A cloud conference-based data information pushing system

Abstract

The application relates to the technical field of big data analysis, in particular to a data information pushing system based on cloud conferences, which comprises a data processing and sequence division module, a conference data correlation analysis module, a participant user behavior pattern recognition module and a conference content pushing formulation and optimization module; through extraction and processing of conference data, accurate and orderly data bases are provided for subsequent analysis; through correlation coefficients and clustering algorithms, the correlation among conference data is deeply mined, conference themes are accurately divided, and strong support is provided for conference content pushing; through definition of behavior tags and training of decision tree classifiers, behavior patterns of participant users are effectively recognized, and important bases are provided for personalized conference content pushing; through calculation of pushing weights and adjustment of pushing strategies, precise and efficient conference content pushing is realized. The application is used for solving the technical problem that content correlation and user behavior analysis are not comprehensive in current information management pushing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data analysis, and particularly relates to a data information pushing system based on cloud conference. BACKGROUND

[0002] Big data analysis technology refers to using advanced computer technology and tools to collect, store, process and analyze massive data to find the rules and values hidden in the data, and mainly includes technologies such as data preprocessing, data mining, data visualization, data analysis and machine learning.

[0003] In the current information management and pushing field, the following problems still exist: conference data set processing is insufficient, conference data is huge, and with the large increase of data, it may not be possible to accurately process the data due to factors such as data diversity, resulting in insufficient pushing reliability; the analysis of the behavior of participating users is not comprehensive, which makes it impossible to push personalized content according to user behavior; the pushing strategy lacks flexibility, and traditional information pushing strategies may be relatively single, lacking the flexibility of dynamic adjustment according to user feedback, thereby affecting the pushing effect and user satisfaction. Therefore, the present application proposes a data information pushing system based on cloud conference. SUMMARY

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

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0006] A data information pushing system based on cloud conference, comprising: a data processing and sequence division module, a conference data correlation analysis module, a participating user behavior pattern recognition module and a conference content pushing formulation and optimization module;

[0007] The data processing and sequence division module extracts a conference data set from a database and processes and sequence divides conference data containing time sequence characteristics for subsequent analysis; wherein the conference data set includes conference timestamps, a set of participating users and conference content;

[0008] The conference data correlation analysis module analyzes the correlation between conference data using a correlation coefficient and filters to obtain a conference data group; a clustering algorithm is used to classify the conference theme of the conference data group and determine the theme attribution of the conference data in each conference data group;

[0009] The participating user behavior pattern recognition module extracts a behavior data set of participating users from a set of participating users and defines a behavior label; a decision tree classifier is trained to map the behavior label of participating users, thereby recognizing the behavior pattern of users;

[0010] Meeting content push design and optimization module: Based on the meeting content and the behavioral tags of the participants, the push weight is calculated and then the meeting content is pushed to the participants.

[0011] The behavior patterns of conference participants are further analyzed. Specifically, the data processing and sequence segmentation module extracts the conference data set from the database and processes and segments the conference data containing time-series characteristics.

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

[0013] d g ={d1,d2,…,d n};

[0014] Where, d g This represents the g-th meeting data entry, which includes 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, where n represents the total number of meeting data entries, and d1, d2, ..., dn. n Let each represent the first meeting data entry, the second meeting data entry, ..., the nth meeting data entry, respectively.

[0015] For meeting data containing time-series characteristics (such as meeting timestamps), a sliding window technique 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 we have a time series data set T = {t1, t2, ..., t} m} and was 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 Let t represent the first time series data, the second time series data, ..., the m-th time series data, respectively. (r-1)w+1 ,t (r-1)w+2 ,...,t rw Let r and w represent the (r-1)w+1th time subsequence, the (r-1)w+2th time subsequence, ..., the rwth time subsequence, respectively. indicates that the is rounded up.

[0020] Further, the conference data correlation analysis module analyzes the correlation between the conference data by using the correlation coefficient, and the process of screening the conference data group comprises:

[0021] For any two conference data, i.e. i , d j , the Pearson correlation coefficient PCR ij is used to analyze the correlation between the conference data d i and d j ; wherein the calculation formula of the Pearson correlation coefficient is:

[0022]

[0023] In the formula, PCR ij indicates the Pearson correlation coefficient, i and j indicate the indexes of any two conference data, a 1H , a 2H correspond to the Hth numerical attribute of any two conference data d i and d j , respectively, and H indicates the index of the numerical attribute. i , j respectively indicate the mean values of any two conference data d i and d j ;

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

[0025] The conference data whose correlation coefficient is positively correlated is screened out from any two conference data, and is collected as a conference data group, which is used for conference theme classification.

[0026] Further, the process of conference theme classification of the conference data group by the conference data correlation analysis module through the clustering algorithm comprises:

[0027] Z1, determine the number of conference themes K;

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

[0029] Z3, for each data point x in the conference data set q Calculate x for each data point q With each initial centroid μ k Distance E(x) q ,μ k ); where q represents the data point index;

[0030] Z4, transfer data point x q Assigned to the initial centroid μ with the smallest distance k The corresponding set of conference topics C k In, that is In the formula, k represents the final determination of the data point x. q The index of the initial centroid with the smallest distance, where k = 1, 2, ..., K, and s represents the indices of all other initial centroids, used to ensure that a centroid with the smallest distance is found relative to the data point x. q Find the initial centroid with the smallest distance and assign it to the corresponding set of conference topics C. k middle;

[0031] Z5. Update the initial centroid. In the formula, The initial centroid is represented by h, which is the number of iterations.

[0032] Z6. Repeat steps Z3 to Z5 until the position of the initial centroid no longer changes significantly or the preset number of iterations h' is reached;

[0033] Z7. Represent the clustering process using a minimization objective function:

[0034]

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

[0036] Furthermore, the process by which the conference data correlation analysis module determines the topic affiliation of conference data within each conference data group includes:

[0037] For the text data set of the meeting content WL={L1,L2,...,L... e The text vectors of the meeting content are converted into vector representations using the TF-IDF method. Let the set of text vectors of the meeting content be WV = {V1, V2, ..., V}. e}; where WV is the vector representation of WL, e represents the number of texts in the meeting content, and L1, L2, ..., L e V1, V2, ..., Ve represent the first text data, the second text data, ..., the e-th text data, respectively. e Let represent the 1st text vector, the 2nd text vector, ..., the eth text vector, respectively;

[0038] Combine the set of text vectors V of the meeting content with the set of meeting topics C. k The matching is performed, and the LDA algorithm is used to process the data, outputting a conference topic term matrix Φ∈I. K×WV A text topic distribution matrix Θ∈I e×WV As a feature representation of the meeting content, we can obtain the meeting content feature A=(Φ,Θ);

[0039] Understandably, in the conference topic term matrix Φ∈I K×WV In the text topic distribution matrix, each row represents a conference topic, and each column represents a word from the vocabulary. The element values ​​in the conference topic term matrix represent the probability of the corresponding word under that topic; in the text topic distribution matrix Θ∈I e×WV In the matrix, each row represents a text data point, each column represents a conference topic, and the element values ​​in the text topic distribution matrix represent 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 behavior pattern recognition module extracts the participant behavior dataset from the participant set and defines behavior labels; the process of mapping participant behavior labels to participant behavior labels by training a decision tree classifier, thereby recognizing participant behavior patterns, includes:

[0041] The behavioral dataset of the attendees is extracted from the set of attendees and is defined as U = {(u1,y1),(u2,y2),...,(u...}. N ,y N In the formula, u represents the participant user ID, y represents the set of participant user behavior attributes (such as the number of times they attended the meeting, the number of times they spoke, etc.), and N represents the total number of participants.

[0042] Obtain the behavioral feature set F = {f1, f2, ..., f...} of the participants. M}, where f1, f1, ..., f M Let $\mathbf$ represent the behavioral characteristics of the first participating user, $\mathbf$ the behavioral characteristics of the second participating user, $\mathbf$ the behavioral characteristics of the Mth participating user, and so on, where $M$ represents the total number of behavioral characteristics of the participating users.

[0043] For each participant in the participant behavior dataset U, define the participant behavior feature function. This function maps attendees to their behavioral feature vectors. Above, that is:

[0044]

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

[0046] Based on the extracted behavioral feature vectors of participating users Define a set of behavioral labels to represent the behavioral patterns of different participants, and obtain the behavioral label 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] A training set is constructed by combining the defined behavioral labels and the extracted behavioral feature vectors of the participating users. In the formula, B represents the number of training samples;

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

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

[0050] Furthermore, the meeting content delivery planning and optimization module calculates and obtains delivery weights based on meeting content and the behavioral tags of participating users, and then delivers meeting content to participating users in the following process:

[0051] Obtain the meeting content features corresponding to the meeting content and the set of behavioral tags corresponding to the behavioral tags of the participating users;

[0052] Based on the characteristics of the meeting content and the behavioral tags of the participants, the push weight PH between the meeting content characteristics and the behavioral characteristics of the participants is calculated and analyzed. bp :

[0053]

[0054] In the formula, D represents the total number of meeting content features, b represents the meeting content feature index, G represents the total number of behavioral features of participating users after completing behavioral tag classification, p represents the behavioral feature index of participating users after completing behavioral tag classification, and A brepresenting the conference content feature, F p representing the behavior feature of the conference user after the completion of the behavior label classification, a b p representing the weight coefficient of A b and F p , respectively, and σ(A b , F p ) is an adjustment function for adjusting the push weight according to the values of A b and F p ;

[0055] It can be understood that when A b and F p represent similar topics or behaviors, then σ(A b , F p ) can take a larger value to enhance the push weight, and vice versa, to weaken the push weight;

[0056] After obtaining the push weight, PH bp , the conference content is pushed to the conference user according to the push weight, and the specific push strategy includes:

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

[0058] A push threshold is set, and if the push weight is greater than the push threshold, the corresponding conference content is pushed to the conference user, and vice versa, the corresponding conference content is not pushed;

[0059] According to the historical feedback of the conference user, quantitative analysis (such as click rate, reading time, etc.) is performed to dynamically adjust the push weight and the push threshold.

[0060] Further, according to the historical feedback of the conference user, quantitative analysis is performed to dynamically adjust the push weight and the push threshold to optimize the push effect, which includes:

[0061] The click rate, reading time, and like number in the historical feedback are obtained, and the quantile rank corresponding to the click rate, reading time, and like number is calculated;

[0062] The historical feedback index HFI of the conference user is calculated by using the formula:

[0063]

[0064] ​In the formula, HFI represents the historical feedback index of the participating user, which is used to reflect the interest and participation of the user in the push content; Ra(CR), Ra(RD), and Ra(LC) represent the quantile ranking of the click rate, reading time, and like number, respectively; R represents the total number of samples, which is used for the normalization of the quantile ranking; η, ψ, and λ are three dynamic learning coefficients corresponding to the click rate, reading time, and like number, respectively, which are adaptively adjusted according to the changes of historical data to capture the influence of long-term trends and short-term fluctuations and reflect the preference trend of the participating user changing 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 the nonlinear characteristics;

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

[0066] Set a push weight update function for dynamically adjusting the push weight according to the historical feedback index of the participating user, and the specific formula is as follows:

[0067]

[0068] In the formula, PH bp(l,v+1) represents the push weight of the participating user l at the next time point, PH bp(l,v) represents the push weight of the participating user l at the current time point, δ represents the slope for controlling the speed of weight update, HFI l represents the historical feedback index of the participating user l, represents the average value of the historical feedback indexes of all participating users, and v represents the time point.

[0069] Adjust the push threshold dynamically in combination with the distribution of the push weight and the historical feedback index of all participating users: set a push threshold update function for dynamically adjusting the push threshold according to the statistical characteristics of the push weight and the historical feedback index at the current time point, and 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 represents the push threshold at the current time point, and γ is an adjustment coefficient for controlling the amplitude of threshold update. represents the average value of the push weights of all participating users. represents the standard deviation of the historical feedback index.

[0072] Compared with the existing technology, the application provides a data information push system and method based on a cloud conference, and the advantages are as follows:

[0073] 1. This invention extracts a set of meeting data from a database and processes and partitions the meeting data containing time-series characteristics to provide an accurate and orderly set of meeting data for subsequent analysis, ensuring the quality and availability of the data;

[0074] 2. This invention analyzes the correlation between meeting data using correlation coefficients and filters to obtain meeting data groups; it uses clustering algorithms to classify the meeting data groups by meeting theme, determines the theme affiliation of the meeting data in each meeting data group, and achieves accurate description and classification of meeting content, providing a scientific basis for subsequent content push and helping to improve the targeting and effectiveness of push.

[0075] 3. This invention extracts the behavioral dataset of participating users from the set of participating users and defines behavioral labels; it uses a trained decision tree classifier to map the behavioral labels of participating users, thereby identifying user behavior patterns and providing a personalized basis for subsequent content push; it calculates and obtains push weights, and then pushes meeting content to participating users, improving the accuracy and effectiveness of push, thereby improving the performance of the entire system and user experience.

[0076] In summary, this invention achieves precise and efficient meeting information delivery by comprehensively analyzing meeting content and participant behavior patterns, thereby improving the dissemination effect of meeting information and the experience of participants. Furthermore, this invention can dynamically adjust the delivery strategy based on participants' historical feedback, further enhancing the targeting and effectiveness of the delivery and ensuring the efficient and stable operation of a subsequent cloud-based data information delivery system. Attached Figure Description

[0077] Figure 1 This is a block diagram of a cloud-based data information push system proposed in this invention. Detailed Implementation

[0078] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0079] Reference Figure 1 A cloud-based data information push system, comprising a data processing and sequence segmentation module, a meeting data correlation analysis module, a participant user behavior pattern recognition module, and a meeting content push formulation and optimization module;

[0080] Data Processing and Sequence Partitioning Module: Extracts meeting data sets from the database and processes and partitions the meeting data containing time-series characteristics for subsequent analysis; the meeting data sets include meeting timestamps, participant sets, and meeting content;

[0081] The conference data correlation analysis module uses correlation coefficients to analyze the correlation between conference data and filters and obtains conference data groups; it uses clustering algorithms to classify conference data groups by topic and determines the topic affiliation of the conference data in each group.

[0082] The participant behavior pattern recognition module extracts the participant behavior dataset from the participant set and defines behavior labels; it then uses a trained decision tree classifier to map the participant behavior labels, thereby recognizing the user's behavior patterns.

[0083] Meeting content push design and optimization module: Based on the meeting content and the behavioral tags of the participants, the push weight is calculated and then the meeting content is pushed to the participants.

[0084] It should be noted that the application of the cloud-based data information push system in this embodiment of the invention can be a meeting scenario in the field of information management and push. Specifically, it can be a system that uses big data analysis technology to comprehensively analyze and process meeting data in order to classify the topics of meeting content and identify user behavior patterns. Through multiple steps such as data processing and sequence division, meeting data correlation analysis, identification of the behavior patterns of participating users, and push formulation and optimization, a highly integrated and intelligent data information precision push system is formed. This system can accurately push meeting content based on the characteristics of meeting content and the behavior patterns of participating users, and dynamically adjust the push strategy based on historical user feedback to optimize the push effect.

[0085] The data processing and sequence partitioning module extracts the conference data set from the database and performs processing and sequence partitioning on the conference data containing time-series characteristics, including the following steps:

[0086] Step 101: Extract the meeting data set from the database and label it as...

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

[0088] Where, d g This represents the g-th meeting data entry, which includes 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, where n represents the total number of meeting data entries, and d1, d2, ..., dn. nLet each represent the first meeting data entry, the second meeting data entry, ..., the nth meeting data entry, respectively.

[0089] Step 102: For meeting data containing time-series characteristics (such as meeting timestamps), process it using the sliding window technique:

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

[0091] Then we have a time series data set T = {t1, t2, ..., t} m} and was 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 Let t represent the first time series data, the second time series data, ..., the m-th time series data, respectively. (r-1)w+1 ,t (r-1)w+2 ,...,t rw Let these represent the (r-1)w+1th time subsequence, the (r-1)w+2th time subsequence, ..., the rwth time subsequence, respectively. Indicates to Round up.

[0094] The conference data correlation analysis module uses correlation coefficients to analyze the correlation between conference data and filters and obtains conference data groups; it then uses clustering algorithms to classify the conference data groups by topic and determines the topic affiliation of the conference data in each group. The steps include:

[0095] Step 201: For any two meeting data, i.e. d i ,d j PCR using Pearson correlation coefficient ij Analyze meeting data d i and d j The correlation between them; where the Pearson correlation coefficient is calculated using the following formula:

[0096]

[0097] In the formula, PCR ij This represents the Pearson correlation coefficient, where i and j represent the indices of any two conference data points, and a... 1H ,a 2Hrespectively correspond to any two conference data d i and d j , H represents the index of the numerical attribute. respectively represent the mean of 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 according to the calculation result of the Pearson correlation coefficient, if PCR ij > 0, it is judged that any two conference data are positively correlated; if PCR ij < 0, it is judged that any two conference data are negatively correlated; if | PCR ij | is closer to 1, it is judged that the correlation between the two conference data is stronger; if PCR ij = 0, it is judged that there is no linear relationship between the two conference data.

[0099] Step 203, screen out the conference data whose correlation coefficient is positively correlated, and collect them into a conference data group, which is used for conference theme classification.

[0100] Step 204, the steps of conference theme classification of the conference data group by clustering algorithm are as follows:

[0101] Z1, determine the number of conference themes K;

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

[0103] Z3, for each data point x q in the conference data group, calculate the distance E(x q , μ k ) between each data point x q and each initial centroid μ k ; wherein q represents the index of the data point.

[0104] Z4, assign the data point x q to the conference theme set C k corresponding to the initial centroid μ k with the smallest distance, that is wherein k represents the index of the initial centroid with the smallest distance to the data point x q , 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 smallest distance to the data point x q is found and assigned to the corresponding conference theme set C k .

[0105] Z5, updating the initial centroid wherein, represents the updated initial centroid, and h is the iteration number;

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

[0107] Z7, using the minimization objective function to represent the clustering process:

[0108]

[0109] wherein, OF represents the result value of the minimization objective function;

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

[0111] For the text data set of the conference content WL = {L1, L2,..., Le}, the TF-IDF method is used to convert it into a vector representation, and the text vector set of the conference content is set as WV = {V1, V2,..., Ve}; wherein, WV is the vector representation of WL, e represents the number of text data of the conference content, L1, L2,..., Le respectively represent the 1st text data, the 2nd text data,..., the e-th text data, V1, V2,..., Ve respectively represent the 1st text vector, the 2nd text vector,..., the e-th text vector. e e e e

[0112] The text vector set V of the conference content is matched with the conference theme set C k , and the LDA algorithm is used for processing, to output a conference theme word matrix Φ ∈ I K×WV and a text theme distribution matrix Θ ∈ I e×WV as the feature representation of the conference content, and further obtain the conference content feature A = (Φ, Θ).

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

[0114] ​​​​The participant behavior pattern recognition module extracts the participant behavior dataset from the participant set and defines behavior labels; the steps of identifying user behavior patterns by training a decision tree classifier to map the participant behavior labels include:

[0115] Step 301: Extract the behavior dataset of the participants from the set of participants as U = {(u1,y1),(u2,y2),...,(u...} N ,y N In the formula, u represents the participant user ID, y represents the set of participant user behavior attributes (such as the number of times they attended the meeting, the number of times they spoke, etc.), and N represents the total number of participants.

[0116] Step 302: Obtain the behavioral feature set F = {f1, f2, ..., f...} of the participating users. M}, where f1, f1, ..., f M Let $\mathbf$ represent the behavioral characteristics of the first participating user, $\mathbf$ the behavioral characteristics of the second participating user, $\mathbf$ the behavioral characteristics of the Mth participating user, and so on, where $M$ represents the total number of behavioral characteristics of the participating users.

[0117] Step 303: For each participant in the participant behavior dataset U, define the participant behavior feature function. This function maps attendees to their behavioral feature vectors. Above, that is:

[0118]

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

[0120] Step 304: Based on the extracted behavioral feature vectors of participating users Define a set of behavioral labels to represent the behavioral patterns of different participants, and obtain the behavioral label 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 label and the extracted behavior feature vector of the participating user to construct a training set In the formula, B represents the number of training samples;

[0122] Step 306, select a decision tree classifier to train the constructed training set ζ, and map the behavior feature vector of the participating user to the corresponding behavior label;

[0123] Step 307, for a new participating user u', extract the behavior feature vector of the new participating user and use the trained decision tree classifier to classify the new participating user u', and output a new behavior label X' representing that the behavior mode of the new participating user u' is closest to X'.

[0124] The conference content pushing formulation and optimization module calculates the pushing weight based on the conference content and the behavior label of the participating user, and then pushes the conference content to the participating user. The steps include:

[0125] Step 401, obtain the conference content features corresponding to the conference content and the behavior label set corresponding to the behavior label of the participating user;

[0126] Step 402, according to the conference content features and the behavior label set of the participating user, calculate and analyze the pushing weight PH between the conference content features and the behavior features of the participating user 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 behavior features of the participating user after behavior label classification, p represents the behavior feature index of the participating user after behavior label classification, A b represents the conference content feature, F p represents the behavior feature of the participating user after behavior label classification, α b , β p respectively represent the weight coefficients of A b and F p , σ(A b , F p ) is an adjustment function for adjusting the pushing weight according to the values of A b and F p ;

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

[0130] Step 403, after getting the push weight, PH bp According to the push weight, the conference content is pushed to the participating user, and the specific push strategy includes:

[0131] According to the push weight from high to low, the conference content is pushed to each participating user;

[0132] Set a push threshold, if the push weight is greater than the push threshold, the corresponding conference content is pushed to the participating user, otherwise, the corresponding conference content is not pushed;

[0133] According to the historical feedback of the participating user, the push weight and the push threshold are dynamically adjusted to optimize the push effect;

[0134] Wherein, according to the historical feedback of the participating user, the push weight and the push threshold are dynamically adjusted as follows:

[0135] Get the click rate, reading time and like number in the historical feedback, and calculate the quantile rank corresponding to the click rate, reading time and like number;

[0136] The historical feedback index HFI of the participating user is calculated by the formula:

[0137]

[0138] In the formula, HFI represents the historical feedback index of the participating user, which is used to reflect the interest and participation of the user to the push content; Ra(CR), Ra(RD) and Ra(LC) represent the quantile rank of click rate, reading time and like number respectively; R represents the total number of samples, which is used for quantile rank standardization; η, ψ and λ are three dynamic learning coefficients, which correspond to click rate, reading time and like number respectively, and are self-adaptively adjusted according to the change of historical data to capture the influence of long-term trend and short-term fluctuation, and reflect the preference trend of participating user changing with time; c1, c2 and c3 are preset nonlinear parameters, which can determine the best value through test to optimize the fitting effect of nonlinear characteristics;

[0139] Based on the historical feedback index of the participating user, the weight and priority of the push content are adjusted:

[0140] Set the push weight update function, which is used to dynamically adjust the push weight according to the historical feedback index of the participating user, and the specific formula is as follows:

[0141]

[0142] In the formula, PH bp(l,v+1) PH represents the push weight of the conference participant I at the next time point, PH bp(l,v) PH represents the push weight of the conference participant I at the current time point, and delta represents the slope, which is used to control the speed of weight update, HFI l HFI represents the historical feedback index of the conference participant I, HFI represents the average value of the historical feedback index of all conference participants, and v represents the time point.

[0143] The push threshold is dynamically adjusted in combination with the distribution of the push weight and the historical feedback index of all conference participants: a push threshold update function is set, which is used to dynamically adjust the push threshold according to the statistical characteristics of the push weight and the historical feedback index at the current time point, and the specific formula is as follows:

[0144]

[0145] In the formula, theta v+1 represents the push threshold at the next time point; theta v represents the push threshold at the current time point; and gamma is an adjustment coefficient, which is used to control the amplitude of threshold update. represents the average value of the push weight of all conference participants. represents the standard deviation of the historical feedback index.

[0146] In the embodiment of the present application, the conference data set is extracted, and the conference data containing time sequence characteristics are processed and sequenced, thereby providing accurate and ordered data basis for subsequent analysis. The correlation between the conference data is analyzed by using the correlation coefficient, and the conference data group is obtained by screening. The clustering algorithm is used to classify the conference theme of the conference data group, and the theme attribution of the conference data in each conference data group is determined, thereby providing strong support for conference content push, realizing accurate feature representation of conference content, extracting the behavior data set of the conference participant, defining the behavior label, and using the trained decision tree classifier to map the behavior label of the conference participant, thereby effectively identifying the behavior mode of the conference participant, providing an important basis for personalized conference content push, combining the conference content features and the behavior mode of the conference participant, calculating the push weight, and dynamically adjusting the push strategy according to the historical feedback of the conference participant, thereby realizing accurate and efficient conference content push. In summary, the present application solves the problem of incomplete content correlation 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 schemes.

[0147] In addition, the formula involved in the above is to remove the dimension and take the numerical value, which is obtained by collecting a large amount of data to simulate the closest real situation by software. The proportional coefficient in the formula and the preset threshold in the analysis process are set by the person skilled in the art according to the actual situation or obtained by a large amount of data simulation. The size of the proportional coefficient is to obtain a specific numerical value of each parameter for subsequent comparison. The size of the proportional coefficient depends on the amount of sample data and the corresponding processing coefficient preliminarily set by the person skilled in the art for each group of sample data. As long as it does not affect the proportional relationship between the parameters and the quantized numerical value.

[0148] Each embodiment in the specification is described in a progressive manner, and the same and similar parts between each embodiment can be referred to each other. Each embodiment mainly describes the difference 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 related parts can be referred to the part of the method embodiment.

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

[0150] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer usable program code.

[0151] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The device that implements the functions specified in one flow or multiple flows and / or blocks Figure 1 The device that implements the functions specified in one block or multiple blocks.

[0152] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 of the flow or flows and / or blocks Figure 1 of the block or blocks specified in the flow.

[0153] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that are executed on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 of the flow or flows and / or blocks Figure 1 of the block or blocks specified in the flow.

[0154] Secondly: the drawings of the embodiments disclosed in the present application only involve the structures related to the embodiments disclosed in the present application, other structures can refer to the general design, and in the case of no conflict, the same embodiments and different embodiments of the present application can be combined with each other;

[0155] Finally: the above only describes the preferred embodiments of the present application, but the protection scope of the present application is not limited to this, any person skilled in the art can make equivalent replacement or change according to the technical scheme and the inventive concept of the present application within the technical range disclosed in the present application, which should be covered in the protection scope of the present application.

Claims

1. A data information push system based on cloud conferencing, characterized in that: The system comprises a data processing and sequence division module, a conference data correlation analysis module, a participant user behavior pattern recognition module, and a conference content push formulation and optimization module. The data processing and sequence division module extracts a conference data set from a database, and processes and sequence divides conference data containing time sequence characteristics; the conference data set comprises a conference timestamp, a participant user set, and conference content. The conference data correlation analysis module analyzes the correlation between conference data using a correlation coefficient, and screens and obtains a conference data group; the conference data group is classified by conference theme using a clustering algorithm, and the theme attribution of conference data in each conference data group is determined. The participant user behavior pattern recognition module extracts a participant user behavior data set from a participant user set, and defines a behavior label; the behavior label of a participant user is mapped by training a decision tree classifier, so as to recognize the behavior pattern of the participant user. The conference content push formulation and optimization module calculates a push weight based on conference content and a behavior label of a participant user, and then pushes conference content to the participant user. The process of the conference content push formulation and optimization module based on conference content and a behavior label of a participant user, calculating a push weight, and then pushing conference content to the participant user, comprises: obtaining conference content features corresponding to conference content and a behavior label set corresponding to a behavior label of a participant user; According to the conference content features and the behavior label set of the conference user, a push weight between the conference content features and the behavior features of the conference user is calculated and analyzed : ； In the formula, represents the total number of conference content features, represents the conference content feature index, represents the total number of behavior features of the participating user after the behavior label classification is completed, represents the behavior feature index of the participating user after the behavior label classification is completed, represents the conference content feature, represents the behavior feature of the participating user after the behavior label classification is completed, respectively represent and weight coefficients of is an adjustment function used to adjust the push weight according to the values of and ; After the push weight is obtained According to the push weight, the conference content is pushed to the participating user, and the specific push strategy includes: for each participant user, pushing conference content to the participant user in descending order of push weight; setting a push threshold value, if the push weight is greater than the push threshold value, the corresponding conference content is pushed to the participant user, otherwise, the corresponding conference content is not pushed to the participant user; quantitatively analyzing historical feedback of a participant user, and dynamically adjusting the push weight and the push threshold value.

2. The data information pushing system based on cloud conference according to claim 1, characterized in that: The process of the data processing and sequence division module extracting a conference data set from a database, and processing and sequence dividing conference data containing time sequence characteristics, comprises: extracting a conference data set from a database, and marking it as ； wherein, represents the first meeting data, containing a meeting timestamp, a set of participating users, and meeting content, represents an index of the meeting data, , represents a total number of the meeting data, respectively represent the first meeting data, the second meeting data, …, the meeting data. for conference data containing time sequence characteristics, using a sliding window technique for processing: The size of the sliding window is defined as Then there is a time series data set and is segmented into a plurality of time subsequences; , wherein ; in, Index representing a time series, Indices representing the indexes of time subsequences. These represent the first time series data, the second time series data, ..., the ... A time series data, They represent the first The time subsequence, the th time subsequence The time subsequence, ..., the nth time subsequence A time subsequence, Indicates to Round up.

3. The data information pushing system based on cloud conference according to claim 2, characterized in that: The process of the conference data correlation analysis module analyzing the correlation between conference data using a correlation coefficient, and screening and obtaining a conference data group, comprises: For any two meeting data, i.e. Using Pearson correlation coefficient Analyze meeting data and The correlation between them; where the Pearson correlation coefficient is calculated using the following formula: ； wherein, denotes the Pearson correlation coefficient, denotes an index of any two conference data, corresponding to any two conference data and the first numerical attribute, denotes the mean of any two conference data and the mean, denotes an index of numerical attributes; According to the calculation result of the Pearson correlation coefficient, the correlation between the conference data is analyzed, if , it is judged that any two conference data are positively correlated; if , it is judged that any two conference data are negatively correlated; if , the closer to 1, the stronger the correlation between the two conference data; if , it is judged that there is no linear relationship between the two conference data; screening conference data whose correlation coefficient with any other conference data is positive, and collecting the conference data into a conference data group for conference theme classification.

4. The data information pushing system based on cloud conference according to claim 1, characterized in that: The process of the conference data correlation analysis module classifying a conference data group by conference theme using a clustering algorithm, comprises: Z1, determine number of conference topics ; Z2, randomly select from the set of meeting data points as the initial centroid and label as ; Z3, for each data point in the meeting data set , compute the distance of each data point to each initial centroid ; wherein, denotes the data point index; Z4, the data point is assigned to the initial centroid with the smallest distance The corresponding set of meeting topics In other words, ; wherein, represents the final determination of the initial centroid with the smallest distance to the data point , and , represents the index of all other initial centroids, to ensure that the initial centroid with the smallest distance to the data point is found and assigned to the corresponding set of meeting topics ; Z5, updating the initial centroid , wherein, represents the updated initial centroid, is the iteration number, is the number of clusters, is the total number of data points contained in the th cluster at the th iteration. Z6, repeating steps Z3 to Z5 until the position of the initial centroid no longer changes significantly or a preset number of iterations is reached ; Z7, using a minimum objective function to represent the clustering process: ； In the formula, represents the result value of the minimized objective function.

5. The data information pushing system based on cloud conference according to claim 4, characterized in that: The process of the conference data correlation analysis module determining the theme attribution of conference data in each conference data group, comprises: For the text data set of meeting content ,use The method converts it into a vector representation, let the set of text vectors of the meeting content be . ;in, for The vector representation of , The number of texts representing the meeting content. These represent the first text data, the second text data, ..., the ... Text data, Let them represent the first text vector, the second text vector, ..., the third text vector respectively. Text vector; text vector set of conference content is matched with conference topic set , processed by LDA algorithm, outputting a conference topic word matrix and a text topic distribution matrix , as a feature representation of conference content, and further obtaining conference content features ; wherein in the conference topic-word matrix 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; in the text topic distribution matrix 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; denotes the size of the vocabulary, denotes the number of conference topics.

6. The data information pushing system based on cloud conference according to claim 5, characterized in that: The participant user behavior pattern recognition module extracts a participant user behavior data set from a participant user set, and defines a behavior label. The process of the participant user behavior pattern recognition module mapping the behavior label of a participant user by training a decision tree classifier, so as to recognize the behavior pattern of the participant user, comprises: A behavior data set of the participating user is extracted from a participating user set as , wherein, represents a participating user ID, represents a participating user behavior attribute set, represents a total number of participating users; Obtaining behavior feature set of participating user , wherein, respectively represent behavior feature values of the first participating user, the second participating user, …, and the th participating user, represent the total number of behavior feature values of the participating users. For each attending user in the set of behavior data of attending users , define a behavior feature function of the attending user , use the function to map the attending user to a behavior feature vector of the attending user , namely: ； In the formula, represents a dimensional real vector space, is the dimension of the behavior feature vector of the participating user, indicating the total number of behavior feature values of the participating user; represents the value of the participating user on the behavior feature value of the participating user; Based on the extracted behavior feature vector of the participating user , define a set of behavior labels for representing different behavior patterns of the participating user, obtain a behavior label set ; wherein represents the number of behavior labels; Combine the defined behavior label and the extracted behavior feature vector of the participant user to construct a training set , wherein, represents the number of training samples; A decision tree classifier is selected, and the constructed training set Training is performed to map the behavior feature vectors of the participating users to the corresponding behavior labels. For a new participant user , extract the behavior feature vector of the new participant user , and use the trained decision tree classifier to classify the new participant user , output a new behavior label , indicating that the behavior mode of the new participant user is closest to .

7. The data information pushing system based on cloud conference according to claim 1, characterized in that: The process of quantitatively analyzing historical feedback of a participant user, and dynamically adjusting the push weight and the push threshold value, comprises: Obtain the click rate, reading duration and like number in the historical feedback, and calculate the quantile rank corresponding to the click rate, reading duration and like number; Calculating a historical feedback index for a participating user using a formula : ； In the formula, represents the historical feedback index of the participating user, and is used to reflect the interest and participation of the user in the pushed content; respectively represent the quantile ranking of the click rate, the reading time length, and the number of likes; represents the total number of samples, and is used for standardization of the quantile ranking; are three dynamic learning coefficients, respectively corresponding to the click rate, the reading time length, and the number of likes; is a preset nonlinear parameter; Adjust the weight and priority of the pushed content based on the historical feedback index of the participating user: Set the pushed weight update function, and the specific formula is as follows: ； wherein, represents a participating user a push weight at a next time point, represents a participating user a push weight at a current time point, represents a learning rate for controlling a speed of weight update, represents a participating user a historical feedback index of the participating user, represents an average value of historical feedback indices of all participating users, represents a time point; Adjust the push threshold dynamically in combination with the distribution of the push weight and the historical feedback index of all participating users: set the push threshold update function, and the specific formula is as follows: Adjust the push threshold dynamically in combination with the distribution of the push weight and the historical feedback index of all participating users: set the push threshold update function, and the specific formula is as follows: ； In the formula, represents the push threshold of the next time point; represents the push threshold of the current time point; is an adjustment coefficient, used to control the amplitude of the threshold update; represents the average value of the push weights of all participating users; represents the standard deviation of the historical feedback index.

Images