An entertainment cultural content propagation mode recognition method based on an improved FP-GROWTH algorithm

Abstract

This invention discloses a method for recognizing patterns in the dissemination of entertainment and cultural content based on an improved FP-GROWTH algorithm, belonging to the field of data mining and pattern recognition technology. This method addresses the problems of high memory consumption, low computational efficiency, and lack of semantic understanding in existing FP-Growth algorithms when processing large-scale, high-dimensional dissemination data. It reduces resource consumption by using a hash table to split the transaction database and combining node swapping technology to optimize FP-tree construction; it introduces a cultural resonance index and uses a deep learning model to extract user behavior and content features to quantify dissemination appeal; finally, it combines time stamps and dynamic thresholds to filter out dissemination patterns with high confidence and high cultural resonance. This invention significantly improves the efficiency of pattern recognition and the practical application value of the results, making it suitable for content recommendation and marketing strategy optimization.

Description

Technical Field

[0001] This invention belongs to the field of data mining and pattern recognition technology, and in particular relates to a method for recognizing the dissemination patterns of entertainment and cultural content based on an improved FP-GROWTH algorithm. Background Technology

[0002] With the rapid development of the internet and social media, the dissemination of entertainment and cultural content, such as films, television dramas, music, short videos, and live streams, on online platforms has experienced explosive growth. This content spreads through complex, multi-dimensional means, forming diverse dissemination patterns. This process is deeply influenced by a combination of factors, including user behavior such as likes, shares, and comments, as well as regional cultural, socio-psychological, and technological platform characteristics. Against this backdrop, mining and identifying these dissemination patterns is crucial for achieving accurate content recommendation, optimizing marketing strategies, predicting cultural trends, and improving platform operational efficiency. However, due to the massive, high-dimensional, and dynamically changing nature of dissemination data, traditional pattern mining methods face significant challenges in practical applications.

[0003] In the field of data mining, association rule mining is a key technology for identifying inherent relationships between data. The Apriori algorithm, a classic association rule mining algorithm, generates candidate frequent itemsets by scanning the database layer by layer, but it has significant limitations when processing large-scale data. This algorithm requires multiple scans of the entire database, resulting in extremely high computational complexity, especially when dealing with entertainment and cultural content dissemination data that often reaches hundreds of millions of records, where its efficiency falls short of practical needs. Furthermore, the number of candidate frequent itemsets generated by the algorithm grows exponentially with the increase in data dimensionality, which drastically consumes memory and computing resources, severely restricting its application in today's high-concurrency, big data scenarios.

[0004] To address the efficiency bottleneck of the Apriori algorithm, the FP-Growth algorithm was proposed. It reduces the number of database scans to two by constructing a frequent pattern tree, thus improving computational complexity to some extent and making it suitable for medium-sized datasets. However, in the specific scenario of entertainment and cultural content dissemination, the data is not only massive in scale but also highly dimensional, encompassing multiple dimensions such as content type, user behavior time series, and geographic tags, while also exhibiting sparsity and dynamic evolution characteristics. Faced with such complex data characteristics, the traditional FP-Growth algorithm still reveals its shortcomings in practical applications, namely excessive memory consumption and insufficient computational efficiency. More importantly, this method lacks consideration for the temporal consistency of dissemination events and the correlation of user groups, resulting in weak generalization ability of the patterns it uncovers. It struggles to accurately reflect the dynamic patterns and underlying drivers of content dissemination in the real world, failing to meet the needs of in-depth analysis of large-scale, high-dimensional, dynamic dissemination data. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a method for identifying patterns in the dissemination of entertainment and cultural content based on an improved FP-GROWTH algorithm, thereby resolving the issues present in the existing technologies.

[0006] Firstly, to achieve the above objectives, this invention provides a method for identifying entertainment and cultural content dissemination patterns based on an improved FP-GROWTH algorithm, comprising the following steps:

[0007] Collect data on the dissemination of entertainment and cultural content from social media or content platforms, and preprocess it to generate a transaction database;

[0008] User behavior and content features are extracted using a deep learning model, and the cultural resonance score (CRS) is calculated.

[0009] The transaction database is split into multiple subsets using a hash table, and the data is transformed into vector form by combining cultural resonance score (CRS) and propagation event characteristics.

[0010] Calculate the support of propagated events in a subset of data and filter frequent itemsets that meet the minimum support threshold;

[0011] An optimized FP-tree is built based on frequent itemsets, and node swapping techniques are used to reduce memory usage.

[0012] Propagation patterns are mined from FP trees, and confidence scores are calculated by combining time markers and cultural resonance CRS.

[0013] By combining prior knowledge, we select propagation patterns with confidence levels higher than the minimum confidence threshold and cultural resonance scores (CRS) higher than the minimum cultural resonance threshold, and generate the final pattern set.

[0014] Optionally, the process of collecting entertainment and cultural content dissemination data from social media or content platforms and preprocessing it to generate a transaction database includes:

[0015] Collect entertainment and cultural content dissemination data from social media or content platforms to construct a transaction database D, where each transaction represents a dissemination event and includes content identifier, dissemination time, user behavior, and user profile;

[0016] The content identifier includes the short video ID, video title, or video content;

[0017] The user behavior includes liking, forwarding, or commenting;

[0018] The user profile includes age, gender, region, or language;

[0019] The data is preprocessed, including removing duplicate and invalid data, standardizing the time format to UTC timestamps, and generating hourly time stamps based on propagation time.

[0020] Optionally, the process of extracting user behavior and content features and calculating the Cultural Resonance Rating (CRS) using a deep learning model includes:

[0021] The behavior score is calculated by taking a sequence of user behavior as input, processing the behavior sequence using a Long Short-Term Memory (LSTM) network model to obtain a user embedding vector, and then calculating a weighted sum of forwarding rate and comment frequency per unit time based on the user embedding vector and statistical indicators.

[0022] The content score is calculated by taking the content title and keywords as input, processing the text using the BERT model to obtain the embedding vector, and then calculating the weighted sum of sentiment and semantic relevance based on the embedding vector.

[0023] The cultural resonance (CRS) value is obtained by weighting and summing the behavioral score and the content score.

[0024] Optionally, the process of using a hash table to split the transaction database into multiple subsets and combining cultural resonance (CRS) and propagation event features to transform the data into vector form includes:

[0025] Choose a hash function and use open addressing to resolve hash address collisions to construct a hash table;

[0026] The transaction database D is split into k subsets based on content identifiers or user groups using a hash table;

[0027] For each subset of data, the features extracted for the propagation events include content category, time stamp, user behavior, user profile, propagation speed, user activity, and cultural resonance (CRS).

[0028] The content categories include funny videos, game videos, or food videos;

[0029] The user group includes age or region;

[0030] The propagation speed is the number of forwards per unit time;

[0031] The user activity level refers to the frequency of user interaction.

[0032] The extracted features are converted into vector form, where each vector contains the feature value.

[0033] Optionally, the process of constructing an optimized FP-tree based on frequent itemsets and using node swapping techniques to reduce memory usage includes:

[0034] Sort frequent itemsets by support from high to low;

[0035] The transactions are assigned to corresponding subsets using a hash table based on the last item of each transaction.

[0036] For each transaction in a subset, recursively insert it into the FP tree, including retrieving the first propagated event from the transaction, checking if a node with the same name exists in the FP tree, incrementing the count of the node if it exists, and creating a new node and setting the count to 1 if it does not exist.

[0037] When a new node is inserted, if the difference in support between the node and its successor is within a predetermined range and the current node has only a single subtree, then a node swap operation is performed to adjust the node's position and reduce the tree depth.

[0038] Optionally, the process of mining propagation patterns from FP trees and calculating confidence levels by combining time markers and cultural resonance CRS includes:

[0039] If the FP-tree contains a single path, then a propagation pattern is generated for all possible combinations of nodes in the path, and the support of the pattern is the minimum support of the nodes in the combination.

[0040] If the FP-tree contains multiple paths, then recursively call the FP-Growth algorithm to generate a frequent itemset for each frequent item in the head table;

[0041] For each generated frequent itemset, a confidence score is calculated, which is the support of the frequent itemset co-occurring with the condition divided by the support of the frequent itemset, where the condition includes a time stamp and cultural resonance (CRS).

[0042] Secondly, the present invention also provides an entertainment culture content dissemination pattern recognition system based on an improved FP-GROWTH algorithm, used to implement an entertainment culture content dissemination pattern recognition method based on an improved FP-GROWTH algorithm, the system comprising:

[0043] The data acquisition and preprocessing module is used to collect entertainment and cultural content dissemination data from social media or content platforms and preprocess it to generate a transaction database;

[0044] The feature extraction and CRS calculation module is used to extract user behavior and content features through deep learning models and calculate cultural resonance (CRS).

[0045] The data splitting and vectorization module is used to split the transaction database into multiple subsets using a hash table, and to transform the data into vector form by combining cultural resonance score (CRS) and dissemination event characteristics.

[0046] The support calculation and filtering module is used to calculate the support of propagated events in a subset of data and filter frequent itemsets that meet the minimum support threshold.

[0047] The FP-tree construction and optimization module is used to build optimized FP-trees based on frequent itemsets and to reduce memory usage using node swapping techniques.

[0048] The pattern mining and confidence calculation module is used to mine propagation patterns from FP trees and calculate confidence by combining time stamps and cultural resonance CRS.

[0049] The pattern filtering module is used to combine prior knowledge to filter propagation patterns with confidence levels higher than the minimum confidence threshold and cultural resonance scores (CRS) higher than the minimum cultural resonance threshold, generating the final pattern set.

[0050] Thirdly, the present invention also provides a computer terminal device, comprising:

[0051] One or more processors;

[0052] A memory, coupled to the processor, for storing one or more programs;

[0053] When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the entertainment and cultural content dissemination pattern recognition method based on the improved FP-GROWTH algorithm in the first aspect described above.

[0054] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the entertainment and cultural content dissemination pattern recognition method based on the improved FP-GROWTH algorithm in the first aspect described above.

[0055] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the entertainment and cultural content dissemination pattern recognition method based on the improved FP-GROWTH algorithm described in the first aspect above.

[0056] Compared with the prior art, the present invention has the following advantages and technical effects:

[0057] This invention provides a method for identifying entertainment and cultural content dissemination patterns based on an improved FP-GROWTH algorithm. By introducing hash table data splitting and node swapping techniques, this invention optimizes the FP-tree construction process, significantly reducing memory usage and computational complexity, providing an efficient solution for large-scale data mining. Simultaneously, it introduces the Cultural Resonance Index (CRS) to quantify the content's appeal to specific user groups, and combines deep learning to extract user behavior and content features, ensuring the mined dissemination patterns exhibit temporal consistency and group relevance, thus enhancing the pattern's confidence and practical application value. Furthermore, it supports dynamic adjustment of the minimum support and confidence thresholds to adapt to different data scales and business needs, demonstrating high scalability. This invention can capture dynamic changes in dissemination patterns, identify high-frequency dissemination behaviors within specific time periods, and, by combining user behavior characteristics and user group attributes, improve user experience, accurately predict dissemination trends, and reduce prediction costs. Attached Figure Description

[0058] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0059] Figure 1 This is a flowchart illustrating the identification of entertainment and cultural content dissemination patterns according to an embodiment of the present invention.

[0060] Figure 2 This is a schematic diagram illustrating the construction process of the FP-tree in an embodiment of the present invention;

[0061] Figure 3 This is a flowchart of node swapping in an FP tree according to an embodiment of the present invention. Detailed Implementation

[0062] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0063] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0064] Example 1

[0065] like Figure 1 As shown, this embodiment provides a method for identifying entertainment and cultural content dissemination patterns based on an improved FP-GROWTH algorithm, including:

[0066] Collect data on the dissemination of entertainment and cultural content from social media or content platforms, and preprocess it to generate a transaction database;

[0067] User behavior and content features are extracted using a deep learning model, and the cultural resonance score (CRS) is calculated.

[0068] The transaction database is split into multiple subsets using a hash table, and the data is transformed into vector form by combining cultural resonance score (CRS) and propagation event characteristics.

[0069] Calculate the support of propagated events in a subset of data and filter frequent itemsets that meet the minimum support threshold;

[0070] An optimized FP-tree is built based on frequent itemsets, and node swapping techniques are used to reduce memory usage.

[0071] Propagation patterns are mined from FP trees, and confidence scores are calculated by combining time markers and cultural resonance CRS.

[0072] By combining prior knowledge, we select propagation patterns with confidence levels higher than the minimum confidence threshold and cultural resonance scores (CRS) higher than the minimum cultural resonance threshold, and generate the final pattern set.

[0073] Furthermore, the above overall process specifically includes the following seven steps:

[0074] S1. Collect entertainment and cultural content dissemination data from social media or content platforms, preprocess and generate a transaction database D, which includes content identifiers, time stamps, user behavior and user profiles, etc.

[0075] S2. Extract user behavior and content features through deep learning models, calculate cultural resonance (CRS), and quantify the appeal of content to specific user groups.

[0076] S3. Use a hash table to split D into subsets, and combine CRS and propagation event features (content category, propagation speed) to transform them into vector form;

[0077] S4. Calculate the support of propagated events in the subset and filter out the frequent itemsets L that meet the minimum support threshold;

[0078] S5. Optimize the FP tree based on L and reduce memory usage by using node swapping;

[0079] S6. Discover the FP-tree generation and propagation patterns, and calculate the confidence level by combining time stamps and CRS;

[0080] S7. Combine prior knowledge to select propagation patterns with high confidence and high CRS, and generate the final pattern set C.

[0081] As one implementation method in this embodiment, the process of collecting entertainment and cultural content dissemination data from social media or content platforms and preprocessing it to generate a transaction database includes:

[0082] Collect entertainment and cultural content dissemination data from social media or content platforms to construct a transaction database D, where each transaction represents a dissemination event and includes content identifier, dissemination time, user behavior, and user profile;

[0083] The content identifier includes the short video ID, video title, or video content;

[0084] The user behavior includes liking, forwarding, or commenting;

[0085] The user profile includes age, gender, region, or language;

[0086] The data is preprocessed, including removing duplicate and invalid data, standardizing the time format to UTC timestamps, and generating hourly time stamps based on propagation time.

[0087] Furthermore, step S1 includes:

[0088] Collect data on the dissemination of entertainment and cultural content from social media or content platforms (such as Weibo and Douyin) to build a transaction database. Each transaction This represents a propagation event, including content identifiers (such as short video ID, video title, and video content), propagation time, user behavior (likes, shares, comments, etc.), and user profiles (age, gender, region, language). Preprocessing includes removing duplicate and invalid data, standardizing the time format (converting to UTC timestamps), and generating time stamps based on the propagation time (such as dividing time windows by hours) to capture the dynamic characteristics of propagation.

[0089] Where D represents a transactional database, Let represent the i-th propagation event, and n represent the total number of transactions.

[0090] As one implementation method in this embodiment, the process of extracting user behavior and content features and calculating the cultural resonance score (CRS) using a deep learning model includes:

[0091] The behavior score is calculated by taking a sequence of user behavior as input, processing the behavior sequence using a Long Short-Term Memory (LSTM) network model to obtain a user embedding vector, and then calculating a weighted sum of forwarding rate and comment frequency per unit time based on the user embedding vector and statistical indicators.

[0092] The content score is calculated by taking the content title and keywords as input, processing the text using the BERT model to obtain the embedding vector, and then calculating the weighted sum of sentiment and semantic relevance based on the embedding vector.

[0093] The cultural resonance (CRS) value is obtained by weighting and summing the behavioral score and the content score.

[0094] Furthermore, step S2 includes:

[0095] A CRS value is generated for each content-user pair. A behavior score is calculated based on user behavior data, and a content score is calculated using the BERT model. The two scores are then weighted to obtain the CRS value.

[0096] This section details the calculation of behavioral scores, which assess the intensity of user interaction with entertainment content (such as short videos or social media posts), reflecting user engagement with the content. The input is a sequence of user actions, i.e., a list of user actions arranged chronologically, such as [likes, comments, shares]. These actions capture the dynamic interaction patterns between users and content. The calculation process first uses a Long Short-Term Memory (LSTM) network model to process the behavioral sequence, transforming it into user embedding vectors. Then, based on the user embeddings and statistical metrics, the behavioral score is calculated.

[0097]

[0098] Where 'c' represents content (e.g., short video), 'u' represents users, and 'F' represents the sharing rate, which is the number of shares divided by the total number of interactions, reflecting the content's potential for dissemination. 'C' also represents the comment frequency per unit of time, reflecting the level of user enthusiasm for discussion. , The weights are set according to the actual situation.

[0099] To calculate the content score, the text information, including the title and keywords, is first preprocessed and then input into the BERT model to obtain embedding vectors. The embedding vectors are then weighted by sentiment and semantic relevance to arrive at the final score. The formula for calculating semantic relevance is as follows:

[0100]

[0101] in The vector obtained through the model, This is a predefined reference vector.

[0102] The proportion of positive sentiment words in the title was calculated using a Chinese sentiment lexicon.

[0103]

[0104] in For the number of positive words, This represents the total number of words. The content score is calculated using the following formula:

[0105]

[0106] in and These are the weight parameters.

[0107] Most importantly, the CRS value is calculated from the content score and the behavior score:

[0108]

[0109] Where 'c' represents content (such as short videos) and 'u' represents users. and The weights are set according to the actual situation.

[0110] As one implementation method in this embodiment, the process of splitting the transaction database into multiple subsets using a hash table and transforming the data into vector form by combining the cultural resonance score (CRS) and propagation event characteristics includes:

[0111] Choose a hash function and use open addressing to resolve hash address collisions to construct a hash table;

[0112] The transaction database D is split into k subsets based on content identifiers or user groups using a hash table;

[0113] For each subset of data, the features extracted for the propagation events include content category, time stamp, user behavior, user profile, propagation speed, user activity, and cultural resonance (CRS).

[0114] The content categories include funny videos, game videos, or food videos;

[0115] The user group includes age or region;

[0116] The propagation speed is the number of forwards per unit time;

[0117] The user activity level refers to the frequency of user interaction.

[0118] The extracted features are converted into vector form, where each vector contains the feature value.

[0119] Furthermore, step S3 includes:

[0120] Choose a suitable hash function, use open addressing to resolve hash collisions, and construct a hash table. Use the hash table to split the transaction database (D) into multiple subsets, extract features of the propagation events, including content category (e.g., funny videos, game videos, food videos), timestamps, user behavior (likes, shares, comments), and user profiles (age, gender, region, language). Finally, convert the features and CRS values ​​into vector form.

[0121] Specifically, this includes: using a hash function to split the transaction database D into k subsets based on content identifiers or user groups (such as age, region). This reduces the size of the data processed in a single batch. For each subset of data... Features are extracted from the propagation events, including content category (e.g., funny videos, gaming videos, food videos), propagation speed (number of reposts per unit time), user activity (frequency of user interaction), and CRS, and these features are transformed into vector form:

[0122]

[0123] in This represents the feature vector of the i-th propagation event. Let m represent the j-th feature value (such as the propagation speed value), and m be the feature dimension.

[0124] Furthermore, step S4 includes:

[0125] Calculate the support of propagated events in the subset of data, and filter out frequent itemsets that meet the minimum support threshold. Delete itemsets with support lower than [a certain threshold]. The events generate frequent itemsets L.

[0126] Step S4 specifically includes:

[0127] For each subset of datasets The support of propagation events is calculated, and frequent itemsets (L) that meet the minimum support threshold are selected. The formula is:

[0128]

[0129] in Indicates an event Support Indicates an event The number of times it appears, The total number of transactions in the subset Di is represented by min_sup, which represents the minimum support threshold. Events with support lower than min_sup are removed, and the frequent itemset L is obtained.

[0130] As one implementation method in this embodiment, the process of constructing an optimized FP-tree based on frequent itemsets and using node swapping techniques to reduce memory usage includes:

[0131] Sort frequent itemsets by support from high to low;

[0132] The transactions are assigned to corresponding subsets using a hash table based on the last item of each transaction.

[0133] For each transaction in a subset, recursively insert it into the FP tree, including retrieving the first propagated event from the transaction, checking if a node with the same name exists in the FP tree, incrementing the count of the node if it exists, and creating a new node and setting the count to 1 if it does not exist.

[0134] When a new node is inserted, if the difference in support between the node and its successor is within a predetermined range and the current node has only a single subtree, then a node swap operation is performed to adjust the node's position and reduce the tree depth.

[0135] Furthermore, step S5 includes:

[0136] Frequent itemsets L are sorted in descending order of support, and transactions are allocated to hash table V. Then, an FP-tree is constructed, and the tree structure is optimized through node swapping to reduce memory usage and improve computational efficiency. If node r exists, its count is incremented by 1; otherwise, a new node is created to optimize the structure.

[0137] Step S5 specifically includes:

[0138] Reference Figure 2 To construct an FP-tree, firstly, sort the propagated events in the frequent itemset L according to their support and CRS value from high to low, placing the most frequent propagated events (such as "like" or "share") at the top to ensure high-frequency events are processed first. Then, based on the last item of each transaction, use a hash table V to assign the transaction to the corresponding subset. For each transaction in hash table V, construct a sub-FP-tree step by step: extract the first propagated event r (e.g., "like") from the current transaction, denote the remaining events as R (e.g., "share, weekend"), and check the existing FP-tree T, determining the node's status and inserting it into the FP-tree accordingly.

[0139]

[0140] Where r is the currently processed propagation event, R is the remaining events in the transaction, and T is the current FP tree, labeled c.

[0141] If a node c with the same name as r already exists in T (e.g., there is already a "like" node), then increment the count of c by 1. If no node with the same name as r exists, create a new node N with the same name as node r, and set the count of N to 1. If either condition is met and And if node c has only a single subtree (where e is a propagation event after r), then the count of node c is incremented by 1 and node swapping is performed (see [reference]). Figure 3 To reduce tree depth; if not satisfied or support difference > If R is not empty, then create a new node q, which will be a subtree of the parent node and linked to the linked list of nodes with the same name. Ultimately, this is used to construct the target tree.

[0142] As one implementation method in this embodiment, the process of mining propagation patterns from FP trees and calculating confidence levels by combining time stamps and cultural resonance (CRS) includes:

[0143] If the FP-tree contains a single path, then a propagation pattern is generated for all possible combinations of nodes in the path, and the support of the pattern is the minimum support of the nodes in the combination.

[0144] If the FP-tree contains multiple paths, then recursively call the FP-Growth algorithm to generate a frequent itemset for each frequent item in the head table;

[0145] For each generated frequent itemset, a confidence score is calculated, which is the support of the frequent itemset co-occurring with the condition divided by the support of the frequent itemset, where the condition includes a time stamp and cultural resonance (CRS).

[0146] Furthermore, step S6 includes:

[0147] Discover FP-tree generation and propagation patterns, and calculate pattern confidence using time stamps. Generate single-path P-pair node combination patterns. The support is the minimum node support; FP-Growth is recursively called in multiple paths to generate frequent itemsets, and the confidence of each frequent itemset is calculated by combining the time stamp of the propagation event and CRS.

[0148] Step S6 specifically includes:

[0149] Perform frequent itemset mining on the FP-tree to generate propagation patterns. If the FP-tree contains a single path P, then for each combination of nodes in P... Generation mode Its support rate is Minimum support of nodes in the middle; if the FP-tree contains multiple paths, then for each frequent item in the head table Recursive call to FP_Growth(FP-tree, Generate frequent itemsets and calculate the confidence score for each frequent itemset using the following formula:

[0150]

[0151] in This represents all possible combinations of actions in path P (e.g., the path "like → share" can generate three combinations: "like", "share", and "like + share"). These are added conditions (like "short videos" or "user activity"), Representation pattern Confidence level, Indicate combination Support that appears together with condition α Representation pattern The level of support is used to filter the final propagation pattern by calculating confidence levels.

[0152] Furthermore, step S7 includes:

[0153] By combining prior knowledge to select high-confidence propagation patterns, a final pattern set is generated. If the confidence level is higher than... The text has a higher resonance level than If the data is selected, it is added to set C, and the selected data represents the propagation patterns with high propagation potential.

[0154] Step S7 specifically includes:

[0155] Based on prior knowledge (such as cultural background, regional preferences, and platform characteristics), the propagation patterns generated by S5 are filtered, retaining those with a confidence level higher than [missing information]. and those with higher cultural resonance The pattern generates the final propagation pattern set C, with the formula:

[0156]

[0157] Where C is the final set of modes, and β is the candidate propagation mode. The confidence level of pattern β. The minimum confidence threshold, For the cultural resonance of pattern β, This is the minimum cultural resonance threshold. Through screening, the patterns in C are ensured to have high reliability and practical application value, suitable for content recommendation and marketing strategies. For example, if "Shanghai users love to share funny videos on weekends" is highly reliable, it will be retained, while less reliable patterns will be eliminated. This process ensures that the patterns in C are reliable and can be practically used to recommend content, optimize marketing plans, or predict trends.

[0158] Based on this, the present invention provides a method for identifying entertainment and cultural content dissemination patterns based on an improved FP-GROWTH algorithm, the technical effects of which include:

[0159] (1) This invention introduces hash table splitting technology to decompose a large-scale transactional database into multiple subsets, effectively reducing the computational load per processing session and lowering the dependence on computing resources. Simultaneously, it utilizes node swapping technology to optimize the FP-tree structure, reducing tree depth by adjusting node positions, compressing data storage space, and significantly reducing memory usage. This technology provides an efficient solution for large-scale data mining, and is particularly suitable for the high-concurrency data processing needs of social media platforms.

[0160] (2) This invention introduces the Cultural Resonance Index (CRS), which ensures that the pattern matches the user's cultural background and quantifies the attractiveness of content dissemination, thereby enhancing the semantic expressiveness of the pattern and distinguishing it from traditional association rule mining.

[0161] (3) This invention supports dynamic adjustment of minimum support and confidence thresholds to adapt to different data scales and business needs. It has high scalability and provides a general and efficient solution for platform operation and data-driven decision-making.

[0162] (4) This invention can capture the dynamic changes in dissemination patterns, such as identifying high-frequency dissemination behaviors within specific time periods (e.g., holidays). Simultaneously, it combines user behavior characteristics and user group attributes. The discovered dissemination patterns not only reflect the frequency of content dissemination but also embody the correlation of user groups, significantly improving user experience, enabling more accurate prediction of dissemination trends, and reducing prediction costs.

[0163] Example 2

[0164] In this embodiment, a computer terminal device is provided, including:

[0165] One or more processors;

[0166] A memory, coupled to the processor, for storing one or more programs;

[0167] When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the above-described entertainment and cultural content dissemination pattern recognition method based on the improved FP-GROWTH algorithm.

[0168] In this embodiment, a computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the above-described method for identifying the dissemination pattern of entertainment and cultural content based on the improved FP-GROWTH algorithm.

[0169] In this embodiment, an electronic device is also provided, including a memory and a processor. The memory stores a computer program, and the processor is configured to run the computer program to perform the steps of the above-described entertainment and cultural content dissemination pattern recognition method based on the improved FP-GROWTH algorithm.

[0170] In this embodiment, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the above-described method for identifying entertainment and cultural content dissemination patterns based on the improved FP-GROWTH algorithm.

[0171] The aforementioned program can run on a processor or be stored in memory (or a computer-readable medium). Computer-readable media includes both permanent and non-permanent, removable and non-removable media, and information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random-access memory (SRAM), dynamic random-access memory (DRAM), other types of random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0172] These computer programs may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes can be implemented by different modules for different steps.

[0173] This embodiment provides such a device or system. The system, referred to as an entertainment culture content dissemination pattern recognition system based on an improved FP-GROWTH algorithm, includes:

[0174] The data acquisition and preprocessing module is used to collect entertainment and cultural content dissemination data from social media or content platforms and preprocess it to generate a transaction database;

[0175] The feature extraction and CRS calculation module is used to extract user behavior and content features through deep learning models and calculate cultural resonance (CRS).

[0176] The data splitting and vectorization module is used to split the transaction database into multiple subsets using a hash table, and to transform the data into vector form by combining cultural resonance score (CRS) and dissemination event characteristics.

[0177] The support calculation and filtering module is used to calculate the support of propagated events in a subset of data and filter frequent itemsets that meet the minimum support threshold.

[0178] The FP-tree construction and optimization module is used to build optimized FP-trees based on frequent itemsets and to reduce memory usage using node swapping techniques.

[0179] The pattern mining and confidence calculation module is used to mine propagation patterns from FP trees and calculate confidence by combining time stamps and cultural resonance CRS.

[0180] The pattern filtering module is used to combine prior knowledge to filter propagation patterns with confidence levels higher than the minimum confidence threshold and cultural resonance scores (CRS) higher than the minimum cultural resonance threshold, generating the final pattern set.

[0181] As one implementation method in this embodiment, the data acquisition and preprocessing module includes:

[0182] The data acquisition unit is used to collect entertainment and cultural content dissemination data from social media or content platforms and construct a transaction database D, where each transaction represents a dissemination event and includes content identifier, dissemination time, user behavior, and user profile;

[0183] The preprocessing unit is used to preprocess the data, including removing duplicate and invalid data, unifying the time format to UTC timestamps, and generating hourly time stamps based on the propagation time.

[0184] As one implementation method in this embodiment, the feature extraction and CRS calculation module includes:

[0185] The behavior score calculation unit is used to calculate the behavior score. The input is a user behavior sequence. The behavior sequence is processed using a Long Short-Term Memory (LSTM) network model to obtain the user embedding vector. Then, based on the user embedding vector and statistical indicators, a weighted sum of forwarding rate and comment frequency per unit time is calculated.

[0186] The content score calculation unit is used to calculate the content score. The input is the content title and keywords. The text is processed using the BERT model to obtain the embedding vector. Then, a weighted sum of sentiment and semantic relevance is calculated based on the embedding vector.

[0187] The CRS calculation unit is used to weight and calculate the behavior score and content score to obtain the cultural resonance CRS value.

[0188] As one implementation method in this embodiment, the data splitting and vectorization module includes:

[0189] The hash table building unit is used to select a hash function and resolve hash address collisions using open addressing to build a hash table.

[0190] The data splitting unit is used to split the transaction database D into k subsets based on content identifiers or user groups using a hash table;

[0191] The feature extraction unit is used to extract features for each subset of the data, including content category, time stamp, user behavior, user profile, propagation speed, user activity, and cultural resonance (CRS).

[0192] The vectorization unit is used to convert the extracted features into vector form, where the vector contains the feature values.

[0193] As one implementation method in this embodiment, the FP-tree construction and optimization module includes:

[0194] The sorting unit is used to sort frequent itemsets from high to low support.

[0195] A transaction allocation unit is used to allocate transactions to corresponding subsets based on the last item of each transaction using a hash table.

[0196] The tree building unit is used to recursively insert transactions into the FP tree for each subset. This includes retrieving the first propagation event from the transaction, checking if a node with the same name exists in the FP tree, incrementing the count of the node if it exists, and creating a new node and setting the count to 1 if it does not exist.

[0197] The node swapping unit is used to perform a node swapping operation to adjust the node position and reduce the tree depth when a new node is inserted, provided that the support difference between the node and its successor is within a predetermined range and the current node has only a single subtree.

[0198] As one implementation method in this embodiment, the pattern mining and confidence calculation module includes:

[0199] A single-path processing unit is used to generate a propagation pattern for all possible combinations of nodes in a path when the FP-tree contains a single path. The support of this pattern is the minimum support of the nodes in the combination.

[0200] The multi-path processing unit is used to recursively call the FP-Growth algorithm to generate frequent itemsets for each frequent item in the head table when the FP-tree contains multiple paths.

[0201] The confidence calculation unit is used to calculate the confidence for each generated frequent itemset, wherein the confidence is the support of the frequent itemset co-occurring with the condition divided by the support of the frequent itemset, wherein the condition includes a time stamp and cultural resonance (CRS).

[0202] The system or apparatus is used to implement the functions of the methods in the above embodiments. Each module in the system or apparatus corresponds to each step in the method, as has been described in the method and will not be repeated here.

[0203] The above implementation method solves the problem of entertainment and cultural content dissemination pattern recognition based on the improved FP-GROWTH algorithm in related technologies, thereby ensuring that the problems existing in the prior art are resolved.

[0204] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An entertainment culture content dissemination pattern recognition method based on an improved FP-GROWTH algorithm, characterized in that, The method comprises the following steps: collecting and preprocessing entertainment culture content dissemination data from social media or content platforms to generate a transaction database; extracting user behavior and content features through a deep learning model and calculating cultural resonance CRS, the process including: calculating behavior scores, inputting user behavior sequences, processing behavior sequences using a long short-term memory network LSTM model to obtain user embedding vectors, and then calculating the weighted sum of forwarding rates and comment frequencies per unit time based on user embedding vectors and statistical indicators; calculating content scores, inputting content titles and keywords, processing text using a BERT model to obtain embedding vectors, and then calculating the weighted sum of sentiment orientation and semantic relevance based on embedding vectors; calculating the weighted sum of behavior scores and content scores to obtain cultural resonance CRS values; splitting the transaction database into multiple sub-datasets using a Hash table and converting data into vector form by combining cultural resonance CRS and dissemination event features, the process including: selecting a Hash function and using open addressing to solve Hash address conflicts to build a Hash table; splitting the transaction database D into k sub-datasets according to content identifiers or user groups through the Hash table; for each dissemination event in a sub-dataset, extracting features including content category, time marker, user behavior, user portrait, dissemination speed, user activity, and cultural resonance CRS; calculating the support of dissemination events in a sub-dataset and filtering frequent item sets meeting a minimum support threshold; building an optimized FP tree based on frequent item sets and using node exchange technology to reduce memory occupancy; mining dissemination patterns from the FP tree and calculating confidence levels in combination with time markers and cultural resonance CRS, the process including: if the FP tree contains a single path, generating a dissemination pattern for all possible combinations of nodes in the path, and the support of the pattern is the minimum support of the nodes in the combination; if the FP tree contains multiple paths, recursively calling the FP-Growth algorithm to generate frequent item sets for each frequent item in the header table; for each generated frequent item set, calculating a confidence level, which is the support of the frequent item set and the condition appearing together divided by the support of the frequent item set, wherein the condition includes time markers and cultural resonance CRS; filtering dissemination patterns with confidence levels higher than a minimum confidence threshold and cultural resonance CRS higher than a minimum cultural resonance threshold in combination with prior knowledge to generate a final pattern set.

2. The method of claim 1, wherein, The process of collecting and preprocessing entertainment culture content dissemination data from social media or content platforms to generate a transaction database comprises: collecting entertainment culture content dissemination data from social media or content platforms and building a transaction database D, wherein each transaction represents a dissemination event and contains content identifiers, dissemination times, user behaviors, and user portraits; the content identifier includes a short video ID, a video title, or video content; the user behavior includes likes, forwards, or comments; the user portrait includes age, gender, region, or language; The data is pre-processed, including removing duplicates and invalid data, unifying the time format to UTC timestamp, and generating hourly time markers based on propagation time.

3. The method of claim 1, wherein, The process of splitting the transaction database into multiple subsets using a Hash table and converting the data into vector form by combining cultural resonance CRS and propagation event features includes: The content categories include funny videos, game videos, or food videos; The user groups include age or region; The propagation speed is the number of forwards per unit time; The user activity is the user interaction frequency; The extracted features are converted into vector form, where the vector contains the feature values.

4. The method of claim 1, wherein, The process of constructing an optimized FP tree based on frequent itemsets and using node exchange techniques to reduce memory usage includes: Sort the frequent itemsets by support from high to low; Use a Hash table to distribute transactions to the corresponding subset according to the last item of each transaction; For transactions in each subset, recursively insert into the FP tree, including taking the first propagation event from the transaction, checking if there is a node with the same name in the FP tree, if there is, increase the count of that node, if not, create a new node and set the count to 1; When a new node is inserted, if the support difference between the node and its successor is within a predetermined range and the current node has only a single subtree, perform a node exchange operation to adjust the node position and reduce the tree depth.

5. An entertainment culture content propagation pattern recognition system based on an improved FP-GROWTH algorithm, characterized by, The system for implementing the method of any one of claims 1-4 includes: A data collection and preprocessing module for collecting entertainment culture content propagation data from social media or content platforms and preprocessing to generate a transaction database; A feature extraction and CRS calculation module for extracting user behavior and content features through a deep learning model and calculating cultural resonance CRS; A data splitting and vectorization module for splitting the transaction database into multiple subsets using a Hash table and converting the data into vector form by combining cultural resonance CRS and propagation event features; A support calculation and filtering module for calculating the support of propagation events in the sub-data set and filtering frequent itemsets that meet the minimum support threshold; An FP tree construction and optimization module for constructing an optimized FP tree based on frequent itemsets and using node exchange techniques to reduce memory usage; A pattern mining and confidence calculation module for mining propagation patterns from the FP tree and calculating confidence in combination with time markers and cultural resonance CRS; A pattern filtering module for filtering propagation patterns with confidence higher than the minimum confidence threshold and cultural resonance CRS higher than the minimum cultural resonance threshold based on prior knowledge to generate a final pattern set.

6. A computer terminal device, characterized by It includes: One or more processors; Memory coupled to the processor for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the method of any one of claims 1-4.

7. A computer-readable storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the steps of the method of any one of claims 1-4.

8. A computer program product comprising a computer program, characterized in that, The computer program, which when executed by the processor, implements the steps of the method of any one of claims 1-4.

Images