Data information pushing system based on cloud conference

Abstract

The invention relates to the technical field of big data analysis, in particular to a data information pushing system based on a cloud conference, which comprises a data processing and sequence division module, a conference data association analysis module, a conference participating user behavior pattern recognition module and a conference content pushing formulation and optimization module. By extracting and processing conference data, an accurate and ordered data basis is provided for subsequent analysis; a correlation coefficient and a clustering algorithm are utilized to deeply mine relevance among conference data, conference themes are accurately divided, and powerful support is provided for conference content pushing; by defining a behavior label and training a decision tree classifier, the behavior pattern of a participant user is effectively identified, and an important basis is provided for personalized conference content pushing; by calculating the pushing weight and adjusting the pushing strategy, accurate and efficient conference content pushing is realized. The method and the device are used for solving the technical problem that content association and user behavior analysis are not comprehensive in current information management pushing.

Description

Technical Field

[0001] The present invention relates to the technical field of big data analysis, and in particular to a data information push system based on cloud conference. Background Art

[0002] Big data analysis technology refers to the use of advanced computer technology and tools to collect, store, process and analyze massive amounts of data in order to discover the patterns and value hidden in the data. It mainly includes data preprocessing, data mining, data visualization, data analysis and machine learning technologies.

[0003] In the current field of information management and push, the following problems still exist: insufficient processing of conference data sets, huge conference data, with the massive increase of data, and the influence of factors such as data diversity, it may not be possible to accurately process data, resulting in insufficient push reliability; incomplete analysis of conference user behavior, resulting in the inability to push personalized content based on user behavior; push strategy lacks flexibility, traditional information push strategy may be relatively simple, lack of flexibility for dynamic adjustment based on user feedback, thus affecting the push effect and user satisfaction. To this end, the present invention proposes a data information push system based on cloud conference. Summary of the invention

[0004] The purpose of the present invention is to solve the problems in the background technology and to propose a data information push system based on cloud conference.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A data information push system based on cloud conference, including: a data processing and sequence division module, a conference data association analysis module, a participant user behavior pattern recognition module, and a conference content push formulation and optimization module;

[0007] Data processing and sequence division module: extracts meeting data sets from the database, processes and sequence divides meeting data containing time series characteristics for subsequent analysis; the meeting data sets include meeting timestamps, meeting user sets, and meeting content;

[0008] Conference data association analysis module: Use correlation coefficients to analyze the association between conference data and filter and obtain conference data groups; use clustering algorithms to classify conference data groups by conference topics and determine the topic attribution of conference data in each conference data group;

[0009] Participant user behavior pattern recognition module: extract the participant user behavior data set from the participant user set and define the behavior label; map the participant user behavior label by training the decision tree classifier, so as to identify the user behavior pattern;

[0010] Conference content push formulation and optimization module: Based on the conference content and the behavioral tags of the participating users, the push weight is calculated and obtained, and then the conference content is pushed to the participating users.

[0011] Behavior patterns of conference users Behavior patterns of conference users Further, the data processing and sequence division module extracts the conference data set from the database, and processes and sequence divides the conference data containing time series characteristics, including:

[0012] Extract the meeting data set from the database and mark it as

[0013] d g ={d1,d2,…,d n};

[0014] Among them, d g represents the g-th meeting data, including the meeting timestamp, the set of participating users (such as ID, name, user behavior records, etc.) and the meeting content. g represents the index of the meeting data, g = 1, 2, ..., n, n represents the total number of meeting data, d1, d2, ..., d n Respectively represent the first meeting data, the second meeting data, ..., the nth meeting data;

[0015] For meeting data containing time series characteristics (such as meeting timestamps), the sliding window technology is used for processing:

[0016] Define the sliding window size as w, which is the length of each subsequence into which the conference data is divided;

[0017] Then there is a time series data set T = {t1, t2, ..., t m} and is divided into multiple time subsequences

[0018] T r ={t (r-1)w+1 ,t (r-1)w+2 ,...,t rw}, where

[0019] Where m represents the index of the time series, r represents the index of the time subsequence, t1, t2, ..., t m Respectively represent the first time series data, the second time series data, ..., the mth time series data, t (r-1)w+1 ,t (r-1)w+2 ,...,t rw They represent the (r-1)w+1th time subsequence, the (r-1)w+2th time subsequence, ..., the rwth time subsequence, Express Round up.

[0020] Furthermore, the conference data association analysis module uses the correlation coefficient to analyze the association between conference data, and the process of screening and obtaining conference data groups includes:

[0021] For any two conference data, d i ,d j , using Pearson correlation coefficient PCR ij Analyze meeting data i and d j The correlation between them; the calculation formula of Pearson correlation coefficient is:

[0022]

[0023] In the formula, PCR ij represents the Pearson correlation coefficient, i and j represent the indexes of any two conference data, and a 1H ,a 2H Corresponding to any two conference data d i and d j The Hth numeric attribute of Represents any two conference data d i and d j , H represents the index of the numerical attribute;

[0024] According to the calculation results of Pearson correlation coefficient, the correlation between conference data is analyzed. ij >0, then any two conference data are judged to be positively correlated; if PCR ij <0, then any two conference data are judged to be negatively correlated; if |PCR ij |The closer it is to 1, the stronger the correlation between the two conference data is; if PCR ij =0, it is judged that there is no linear relationship between the two conference data;

[0025] Screen out any two conference data with positive correlation coefficients and aggregate them into a conference data group for conference topic classification.

[0026] Furthermore, the process of classifying the conference data group into conference topics by the conference data association analysis module through a clustering algorithm includes:

[0027] Z1. Determine the number of conference topics K;

[0028] Z2, randomly select K conference data points from the conference data set as the initial centroids and mark them as μ1,μ2,...,μ K ;

[0029] Z3, for each data point x in the conference data set q , calculate for each data point x q With each initial center of mass μ k The distance E(x q ,μ k ), where q represents the data point index;

[0030] Z4, the data point x q Assign to the initial centroid μ with the smallest distance k The corresponding conference theme set C k In In the formula, k represents the final determination of the data point x q The index of the initial centroid with the smallest distance, and k = 1, 2, ..., K, s represents the index of all other initial centroids, which is used to ensure that the initial centroid with the data point x is found. q The initial centroid with the smallest distance is assigned to the corresponding conference topic set C k middle;

[0031] Z5. Update the initial centroid In the formula, represents the updated initial centroid, and h is the number of iterations;

[0032] Z6, repeating steps Z3 to Z5 until the position of the initial centroid no longer changes significantly or reaches a preset number of iterations h';

[0033] Z7. Use the minimization objective function to represent the clustering process:

[0034]

[0035] In the formula, OF represents the result value of minimizing the objective function.

[0036] Furthermore, the process of determining the subject attribution of the conference data in each conference data group by the conference data association analysis module includes:

[0037] For the text data set of conference content WL={L1,L2,...,L e}, and use the TF-IDF method to convert it into a vector representation. Let the text vector set of the conference content be WV = {V1, V2, ..., V e}; where WV is the vector representation of WL, e is the number of texts in the conference content, L1,L2,...,L e Respectively represent the first text data, the second text data, ..., the eth text data, V1, V2, ..., V e Respectively represent the first text vector, the second text vector, ..., the e-th text vector;

[0038] Combine the text vector set V of the conference content with the conference topic set C k Matching is performed, and the LDA algorithm is used to process and output a conference keyword matrix Φ∈I K×WV and a text topic distribution matrix Θ∈I e×WV , as the feature representation of the meeting content, and then obtain the meeting content feature A = (Φ, Θ);

[0039] It is understandable that in the conference keyword matrix Φ∈I K×WV In the text topic distribution matrix Θ∈I, each row represents a conference topic, each column represents a word in the vocabulary, and the element value in the conference topic word matrix represents the probability of the corresponding word under the topic; e×WV In , each row represents a text data, each column represents a conference topic, and the element value in the text topic distribution matrix represents the probability distribution of the text data on the corresponding topic; I represents the vocabulary size, and K represents the number of conference topics.

[0040] Furthermore, the participant user behavior pattern recognition module extracts the participant user behavior data set from the participant user set and defines the behavior label; the process of mapping the participant user behavior label by training the decision tree classifier to thereby identify the participant user behavior pattern includes:

[0041] The behavioral data set of the participating users is extracted from the participating user set as U = {(u1,y1),(u2,y2),...,(u N ,y N )}, where u represents the ID of the participant user, y represents the set of participant user behavior attributes (such as the number of attendances, the number of speeches, etc.), and N represents the total number of participants;

[0042] Get the behavioral feature set of the participating users F = {f1,f2,...,f M}, where f1,f1,...,f M They respectively represent the behavior characteristic value of the first participating user, the behavior characteristic value of the second participating user, ..., the behavior characteristic value of the Mth participating user, where M represents the total number of the behavior characteristic values ​​of the participating users;

[0043] For each participant in the participant's behavior dataset U, define the participant's behavior feature function Use this function to map the participating users to the behavioral feature vectors of the participating users. On, that is:

[0044]

[0045] In the formula, Q M represents an M-dimensional real vector space, where M is the dimension of the behavioral feature vector of the participating users, and represents the total number of behavioral feature values ​​of the participating users; M (u N ,y N ) represents the participating user u N The behavioral feature value f of the participating users M The value of ;

[0046] Based on the extracted behavioral feature vector of the participants Define a set of behavior tags to represent the behavior patterns of different participants, and obtain the behavior tag set X = {X1, X2, ..., X Y}; where Y represents the number of behavior tags, and each behavior tag represents a specific user behavior pattern;

[0047] Combine the defined behavior labels and the extracted behavior feature vectors of the participating users to construct a training set In the formula, B represents the number of training samples;

[0048] Select a decision tree classifier, train the constructed training set ζ, and map the behavioral feature vectors of the participating users to the corresponding behavioral labels;

[0049] For a new participant u', extract the behavior feature vector of the new participant The trained decision tree classifier is used to classify the new conference user u' and output a new behavior label X', indicating that the behavior pattern of the new conference user u' is closest to X'.

[0050] Furthermore, the conference content push formulation and optimization module calculates and obtains the push weight based on the conference content and the behavioral tags of the participating users, and then pushes the conference content to the participating users, including:

[0051] Obtain the conference content features corresponding to the conference content and the behavior tag set corresponding to the behavior tags of the participating users;

[0052] According to the conference content characteristics and the behavioral tag set of the participating users, the push weight PH between the conference content characteristics and the behavioral characteristics of the participating users is calculated and analyzed. bp :

[0053]

[0054] In the formula, D represents the total number of conference content features, b represents the conference content feature index, G represents the total number of behavioral features of conference participants after the behavioral label classification is completed, p represents the behavioral feature index of conference participants after the behavioral label classification is completed, and A brepresents the content characteristics of the meeting, F p represents the behavioral characteristics of the participating users after completing the behavioral label classification, α b ,β p Respectively represent A b and F p The weight coefficient, σ(A b ,F p ) is the adjustment function, used to adjust the b and F p The value of adjusts the push weight;

[0055] Understandably, when A b and F p When similar themes or behaviors are expressed, then σ(A b ,F p ) can take a larger value to increase the push weight, otherwise, take a smaller value to weaken the push weight;

[0056] After getting the push weight, PH bp , push conference content to participating users based on push weights. Specific push strategies include:

[0057] For each participant, the conference content is pushed to them in descending order of push weight;

[0058] Set a push threshold. If the push weight is greater than the push threshold, the corresponding conference content will be pushed to the participating users. Otherwise, the corresponding conference content will not be pushed.

[0059] Conduct quantitative analysis based on historical feedback from participating users (such as click-through rate, reading time, etc.), and dynamically adjust push weights and push thresholds.

[0060] Furthermore, the process of conducting quantitative analysis based on historical feedback from participating users and dynamically adjusting push weights and push thresholds to optimize push effects includes:

[0061] Obtain the click-through rate, reading time, and number of likes in historical feedback, and calculate the corresponding percentile rankings of the click-through rate, reading time, and number of likes;

[0062] The historical feedback index HFI of the participants is calculated using the formula:

[0063]

[0064] In the formula, HFI represents the historical feedback index of the participating users, which is used to reflect the user's interest and participation in the pushed content; Ra(CR), Ra(RD), and Ra(LC) represent the quantile rankings of click-through rate, reading time, and number of likes, respectively; R represents the total number of samples, which is used to standardize the quantile ranking; η, ψ, and λ are three dynamic learning coefficients, corresponding to the click-through rate, reading time, and number of likes, respectively. They are adaptively adjusted according to the changes in historical data to capture the influence of long-term trends and short-term fluctuations, reflecting the preference trend of participating users that changes over time; c1, c2, and c3 are preset nonlinear parameters, and the optimal values ​​can be determined through experiments to optimize the fitting effect of nonlinear features;

[0065] Adjust the weight and priority of pushed content based on the historical feedback index of participating users:

[0066] Set the push weight update function to dynamically adjust the push weight according to the historical feedback index of the participating users. The specific formula is as follows:

[0067]

[0068] Where PH bp(l,v+1) Indicates the push weight of participant user l at the next time point, PH bp(l,v) represents the push weight of the participant user l at the current time point, δ represents the learning slope, which is used to control the speed of weight update, HFI l represents the historical feedback index of participant l, represents the average historical feedback index of all participating users, and v represents the time point;

[0069] Dynamically adjust the push threshold based on the push weights and historical feedback index distribution of all participating users: Set a push threshold update function to dynamically adjust the push threshold based on the push weights and historical feedback index statistical characteristics at the current time point. The specific formula is as follows:

[0070]

[0071] In the formula, θ v+1 represents the push threshold at the next time point; θ v Indicates the push threshold at the current time point; γ is the adjustment coefficient, which is used to control the amplitude of the threshold update; Indicates the average push weight of all participating users; Represents the standard deviation of the historical feedback index.

[0072] Compared with the existing technology, the advantages of the data information push system and method provided by the present invention based on cloud conference are:

[0073] 1. The present invention extracts a conference data set from a database, processes and sequence-divides the conference data containing time series characteristics, and provides an accurate and ordered conference data set for subsequent analysis, thereby ensuring the quality and availability of the data;

[0074] 2. The present invention analyzes the correlation between conference data by using correlation coefficients, and screens and obtains conference data groups; uses clustering algorithms to classify conference data groups by conference topics, determines the topic attribution of conference data in each conference data group, and achieves accurate description and classification of conference content, providing a scientific basis for subsequent content push, which helps to improve the pertinence and effectiveness of push;

[0075] 3. The present invention extracts the behavioral data set of the participating users from the participating user set and defines the behavioral labels; uses a trained decision tree classifier to map the behavioral labels of the participating users, thereby identifying the user's behavioral pattern and providing a personalized basis for subsequent content push; obtains the push weight through calculation, and then pushes the conference content to the participating users, thereby improving the accuracy and effectiveness of the push, thereby improving the performance of the entire system and user experience.

[0076] To sum up, the present invention realizes accurate and efficient conference information push by comprehensively analyzing the conference content and the behavior patterns of the participants, thereby improving the dissemination effect of conference information and the experience of the participants. At the same time, the present invention can also dynamically adjust the push strategy according to the historical feedback of the participants, further enhancing the pertinence and effectiveness of the push, and ensuring the efficient and stable operation of a subsequent data information push system based on cloud conferences. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 This is a module diagram of a data information push system based on cloud conferencing proposed by the present invention. DETAILED DESCRIPTION

[0078] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the implementation regulations described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0079] Reference Figure 1 , a data information push system based on cloud conference, the system includes a data processing and sequence division module, a conference data association analysis module, a participant user behavior pattern recognition module and a conference content push formulation and optimization module;

[0080] Data processing and sequence division module: extracts meeting data sets from the database, processes and sequence divides meeting data containing time series characteristics for subsequent analysis; the meeting data sets include meeting timestamps, meeting user sets, and meeting content;

[0081] Conference data association analysis module: Use correlation coefficients to analyze the association between conference data and filter and obtain conference data groups; use clustering algorithms to classify conference data groups by conference topics and determine the topic attribution of conference data in each conference data group;

[0082] Participant user behavior pattern recognition module: extract the participant user behavior data set from the participant user set and define the behavior label; map the participant user behavior label by training the decision tree classifier, so as to identify the user behavior pattern;

[0083] Conference content push formulation and optimization module: Based on the conference content and the behavioral tags of the participating users, the push weight is calculated and obtained, and then the conference content is pushed to the participating users.

[0084] It should be noted that the application object of a data information push system based on cloud conference in an embodiment of the present invention can be a conference scenario in the field of information management and push. Specifically, it can be to perform comprehensive analysis and data processing of conference data through big data analysis technology to realize the subject classification of conference content and the identification of user behavior patterns. Through data processing and sequence division, conference data correlation analysis, behavioral pattern identification of participating users, and push formulation and optimization and other links, a highly integrated and intelligent data information precision push system is formed, so that the system can realize the precise push of conference content according to the characteristics of conference content and the behavioral patterns of participating users, and dynamically adjust the push strategy according to user historical feedback to optimize the push effect.

[0085] The data processing and sequence division module extracts the conference data set from the database, and processes and sequence divides the conference data containing time series characteristics, including the following steps:

[0086] Step 101: extract the meeting data set from the database and mark it as

[0087] d g ={d1,d2,...,d n};

[0088] Among them, d g represents the g-th meeting data, including the meeting timestamp, the set of participating users (such as ID, name, user behavior records, etc.) and the meeting content. g represents the index of the meeting data, g = 1, 2, ..., n, n represents the total number of meeting data, d1, d2, ..., d nRespectively represent the first meeting data, the second meeting data, ..., the nth meeting data;

[0089] Step 102: For conference data containing time series characteristics (such as conference timestamps), use sliding window technology to process:

[0090] Define the sliding window size as w, which is the length of each subsequence into which the conference data is divided;

[0091] Then there is a time series data set T = {t1, t2, ..., t m} and is divided into multiple time subsequences

[0092] T r ={t (r-1)w+1 ,t (r-1)w+2 ,...,t rw}, where

[0093] Where m represents the index of the time series, r represents the index of the time subsequence, t1, t2, ..., t m Respectively represent the first time series data, the second time series data, ..., the mth time series data, t (r-1)w+1 ,t (r-1)w+2 ,...,t rw They represent the (r-1)w+1th time subsequence, the (r-1)w+2th time subsequence, ..., the rwth time subsequence, Express Round up.

[0094] The conference data association analysis module uses the correlation coefficient to analyze the association between conference data and screens to obtain conference data groups; the conference data groups are classified into conference themes by clustering algorithms, and the steps of determining the theme attribution of conference data in each conference data group include:

[0095] Step 201: For any two conference data, d i ,d j , using Pearson correlation coefficient PCR ij Analyze meeting data i and d j The correlation between them; the calculation formula of Pearson correlation coefficient is:

[0096]

[0097] In the formula, PCR ij represents the Pearson correlation coefficient, i and j represent the indexes of any two conference data, and a 1H ,a 2HCorresponding to any two conference data d i and d j The Hth numeric attribute of Represents any two conference data d i and d j , H represents the index of the numerical attribute;

[0098] Step 202: Analyze the correlation between the conference data based on the calculation result of the Pearson correlation coefficient. If PCR ij >0, then any two conference data are judged to be positively correlated; if PCR ij <0, then any two conference data are judged to be negatively correlated; if |PCR ij |The closer it is to 1, the stronger the correlation between the two conference data is; if PCR ij =0, it is judged that there is no linear relationship between the two conference data;

[0099] Step 203, screen out any two conference data with positive correlation coefficients, and aggregate them into a conference data group for conference topic classification;

[0100] Step 204: Classifying the conference data group by conference topic using a clustering algorithm is as follows:

[0101] Z1. Determine the number of conference topics K;

[0102] Z2, randomly select K conference data points from the conference data set as the initial centroids and mark them as μ1,μ2,...,μ K ;

[0103] Z3, for each data point x in the conference data set q , calculate for each data point x q With each initial center of mass μ k The distance E(x q ,μ k ), where q represents the data point index;

[0104] Z4, the data point x q Assign to the initial centroid μ with the smallest distance k The corresponding conference theme set C k In In the formula, k represents the final determination of the data point x q The index of the initial centroid with the smallest distance, and k = 1, 2, ..., K, s represents the index of all other initial centroids, which is used to ensure that the initial centroid with the data point x is found. q The initial centroid with the smallest distance is assigned to the corresponding conference topic set C k middle;

[0105] Z5. Update the initial centroid In the formula, represents the updated initial centroid, and h is the number of iterations;

[0106] Z6, repeating steps Z3 to Z5 until the position of the initial centroid no longer changes significantly or reaches a preset number of iterations h';

[0107] Z7. Use the minimization objective function to represent the clustering process:

[0108]

[0109] In the formula, OF represents the result value of minimizing the objective function;

[0110] Step 205, the steps of determining the subject attribution of the conference data in each conference data group are as follows:

[0111] For the text data set of conference content WL={L1,L2,...,L e}, and use the TF-IDF method to convert it into a vector representation. Let the text vector set of the conference content be WV = {V1, V2, ..., V e}; where WV is the vector representation of WL, e is the number of texts in the conference content, L1,L2,...,L e Respectively represent the first text data, the second text data, ..., the eth text data, V1, V2, ..., V e Respectively represent the first text vector, the second text vector, ..., the e-th text vector;

[0112] Combine the text vector set V of the conference content with the conference topic set C k Matching is performed, and the LDA algorithm is used to process and output a conference keyword matrix Φ∈I K×WV and a text topic distribution matrix Θ∈I e×WV , as the feature representation of the meeting content, and then obtain the meeting content feature A = (Φ, Θ);

[0113] In step 205, in the conference keyword matrix Φ∈I K×WV In the text topic distribution matrix Θ∈I, each row represents a conference topic, each column represents a word in the vocabulary, and the element value in the conference topic word matrix represents the probability of the corresponding word under the topic; e×WV In , each row represents a text data, each column represents a conference topic, and the element value in the text topic distribution matrix represents the probability distribution of the text data on the corresponding topic; I represents the vocabulary size, and K represents the number of conference topics.

[0114] The participant user behavior pattern recognition module extracts the participant user behavior data set from the participant user set and defines the behavior label; the steps of mapping the participant user behavior label by training the decision tree classifier to identify the user behavior pattern include:

[0115] Step 301: Extract the behavior data set of the participating users from the participating user set as U = {(u1, y1), (u2, y2), ..., (u N ,y N )}, where u represents the ID of the participant user, y represents the set of participant user behavior attributes (such as the number of attendances, the number of speeches, etc.), and N represents the total number of participants;

[0116] Step 302: Obtain the behavioral feature set F of the participating users = {f1, f2, ..., f M}, where f1,f1,...,f M They respectively represent the behavior characteristic value of the first participating user, the behavior characteristic value of the second participating user, ..., the behavior characteristic value of the Mth participating user, where M represents the total number of the behavior characteristic values ​​of the participating users;

[0117] Step 303: For each participant in the participant's behavior data set U, define the participant's behavior feature function Use this function to map the participating users to the behavioral feature vectors of the participating users. On, that is:

[0118]

[0119] In the formula, Q M represents an M-dimensional real vector space, where M is the dimension of the behavioral feature vector of the participating users, and represents the total number of behavioral feature values ​​of the participating users; M (u N ,y N ) represents the participating user u N The behavioral feature value f of the participating users M The value of ;

[0120] Step 304: Based on the extracted behavioral feature vector of the participating users Define a set of behavior tags to represent the behavior patterns of different participants, and obtain the behavior tag set X = {X1, X2, ..., X Y}; where Y represents the number of behavior tags, and each behavior tag represents a specific user behavior pattern;

[0121] Step 305: Combine the defined behavior labels and the extracted behavior feature vectors of the participating users to construct a training set In the formula, B represents the number of training samples;

[0122] Step 306: Select a decision tree classifier, train the constructed training set ζ, and map the behavior feature vectors of the participating users to corresponding behavior labels;

[0123] Step 307: For a new conference participant u', extract the behavior feature vector of the new conference participant The trained decision tree classifier is used to classify the new conference user u' and output a new behavior label X', indicating that the behavior pattern of the new conference user u' is closest to X'.

[0124] The conference content push formulation and optimization module calculates the push weight based on the conference content and the behavioral tags of the participating users, and then pushes the conference content to the participating users in the following steps:

[0125] Step 401: Obtain conference content features corresponding to the conference content and a behavior tag set corresponding to the behavior tags of the participating users;

[0126] Step 402: Calculate and analyze the push weight PH between the conference content features and the behavioral features of the conference users according to the conference content features and the behavioral tags of the conference users. bp :

[0127]

[0128] In the formula, D represents the total number of conference content features, b represents the conference content feature index, G represents the total number of behavioral features of conference participants after the behavioral label classification is completed, p represents the behavioral feature index of conference participants after the behavioral label classification is completed, and A b represents the content characteristics of the meeting, F p represents the behavioral characteristics of the participating users after completing the behavioral label classification, α b ,β p Respectively represent A b and F p The weight coefficient, σ(A b ,F p ) is the adjustment function, used to adjust the b and F p The value of adjusts the push weight;

[0129] In step 402, when A b and F p When similar themes or behaviors are expressed, then σ(A b ,F p) can take a larger value to increase the push weight, otherwise, take a smaller value to weaken the push weight;

[0130] Step 403: After getting the push weight, PH bp , push conference content to participating users based on push weights. Specific push strategies include:

[0131] For each participant, the conference content is pushed to them in descending order of push weight;

[0132] Set a push threshold. If the push weight is greater than the push threshold, the corresponding conference content will be pushed to the participating users. Otherwise, the corresponding conference content will not be pushed.

[0133] Conduct quantitative analysis based on historical feedback from participating users (such as click-through rate, reading time, etc.), and dynamically adjust push weights and push thresholds to optimize push effects;

[0134] The steps for conducting quantitative analysis based on historical feedback from participating users and dynamically adjusting push weights and push thresholds are as follows:

[0135] Obtain the click-through rate, reading time, and number of likes in historical feedback, and calculate the corresponding percentile rankings of the click-through rate, reading time, and number of likes;

[0136] The historical feedback index HFI of the participants is calculated using the formula:

[0137]

[0138] In the formula, HFI represents the historical feedback index of the participating users, which is used to reflect the user's interest and participation in the pushed content; Ra(CR), Ra(RD), and Ra(LC) represent the quantile rankings of click-through rate, reading time, and number of likes, respectively; R represents the total number of samples, which is used to standardize the quantile ranking; η, ψ, and λ are three dynamic learning coefficients, corresponding to the click-through rate, reading time, and number of likes, respectively. They are adaptively adjusted according to the changes in historical data to capture the influence of long-term trends and short-term fluctuations, reflecting the preference trend of participating users that changes over time; c1, c2, and c3 are preset nonlinear parameters, and the optimal values ​​can be determined through experiments to optimize the fitting effect of nonlinear features;

[0139] Adjust the weight and priority of pushed content based on the historical feedback index of participating users:

[0140] Set the push weight update function to dynamically adjust the push weight according to the historical feedback index of the participating users. The specific formula is as follows:

[0141]

[0142] Where PH bp(l,v+1) Indicates the push weight of participant user l at the next time point, PH bp(l,v) represents the push weight of the participant user l at the current time point, δ represents the learning slope, which is used to control the speed of weight update, HFI l represents the historical feedback index of participant l, represents the average historical feedback index of all participating users, and v represents the time point;

[0143] Dynamically adjust the push threshold based on the push weights and historical feedback index distribution of all participating users: Set a push threshold update function to dynamically adjust the push threshold based on the push weights and historical feedback index statistical characteristics at the current time point. The specific formula is as follows:

[0144]

[0145] In the formula, θ v+1 represents the push threshold at the next time point; θ v Indicates the push threshold at the current time point; γ is the adjustment coefficient, which is used to control the amplitude of the threshold update; Indicates the average push weight of all participating users; Represents the standard deviation of the historical feedback index.

[0146] In the embodiment of the present invention, by extracting a conference data set, and processing and sequence dividing the conference data containing time series characteristics, an accurate and orderly data basis is provided for subsequent analysis, by using the correlation coefficient to analyze the correlation between conference data, and screening to obtain conference data groups, using clustering algorithms to classify conference data groups by conference topics, and at the same time determining the subject attribution of conference data in each conference data group, providing strong support for conference content push, and realizing accurate feature representation of conference content, by extracting the behavioral data set of participating users, and defining behavioral labels, and at the same time using training decision tree classifiers to map the behavioral labels of participating users, effectively identifying the behavioral patterns of participating users, and providing important basis for personalized conference content push, by combining conference content features and participating users' behavioral patterns, calculating push weights, and dynamically adjusting push strategies according to historical feedback from participating users, accurate and efficient conference content push is realized. In summary, the example of the present invention solves the problem of incomplete content association and user behavior analysis in current information management push. In actual situations, more data and context information may be needed to make specific decisions and optimization plans.

[0147] In addition, the formulas involved in the above are all calculated by removing dimensions and taking their numerical values. They are a formula that is closest to the actual situation obtained by collecting a large amount of data and performing software simulation. The proportional coefficient in the formula and the various preset thresholds in the analysis process are set by technical personnel in this field according to actual conditions or obtained by simulating a large amount of data; the size of the proportional coefficient is to quantify each parameter to obtain a specific value for subsequent comparison. The size of the proportional coefficient depends on the amount of sample data and the preliminary setting of the corresponding processing coefficient for each group of sample data by technical personnel in this field; as long as it does not affect the proportional relationship between the parameter and the quantized value.

[0148] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically based on the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0149] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0150] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0151] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0152] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0153] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0154] Secondly: In the drawings of the embodiments disclosed in the present invention, only the structures related to the embodiments disclosed in the present invention are involved, and other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of the present invention can be combined with each other;

[0155] Finally: The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A data information push system based on cloud conference, characterized by: It includes data processing and sequence division module, conference data association analysis module, participant behavior pattern recognition module, and conference content push formulation and optimization module; Data processing and sequence division module: extracts the meeting data set from the database, and processes and sequence divides the meeting data containing time series characteristics; the meeting data set includes the meeting timestamp, the meeting user set and the meeting content; Conference data association analysis module: Use correlation coefficients to analyze the association between conference data and filter and obtain conference data groups; use clustering algorithms to classify conference data groups by conference topics and determine the topic attribution of conference data in each conference data group; Participant user behavior pattern recognition module: extract the participant user behavior data set from the participant user set and define the behavior label; map the participant user behavior label by training the decision tree classifier, so as to identify the user behavior pattern; Conference content push formulation and optimization module: Based on the conference content and the behavioral tags of the participating users, the push weight is calculated and obtained, and then the conference content is pushed to the participating users.

2. According to claim 1, a data information push system based on cloud conference is characterized in that: The data processing and sequence division module extracts the conference data set from the database, and processes and sequence divides the conference data containing time series characteristics. The process includes: Extract the meeting data set from the database and mark it as d g ={d1,d2,...,d n }; Among them, d g represents the g-th meeting data, including the meeting timestamp, the set of participating users, and the meeting content. g represents the index of the meeting data, g = 1, 2, ..., n, n represents the total number of meeting data, d1, d2, ..., d n Respectively represent the first meeting data, the second meeting data, ..., the nth meeting data; For conference data containing time series characteristics, the sliding window technology is used for processing: Define the sliding window size as w, then there is a time series data set T = {t1, t2, ..., t m }, and is divided into multiple time subsequences; T r ={t (r-1)w+1 ,t (r-1)w+2 ,…,t rw }, where Where m represents the index of the time series, r represents the index of the time subsequence, t1, t2, ..., t m Respectively represent the first time series data, the second time series data, ..., the mth time series data, t (r-1)w+1 ,t (r-1)w+2 ,...,t rw They represent the (r-1)w+1th time subsequence, the (r-1)w+2th time subsequence, ..., the rwth time subsequence, Express Round up.

3. According to claim 2, a data information push system based on cloud conference is characterized in that: The conference data correlation analysis module uses the correlation coefficient to analyze the correlation between conference data, and the process of screening and obtaining conference data groups includes: For any two conference data, d i ,d j , using Pearson correlation coefficient PCR ij Analyze meeting data i and d j The correlation between them; the calculation formula of Pearson correlation coefficient is: In the formula, PCR ij represents the Pearson correlation coefficient, i and j represent the indexes of any two conference data, and a 1H ,a 2H Corresponding to any two conference data d i and d j The Hth numeric attribute of Represents any two conference data d i and d j , H represents the index of the numerical attribute; According to the calculation results of Pearson correlation coefficient, the correlation between conference data is analyzed. ij >0, then any two conference data are judged to be positively correlated; if PCR ij <0, then any two conference data are judged to be negatively correlated; if |PCR ij |The closer it is to 1, the stronger the correlation between the two conference data is; if PCR ij =0, it is judged that there is no linear relationship between the two conference data; Screen out any two conference data with positive correlation coefficients and aggregate them into a conference data group for conference topic classification.

4. According to claim 1, a data information push system based on cloud conference is characterized in that: The process of the conference data association analysis module classifying conference data groups into conference topics through clustering algorithms includes: Z1. Determine the number of conference topics K; Z2, randomly select K conference data points from the conference data set as the initial centroids and mark them as μ1,μ2,...,μ K ; Z3, for each data point x in the conference data set q , calculate for each data point x q With each initial center of mass μ k The distance E(x q ,μ k ), where q represents the data point index; Z4, the data point x q Assign to the initial centroid μ with the smallest distance k The corresponding conference theme set C k In In the formula, k represents the final determination of the data point x q The index of the initial centroid with the smallest distance, and k = 1, 2, ..., K, s represents the index of all other initial centroids, which is used to ensure that the initial centroid with the data point x is found. q The initial centroid with the smallest distance is assigned to the corresponding conference topic set C k middle; Z5. Update the initial centroid In the formula, represents the updated initial centroid, and h is the number of iterations; Z6, repeating steps Z3 to Z5 until the position of the initial centroid no longer changes significantly or reaches a preset number of iterations h'; Z7. Use the minimization objective function to represent the clustering process: In the formula, OF represents the result value of minimizing the objective function.

5. A data information push system based on cloud conference according to claim 4, characterized in that: The process of determining the subject attribution of conference data in each conference data group by the conference data association analysis module includes: For the text data set of conference content WL={L1,L2,...,L e }, and use the TF-IDF method to convert it into a vector representation. Suppose the text vector set of the conference content is WV = {V1, V2, ..., V e }; where WV is the vector representation of WL, e is the number of texts in the conference content, L1,L2,…,L e Respectively represent the first text data, the second text data, ..., the eth text data, V1, V2, ..., V e Respectively represent the first text vector, the second text vector, ..., the e-th text vector; Combine the text vector set V of the conference content with the conference topic set C k Matching is performed, and the LDA algorithm is used to process and output a conference keyword matrix Φ∈I K×WV and a text topic distribution matrix Θ∈I e×WV , as the feature representation of the meeting content, and then obtain the meeting content feature A = (Φ, Θ); Among them, in the conference keyword matrix Φ∈I K×WV In the text topic distribution matrix Θ∈I, each row represents a conference topic, each column represents a word in the vocabulary, and the element value in the conference topic word matrix represents the probability of the corresponding word under the topic; e×WV In , each row represents a text data, each column represents a conference topic, and the element value in the text topic distribution matrix represents the probability distribution of the text data on the corresponding topic; I represents the vocabulary size, and K represents the number of conference topics.

6. A data information push system based on cloud conference according to claim 5, characterized in that: The participant user behavior pattern recognition module extracts the participant user behavior data set from the participant user set and defines the behavior label; The process of mapping the behavior labels of conference participants by training a decision tree classifier and identifying the behavior patterns of conference participants includes: The behavioral data set of the participating users is extracted from the participating user set as U = {(u1,y1),(u2,y2),...,(u N ,y N )}, where u represents the ID of the participating user, y represents the set of participating user behavior attributes, and N represents the total number of participating users; Get the behavioral feature set of the participating users F = {f1,f2,…,f M }, where f1,f1,…,f M They represent the behavior feature value of the first participating user, the behavior feature value of the second participating user, ..., the behavior feature value of the Mth participating user, where M represents the total number of behavior feature values ​​of participating users; For each participant in the participant's behavior dataset U, define the participant's behavior feature function Use this function to map the participating users to the behavioral feature vectors of the participating users. On, that is: In the formula, Q M represents an M-dimensional real vector space, where M is the dimension of the behavioral feature vector of the participating users, and represents the total number of behavioral feature values ​​of the participating users; M (u N ,y N ) represents the participating user u N The behavioral feature value f of the participating users M The value of ; Based on the extracted behavioral feature vector of the participants Define a set of behavior tags to represent the behavior patterns of different participants, and obtain the behavior tag set X = {X1, X2, …, X Y }; where Y represents the number of behavior labels; Combine the defined behavior labels and the extracted behavior feature vectors of the participating users to construct a training set In the formula, B represents the number of training samples; Select a decision tree classifier, train the constructed training set ζ, and map the behavioral feature vectors of the participating users to the corresponding behavioral labels; For a new participant u', extract the behavior feature vector of the new participant The trained decision tree classifier is used to classify the new conference user u' and output a new behavior label X', indicating that the behavior pattern of the new conference user u' is closest to X'.

7. A data information push system based on cloud conference according to claim 6, characterized in that: The conference content push formulation and optimization module calculates the push weight based on the conference content and the behavioral tags of the participating users, and then pushes the conference content to the participating users. The process includes: Obtain the conference content features corresponding to the conference content and the behavior tag set corresponding to the behavior tags of the participating users; According to the conference content characteristics and the behavioral tag set of the participating users, the push weight PH between the conference content characteristics and the behavioral characteristics of the participating users is calculated and analyzed. bp : In the formula, D represents the total number of conference content features, b represents the conference content feature index, G represents the total number of behavioral features of conference participants after the behavioral label classification is completed, p represents the behavioral feature index of conference participants after the behavioral label classification is completed, and A b represents the content characteristics of the meeting, F p represents the behavioral characteristics of the participating users after completing the behavioral label classification, α b ,β p Respectively represent A b and F p The weight coefficient, σ(A b ,F p ) is the adjustment function, used to adjust the b and F p The value of adjusts the push weight; After getting the push weight, PH bp , push conference content to participating users based on push weights. Specific push strategies include: For each participant, the conference content is pushed to them in descending order of push weight; Set a push threshold. If the push weight is greater than the push threshold, the corresponding conference content will be pushed to the participating users. Otherwise, the corresponding conference content will not be pushed. Conduct quantitative analysis based on historical feedback from participating users and dynamically adjust push weights and push thresholds.

8. A data information push system based on cloud conference according to claim 7, characterized in that: The process of quantitatively analyzing the historical feedback from conference users and dynamically adjusting the push weight and push threshold includes: Obtain the click-through rate, reading time, and number of likes in historical feedback, and calculate the corresponding percentile rankings of the click-through rate, reading time, and number of likes; The historical feedback index HFI of the participants is calculated using the formula: In the formula, HFI represents the historical feedback index of the participating users, which is used to reflect the user's interest and participation in the pushed content; Ra(CR), Ra(RD), and Ra(LC) represent the quantile rankings of click-through rate, reading time, and number of likes, respectively; R represents the total number of samples, which is used to standardize the quantile ranking; η, ψ, and λ are three dynamic learning coefficients, corresponding to click-through rate, reading time, and number of likes, respectively; c1, c2, and c3 are preset nonlinear parameters; Adjust the weight and priority of pushed content based on the historical feedback index of participating users: Set the push weight update function. The specific formula is as follows: Where PH bp(l,v+1) Indicates the push weight of participant user l at the next time point, PH bp(l,v) represents the push weight of the participant user l at the current time point, δ represents the learning slope, which is used to control the speed of weight update, and HFI l represents the historical feedback index of participant l, represents the average value of the historical feedback index of all participating users, and v represents the time point; Dynamically adjust the push threshold based on the push weights of all participating users and the distribution of historical feedback indexes: Set the push threshold update function. The specific formula is as follows: In the formula, θ v+1 represents the push threshold at the next time point; θ v Indicates the push threshold at the current time point; γ is the adjustment coefficient, which is used to control the amplitude of threshold update; Indicates the average push weight of all participating users; Represents the standard deviation of the historical feedback index.

Images