set of aligned vectors of the same dimension. The set of aligned vectors of the same dimension refers to the set of feature description vectors and explicit interest label vectors with the same dimension and located in the same feature space. Then, based on the set of aligned vectors of the same dimension, cosine similarity is calculated. The method calculates the cosine similarity between two vectors in the same-dimensional aligned vector group. The calculation first involves summing the products of corresponding elements of the two vectors to obtain the vector dot product. Then, the L2 norm of each vector is calculated and multiplied. Finally, the vector dot product is divided by this product to obtain the cosine similarity value, which is used as the initial explicit similarity value. This initial explicit similarity value refers to the original matching value obtained through vector semantic association. Finally, the minimum-maximum normalization method is used to map the initial explicit similarity value to the [0,1] interval to obtain the user's explicit interest matching degree.

[0079] It should be noted that, in this application, explicit interest matching degree refers to an indicator that measures the degree to which a user's explicit interest aligns with that of new digital media content. The determination of explicit interest matching degree lies in transforming the semantic association between the feature description vector of the new digital media content and the explicit interest tags in the user profile model into a quantifiable standardized value. This solves the problem that discrete user explicit interests and diverse content attributes cannot be directly compared for matching degree. At the same time, explicit interest matching degree can accurately reflect the semantic alignment between digital media content and the user's known interests, avoiding matching bias caused by relying solely on semantic association while ignoring the priority of user interests. This matching degree can also intuitively distinguish the strength of different users' explicit interest in the same new content, providing a core foundational input for the subsequent calculation of the content distribution index.

[0080] In some embodiments, the latent interest preferences in the user profile model are analyzed for potential correlation to obtain the user's potential interest correlation degree through the following steps:

[0081] The implicit interest preferences in the user profile model are processed by feature vector encoding to obtain the implicit interest preference vector;

[0082] The feature description vector and the latent interest preference vector are normalized in dimension and aligned with the feature space to generate a set of same-dimensional adaptive vectors.

[0083] The user's initial potential matching value is calculated based on the same-dimensional adaptation vector group;

[0084] The initial potential matching values ​​are normalized to obtain the user's potential interest correlation.

[0085] In specific implementation, firstly, the implicit interest preferences in the user profile model are processed by feature vector encoding. Implicit interest preferences refer to potential interest tendencies that users do not explicitly express but are obtained through behavioral correlation mining. During encoding, the digital media content corresponding to the implicit interest preferences is encoded into feature vectors using the aforementioned vector mapping method for explicit interest tag vectors, resulting in an implicit interest preference vector. This is not elaborated further here. The implicit interest preference vector refers to transforming discrete implicit interest preferences into a computable high-dimensional numerical vector. Secondly, the aforementioned method of performing dimensionality normalization and feature space alignment on the feature description vector and explicit interest tag vector is used to normalize the dimension of the feature description vector and the implicit interest preference vector and align them in the feature space, generating a set of vectors with the same dimension. This is not elaborated further here. A same-dimensional adaptive vector group refers to a set of feature description vectors and latent interest preference vectors with the same dimension and located in the same feature space. Then, based on the same-dimensional adaptive vector group, the user's initial potential matching value is calculated. During the calculation, the joint occurrence frequency of each dimension value of the two vectors in the same-dimensional adaptive vector group is first counted to estimate the joint probability distribution of the two. Then, the occurrence frequency of each dimension value of each individual vector is counted to calculate their respective marginal probability distribution. Subsequently, the mutual information calculation formula is substituted into the log-likelihood function to calculate the degree of potential information association between the two vectors, and the initial potential matching value is output. The initial potential matching value refers to the value that represents the strength of the potential association between the feature description vector and the latent interest preference vector. Finally, the minimum-maximum normalization method is used to map the initial potential matching value to the [0,1] interval to obtain the user's potential interest association degree.

[0086] It should be noted that, in this application, the latent interest correlation refers to an indicator that characterizes the degree to which a user's latent interest aligns with that of new digital media content. The determination of the latent interest correlation lies in deeply associating the feature description vector of the new digital media content with the implicit interest preferences that are not explicitly expressed in the user profile model, transforming them into quantifiable standardized values. This effectively compensates for the potential user interest alignment points that may be missed if only explicit interest matching is relied upon, and solves the problem that user interests that are not clearly perceived but actually exist cannot be accurately captured. At the same time, this correlation complements explicit interest matching, providing a core input for the calculation of the content distribution index, enabling distribution decisions not only to match users' known preferences but also to accurately reach their latent interest areas, avoiding the homogenization and solidification of content distribution, thereby increasing the probability of users discovering new content that interests them, and thus enhancing user stickiness and satisfaction with the platform.

[0087] In step S104, the new digital media content is subjected to dual preference feedback based on the explicit interest matching degree and the potential interest correlation degree, thereby identifying the user's preference feedback value for the new digital media content.

[0088] In some embodiments, the process of identifying a user's preference feedback value for the new digital media content by performing dual preference feedback based on the explicit interest matching degree and the latent interest correlation degree is achieved through the following steps:

[0089] The entropy weight method is used to determine the weight allocation vector between the explicit interest matching degree and the potential interest correlation degree;

[0090] Based on the weight allocation vector, a weighted fusion operation is performed on the explicit interest matching degree and the potential interest correlation degree to obtain the initial preference feedback value;

[0091] The initial preference feedback value is normalized to obtain the user's preference feedback value for the new digital media content.

[0092] In specific implementation, firstly, the entropy weight method is used to determine the weight allocation vector of the explicit interest matching degree and the potential interest correlation degree. Specifically, when using the entropy weight method, explicit interest matching degree and potential interest correlation degree data of different users in historical distribution scenarios of digital media platforms are collected first. This data is then standardized to eliminate the influence of differences in units. Next, the information entropy of each indicator (i.e., explicit interest matching degree and potential interest correlation degree) is calculated, and the difference coefficient of each indicator is derived based on the information entropy (where the difference coefficient can be determined according to: difference coefficient = 1 - entropy value). The difference coefficient of each indicator is divided by the sum of the difference coefficients of all indicators to obtain the corresponding weight. The weights of the two indicators are combined to form a weight allocation vector. This weight allocation vector refers to a set of values ​​representing the proportion of importance of explicit interest matching degree and potential interest correlation degree in preference feedback calculation. First, based on the weighted allocation vector, a weighted fusion operation is performed on the explicit interest matching degree and the potential interest correlation degree. Specifically, the explicit interest matching degree is multiplied by its corresponding weight, the potential interest correlation degree is multiplied by its corresponding weight, and the two product results are added together to obtain the initial preference feedback value. The initial preference feedback value refers to the original preference value that integrates the matching information of explicit and potential interest dimensions. Finally, the initial preference feedback value is processed using the min-max normalization method. That is, by statistically analyzing the maximum and minimum values ​​of all historical initial preference feedback values ​​on the platform, the difference between the current initial preference feedback value and the minimum value is calculated, and then divided by the difference between the maximum and minimum values. The value is mapped to the [0,1] interval to eliminate the difference in the numerical range under different distribution scenarios, thereby obtaining the user's preference feedback value for the new digital media content.

[0093] It should be noted that the preference feedback value in this application refers to an indicator that measures the degree of alignment between users' explicit and potential interests in new digital media content. The preference feedback value is determined because explicit interest matching can only characterize users' known explicit preferences, and potential interest correlation can only reflect users' unexpressed potential needs. When either exists alone, they are both single-dimensional quantitative indicators of interest, which cannot comprehensively and holistically represent the overall degree of user preference for new digital media content. This makes it difficult for the distribution system to make efficient and accurate distribution decisions based on these indicators, and it is easy to have the problem of one-sided decision-making (only focusing on explicit or potential interests). Therefore, by determining the preference feedback value, the problems of incomplete characterization by a single indicator and lack of a unified comparison standard can be solved, and the user group that is truly interested in new digital media content can be accurately screened.

[0094] In step S105, interest user groups are selected from the user database based on the preference feedback value, and new digital media content is distributed to the selected interest user groups.

[0095] In some embodiments, the process of filtering out interested user groups from the user database based on the preference feedback value and distributing new digital media content to the filtered interested user groups is achieved through the following steps:

[0096] The preference feedback values ​​of each user are associated and bound to the user to generate a user-distribution index mapping set;

[0097] Users whose preference feedback values ​​are greater than the feedback threshold are selected from the user-distribution index mapping set to form a candidate interest user group;

[0098] The candidate interest user groups are distributed and prioritized to generate interest user groups;

[0099] Distribute new digital media content to the aforementioned user groups with interests.

[0100] In specific implementation, firstly, each user's corresponding preference feedback value is associated with a unique user identifier (e.g., a user ID assigned by the platform), ensuring that each user identifier corresponds to only one unique preference feedback value, generating a user-distribution index mapping set. This user-distribution index mapping set is a structured data set integrating user identity information and corresponding preference feedback values. Secondly, all user records with preference feedback values ​​greater than a distribution threshold are extracted from the user-distribution index mapping set to form a candidate interest user group. The feedback value threshold can be set according to actual needs or expert knowledge; no limitation is made here. The candidate interest user group refers to the set of users who initially meet the content preference matching requirements. Then, the candidate interest user group is sorted in descending order for distribution priority, based on the user's preference feedback value. Users are sorted from high to low preference feedback values ​​to form an ordered interest user group. This interest user group refers to the target user set after priority sorting. Finally, the digital media platform's built-in intelligent digital media content distribution system distributes new digital media content to the interest user group according to their priority order, using a user-preset receiving method (e.g., APP push notification).

[0101] It should be noted that, in this application, the "interested user group" refers to the user group that can ensure the acceptance of digital media content.

[0102] Furthermore, in another aspect of this application, in some embodiments, this application provides an intelligent digital media content distribution system based on user profiles, referencing... Figure 3 The figure is a schematic diagram of the structure of a user profile-based intelligent digital media content distribution system according to some embodiments of this application. The user profile-based intelligent digital media content distribution system includes: an acquisition module 201, a processing module 202, and an execution module 203, which are described below:

[0103] The acquisition module 201 in this application is mainly used to acquire the complete interaction records of users with historical digital media content, and to construct a user profile model containing explicit interest tags and implicit interest preferences based on the complete interaction records.

[0104] Processing module 202, in this application, is mainly used to extract the feature description vector of digital media content from the metadata of new digital media content when new digital media content enters the distribution system.

[0105] The processing module 202 is further configured to perform similarity matching between the feature description vector and the explicit interest tags in the user profile model to obtain the explicit interest matching degree of the user, and to perform potential association analysis between the feature description vector and the implicit interest preferences in the user profile model to obtain the potential interest association degree of the user.

[0106] In addition, the processing module 202 is also used to provide dual preference feedback on the new digital media content based on the explicit interest matching degree and the potential interest correlation degree, thereby identifying the user's preference feedback value for the new digital media content;

[0107] The execution module 203 in this application is mainly used to filter out interested user groups from the user library based on the preference feedback value, and distribute new digital media content to the filtered interested user groups.

[0108] In addition, this application also provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described intelligent distribution method for digital media content based on user profiles.

[0109] In some embodiments, reference Figure 4 The figure is a schematic diagram of the structure of a computer device implementing a user profile-based intelligent digital media content distribution method according to some embodiments of this application. The user profile-based intelligent digital media content distribution method in the above embodiments can... Figure 4 The computer device shown is used to implement this, and the computer device includes at least one processor 301, a communication bus 302, a memory 303, and at least one communication interface 304.

[0110] The processor 301 can be a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more devices used to control the execution of the intelligent distribution method for digital media content based on user profiles in this application.

[0111] The communication bus 302 can be used to transmit information between the aforementioned components.

[0112] The memory 303 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 303 may exist independently and be connected to the processor 301 via the communication bus 302. The memory 303 may also be integrated with the processor 301.

[0113] The memory 303 stores program code for executing the solution of this application, and its execution is controlled by the processor 301. The processor 301 executes the program code stored in the memory 303. The program code may include one or more software modules. In the above embodiments, the determination of the intelligent distribution method for digital media content based on user profiles can be implemented by the processor 301 and one or more software modules in the program code in the memory 303.

[0114] Communication interface 304 uses any transceiver-like device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0115] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0116] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0117] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described intelligent distribution method for digital media content based on user profiles.

[0118] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0119] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for intelligent distribution of digital media content based on user profiles, characterized in that, Includes the following steps: Obtain complete user interaction records of historical digital media content, and construct a user profile model containing explicit interest tags and implicit interest preferences based on the complete interaction records; When new digital media content enters the distribution system, feature description vectors of the digital media content are extracted from the metadata of the new digital media content. The feature description vector is matched with the explicit interest tags in the user profile model to obtain the explicit interest matching degree of the user. The feature description vector is also matched with the implicit interest preferences in the user profile model to obtain the potential interest correlation degree of the user. The user's preference feedback value for the new digital media content is identified by performing dual preference feedback on the new digital media content based on the explicit interest matching degree and the potential interest correlation degree. Based on the preference feedback values, select interested user groups from the user database and distribute new digital media content to the selected interested user groups; Specifically, the process of matching the feature description vector with the explicit interest tags in the user profile model to obtain the user's explicit interest matching degree includes: The explicit interest tags in the user profile model are vector-mapped to obtain explicit interest tag vectors; the feature description vectors and explicit interest tag vectors are normalized in dimension and aligned in feature space to generate a set of aligned vectors of the same dimension; the initial explicit similarity value of the user is calculated based on the set of aligned vectors of the same dimension; the initial explicit similarity value is normalized to obtain the explicit interest matching degree of the user, where the explicit interest matching degree is an indicator that measures the degree of fit between the user's explicit interests and those of the user in new digital media content; Specifically, performing a latent association analysis between the feature description vector and the implicit interest preferences in the user profile model to obtain the user's latent interest correlation degree includes: The implicit interest preferences in the user profile model are encoded using feature vectors to obtain implicit interest preference vectors; the feature description vectors and implicit interest preference vectors are normalized in dimension and aligned with the feature space to generate a set of same-dimensional adaptation vectors; the user's initial potential matching value is calculated based on the set of same-dimensional adaptation vectors; the initial potential matching value is normalized to obtain the user's potential interest correlation degree, which is an indicator representing the degree of fit between the user's potential interest in new digital media content; The process of identifying user preference feedback values ​​for new digital media content by applying dual preference feedback based on explicit interest matching degree and potential interest correlation degree is achieved through the following steps: The entropy weight method is used to determine the weight allocation vector of the explicit interest matching degree and the potential interest correlation degree; based on the weight allocation vector, the explicit interest matching degree and the potential interest correlation degree are weighted and fused to obtain the initial preference feedback value; the initial preference feedback value is normalized to obtain the user's preference feedback value for the new digital media content.

2. The method as described in claim 1, characterized in that, Constructing a user profile model containing explicit interest tags and implicit interest preferences based on the complete interaction records specifically includes: The complete interaction record is standardized to obtain structured user interaction data; Based on the structured user interaction data, explicit interaction features of users with digital media content are extracted to generate explicit interest tags for users. Based on the structured user interaction data, latent association features not explicitly expressed by users are mined to obtain users' implicit interest preferences; The user's explicit interest tags and implicit interest preferences are combined to form a user profile model.

3. The method as described in claim 1, characterized in that, When new digital media content enters the distribution system, the feature description vector extracted from the metadata of the new digital media content specifically includes: When new digital media content enters the distribution system, the metadata of the new digital media content is processed to standardize the format and obtain a standardized metadata dataset. The standardized metadata dataset is split into textual, categorical, and numerical feature subsets; The textual feature subset is semantically encoded using a pre-trained language model, the categorical feature subset is one-hot encoded, and the numerical feature subset is normalized to obtain the coded features of each type. The feature description vector of digital media content is determined based on the coding features of each type.

4. The method as described in claim 1, characterized in that, The process of filtering out interested user groups from the user database based on the preference feedback values ​​and distributing new digital media content to these selected groups specifically includes: The preference feedback values ​​of each user are associated and bound to the user to generate a user-distribution index mapping set; Users whose preference feedback values ​​are greater than the feedback threshold are selected from the user-distribution index mapping set to form a candidate interest user group; The candidate interest user groups are distributed and prioritized to generate interest user groups; Distribute new digital media content to the aforementioned user groups with interests.

5. The method as described in claim 1, characterized in that, Obtain complete user interaction records for historical digital media content through server logs.

6. A user-profile-based intelligent digital media content distribution system, used to execute the user-profile-based intelligent digital media content distribution method as described in any one of claims 1 to 5, characterized in that, The system includes: The acquisition module is used to acquire the complete interaction records of users with historical digital media content, and to construct a user profile model containing explicit interest tags and implicit interest preferences based on the complete interaction records. The processing module is used to extract the feature description vector of the digital media content from the metadata of the new digital media content when the new digital media content enters the distribution system. The processing module is further configured to perform similarity matching between the feature description vector and the explicit interest tags in the user profile model to obtain the explicit interest matching degree of the user, and to perform potential association analysis between the feature description vector and the implicit interest preferences in the user profile model to obtain the potential interest association degree of the user. The processing module is also used to provide dual preference feedback on the new digital media content based on the explicit interest matching degree and the potential interest correlation degree, thereby identifying the user's preference feedback value for the new digital media content; The execution module is used to filter out interested user groups from the user database based on the preference feedback value and distribute new digital media content to the filtered interested user groups.

7. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing code, and the processor being configured to retrieve the code and execute the intelligent distribution method for digital media content based on user profiles as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the intelligent distribution method for digital media content based on user profiles as described in any one of claims 1 to 5.

Images