CN111552882B - News influence calculation method and device, computer equipment and storage medium - Google Patents
News influence calculation method and device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN111552882B CN111552882B CN202010386342.7A CN202010386342A CN111552882B CN 111552882 B CN111552882 B CN 111552882B CN 202010386342 A CN202010386342 A CN 202010386342A CN 111552882 B CN111552882 B CN 111552882B
- Authority
- CN
- China
- Prior art keywords
- news
- influence
- data
- index
- historical
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000004364 calculation method Methods 0.000 title claims abstract description 96
- 238000013528 artificial neural network Methods 0.000 claims abstract description 52
- 238000000034 method Methods 0.000 claims abstract description 46
- 230000006399 behavior Effects 0.000 claims abstract description 43
- 238000012549 training Methods 0.000 claims abstract description 22
- 238000012552 review Methods 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 230000015654 memory Effects 0.000 claims description 6
- 230000001502 supplementing effect Effects 0.000 claims description 2
- 238000002372 labelling Methods 0.000 claims 3
- 238000007405 data analysis Methods 0.000 abstract description 2
- 230000003319 supportive effect Effects 0.000 abstract description 2
- 230000008569 process Effects 0.000 description 9
- 238000010586 diagram Methods 0.000 description 8
- 238000004422 calculation algorithm Methods 0.000 description 7
- 238000011156 evaluation Methods 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 4
- 238000005065 mining Methods 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 4
- 210000004027 cell Anatomy 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 238000012216 screening Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 230000000052 comparative effect Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- -1 hydrogen Chemical class 0.000 description 2
- 239000001257 hydrogen Substances 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 238000011524 similarity measure Methods 0.000 description 2
- 241001655736 Catalpa bignonioides Species 0.000 description 1
- 108010074506 Transfer Factor Proteins 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 210000002364 input neuron Anatomy 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 210000004205 output neuron Anatomy 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention belongs to the field of data analysis, and relates to a news influence calculation method, a news influence calculation device, computer equipment and a storage medium; the method comprises the steps of collecting historical news data, and determining an average reading index and a news type heat index of a news source; selecting historical news data with stable user behaviors according to the user behavior growth rate; determining a news comment number index and a news reading index related to user behaviors; dividing different historical news data into a plurality of event numbers; establishing an influence calculation model; determining the influence of the historical news data according to each index from the historical news data with the same event number, and marking a grade label; training the influence calculation model through a BP neural network; and inputting news data to be calculated into the trained influence calculation model, and outputting a result. The method considers that the influence of news has more supportive indexes, so that the accuracy of the influence calculation model is higher.
Description
Technical Field
The invention belongs to the field of data analysis, and particularly relates to a news influence calculation method and device based on a BP neural network, computer equipment and a storage medium.
Background
News influence refers to the influence of news on individual recipients, groups and the whole society after the news is spread. The network news is used as a main source of network public opinion and social public opinion, and the accurate judgment of the influence of the network news is particularly important. The method for calculating and evaluating the influence of the network news is widely applied to the fields of news public opinion discovery and mining, precise delivery of news advertisements, news performance evaluation and assessment and the like.
The key problem faced by news influence calculation is that (1) the indexes related to news are various, and if the indexes related to influence calculation are not screened out, the result is far from the real situation. (2) The current mainstream method for calculating the influence is a linear weighting-based method, and the weight setting mode of the method depends on manual selection and is subjective. (3) In the data acquisition stage, the condition of data missing is easy to occur. If the key index of the influence calculation is missing, the influence cannot be calculated directly.
Researchers have proposed many methods to calculate news influence, mainly the following three methods.
A linear weighting based approach. And calculating the influence by combining four indexes of an influence factor, a news reprinting rate, a news replying rate and a time factor of the information source website and adopting a method of manually weighting the indexes. Specifically, the time factor is obtained by substituting the difference between the release time and the news acquisition time into the time influence function of the corresponding category news; the influence factor of the information source website is determined by the popularity index of each website in the Chinese Internet index system; calculating the news reprint rate, wherein the click rate is difficult to obtain, and the reprint relation is determined by calculating the similarity between news and then is applied to the HIT algorithm to obtain the news reprint rate; and (4) calculating the news response rate, wherein the click rate is also unavailable, and the news response rate is obtained by directly corresponding the number of the comments to the response rate.
A method based on historical events. The popular news is not isolated, a series of continuous reports exist in the popular news, and by utilizing the characteristic, the news is converted into events by a clustering algorithm, and the importance of each event is judged according to a large amount of historical event corpora.
Based on the link structure. The method mainly measures the influence of each website by considering the network relationship among each news website. The network of the reference relationship among news websites needs to be analyzed, nodes in the network are regarded as mechanisms for releasing news, the whole network is a directed graph, and the connection leads to reference media from the guided mechanisms. In the method, the basic structure of the reference relation network is described by calculating average out degree, density, average path length, diameter and clustering coefficient. The average out degree shows that news of each website is referenced by several websites on average; the density represents the density degree of the connecting edges among the nodes in the network, namely the overall contact tightness degree of the news network; the average path length represents the distance between two nodes, namely, one piece of information can reach another website after being propagated for several times on average; the diameter is the maximum value of the distance between any two nodes in the network, and represents the distance between the two farthest network stations. Under the premise, the out-degree of each node is directly used as an important mark for measuring the influence, and meanwhile, the average influence between adjacent nodes is inspected by using the centrality of the degree.
Although these methods can calculate news influence in general and have advantages, the following disadvantages are inevitable: manually weighting the indexes, so that the result is subjective and inconsistent with the actual result; similar news is screened out from historical data and then is matched with the influence of the similar news, and the method needs a huge news data set and has low operability; the method for constructing a network by taking news as nodes and utilizing the node relationship to solve influence also needs a huge data set, and a large amount of news without reference relationship exists.
In summary, it is known that the technical solutions for calculating news influence in the prior art all have the defects of low accuracy, weak operability, complex operation, and the like, and therefore, an improved technical means is needed to solve the above problems.
Disclosure of Invention
In view of this, the invention provides a news influence calculation method, a news influence calculation device, a computer device and a storage medium, which can solve the problem of missing of time sequence indexes in news influence calculation, realize empowerment and influence calculation by adopting a BP (back propagation) neural network, and improve the accuracy of influence calculation.
In a first aspect thereof, the present invention provides a news influence calculation method based on a BP neural network, the method comprising:
collecting historical news data and determining an average reading index of a news source; the historical news data at least comprises a news type popularity index;
selecting historical news data with stable user behaviors according to the user behavior growth rate;
determining a news comment number index and a news reading number index related to user behaviors from stable historical news data;
dividing different historical news data according to the similarity of news contents and the number of news labels, and dividing a plurality of event numbers;
establishing an influence calculation model;
determining the influence of the historical news data through an influence calculation model according to each index from the historical news data with the same event number, and marking a grade label on the historical news data;
training the influence calculation model through a BP neural network based on historical news data labeled with a grade label;
and inputting the collected news data to be calculated into the trained influence calculation model, outputting the news influence to be calculated, and outputting the corresponding news influence according to the grade label.
In a second aspect of the present invention, the present invention provides a news influence calculation apparatus based on a BP neural network, comprising:
the data acquisition module is used for acquiring historical news data and news data to be calculated;
the index calibration module is used for determining each index of the influence calculation model, including a news source average reading index, a news type heat index, a news comment number index and a news reading index;
the model building module is used for building an influence force calculation model;
the tag calibration module is used for performing tag calibration on the influence of the historical news data;
the BP neural network module is used for training the influence calculation model by using historical news data for label calibration;
and the calculation module is used for calculating the influence of the historical news data according to the influence calculation model or calculating the influence of the news data to be calculated according to the trained influence calculation model.
In a third aspect of the present invention, the present invention provides a computer device, which includes a memory, a processor and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement the above-mentioned news influence calculation method based on the BP neural network.
In a fourth aspect of the present invention, the present invention also provides a computer-readable storage medium, on which a computer program is stored, the computer program being executed by a processor to implement the above-mentioned news influence calculation method based on a BP neural network.
The invention has the beneficial effects that: the method considers that the influence of news has more supportive indexes, so that the accuracy of designing the influence calculation model is higher. And the LSTM is used for predicting missing key time series data, so that data waste is avoided. On the setting of the weight in the model, a more accurate BP neural network which is a supervised algorithm is adopted for calculation, and the problem of over subjective artificial weighting is solved. And the number of comments and the number of readings after the user behavior tends to be stable are used, so that errors caused by time factors to results are eliminated. Finally, the purpose of improving the accuracy of news influence calculation is achieved.
Drawings
FIG. 1 is a flow chart of a method for calculating news influence based on a BP neural network according to the present invention;
FIG. 2 is a flow chart of data acquisition in the present invention;
FIG. 3 is a flow chart of index calculation according to the present invention;
FIG. 4 is a topology structure diagram of the present invention employing a three-layer BP neural network;
FIG. 5 is a flow chart of the present invention for calculating influence;
FIG. 6 is a flowchart of a method for calculating the influence of historical news according to the present invention;
FIG. 7 is a graph showing the comparison of the effect model of the present invention with two other prior art;
FIG. 8 is a flowchart of a method for calculating the influence of current news according to the present invention;
FIG. 9 is a block diagram of a news influence computing device based on BP neural network according to the present invention;
fig. 10 is a block diagram of a news influence computer device based on a BP neural network according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The method for calculating the news influence can be applied to various application scenes such as commercial websites, commodity websites or Social Networking Services (SNS). The method can be widely applied to the fields of news public opinion discovery and mining, accurate release of news advertisements, news performance evaluation and assessment and the like.
Fig. 1 is a flowchart of a news influence calculation method based on a BP neural network according to an embodiment of the present invention, including the following steps:
the collected historical news data comprises news data of each plate and each field of collected news; specifically, news of blocks such as time, sports, entertainment, finance, science, military and education can be included; the fields of a collection may include, but are not limited to, title, link, news source, total number of news sources, total number of reads from news source, news release time, collection time, news content, news tag, news genre, number of reviews, and number of reads field.
In this embodiment, the news type is used as one of the indexes for guiding subsequent influence calculation. Further, the popularity of the news types can be replaced by the number of the top lists of the news types in the news popularity list.
According to the news data of the fields, average reading numbers of the news sources can be obtained according to division operation according to the number of articles of the release sources and the total reading numbers of the release sources; in this embodiment, the average reading number of news sources is used as one of the indicators.
Fig. 2 is a process diagram of acquiring data according to the present invention, and acquiring news data may include:
1011. acquiring news data by using a web crawler;
1012. and preprocessing the news data, including deleting and statistically calculating news source information.
102, selecting historical news data with stable user behaviors according to the user behavior growth rate;
selecting historical news data with stable user behaviors according to the user behavior growth rate comprises calculating the user behavior growth rate of each period and calculating the absolute value of the difference value of the growth rates of each period and the adjacent period; when the absolute value of the difference value of the growth rates of a plurality of continuous periods meets the growth rate threshold, regarding the periods as periods which tend to be stable according to the growth rate of the user behavior, and acquiring historical news data in the stable periods, wherein the calculation formula of the growth rate of the user behavior of each period is represented as:
wherein the content of the first and second substances,
mean comment number growth rate, c, representing the k-th cyclekRepresents the average number of reviews for all news in the kth period,
represents the average read increase rate, r, of the k-th cyclekRepresenting the average reading of all news for the kth cycle.
Of course, for convenience of calculation, the embodiment of the present invention takes each day as a period, and the stable period is generally several days.
The purpose of finding out the period in which the user behavior increases and tends to be stable is to avoid the influence of time factors, and the subsequent calculation of influence force directly uses the value in which the user behavior tends to be stable to participate in the calculation.
103, determining a news comment number index and a news reading index related to the user behavior from the stable historical news data;
dividing the collected historical news data to obtain news data with complete user behaviors as a training set; adopting a Long Short Term Memory network (LSTM) to complement the news data with incomplete user behaviors, namely the news data has the same dimension with the complete news data, for example, the complete news data comprises a plurality of dimension data, and if the dimension data is title, link, news source, total number of news sources, total reading times of the news sources, news release time, acquisition time, news content, news label, news type, comment number, reading number segment and the like, and the incomplete news data lacks the dimension of the news comment number, the news comment number dimension is complemented through the LSTM news comment number dimension; therefore, the number of news comments and the number of news readings after the user behavior is stable are determined, and the number of the stable news comments and the number of the stable news readings are used as indexes.
The process of determining the news comment number index and the news reading index related to the user behavior may include:
finding out a period in which the user behavior tends to be stable in the data set according to the growth rate;
predicting the number of missing stable daily news reviews and stable daily news reading using LSTM; and the number of complete stable day news comments and the complete stable day news reading number are respectively used as indexes.
Fig. 3 is a process diagram of determining an index in the present invention, and as an implementation manner, the process of determining an index may further include:
1031. selecting at least five indexes according to historical experience;
1032. and screening the indexes by using the Pearson correlation coefficient, and screening the indexes with pairwise correlation less than 0.5.
By the index mode, the index defined by the claims of the invention can be used for processing, news forwarding numbers, news publishing platform numbers and the like can be used as the indexes, and a corresponding influence calculation model is established according to the selected indexes to calculate the news influence.
because the real news has the condition of multiple times of publication and multi-angle publication of the same news, the real news needs to be divided into events, and subsequent calculation is carried out by taking the events as units. Therefore, the method can well divide news into events by calculating the similarity of news contents and the same number of tags; specifically, different historical news data are calculated, correlation values of the different historical news data are calculated, the historical news data with the correlation value equal to 1 are divided into the same event, the historical news data with the correlation value equal to 0 are divided into different events, and a calculation formula of the correlation values is represented as:
wherein g (i, j) represents the correlation value of the ith news and the jth news; if the function value of g (i, j) is 1, the fact that i news and j news are divided into the same event is represented, and if the function value of g (i, j) is 0, the fact that i news and j news are divided into the same event is not represented; simijRepresenting the similarity between the ith news item and the jth news item, sim' is the set similarity threshold value, eqijIndicating the number of identical tags existing in the ith news and the jth news, and eq' is a threshold value for the number of identical tags set.
105, establishing an influence force calculation model;
in this embodiment, an influence calculation model is established according to the above four indexes, and more generally, the present invention may not be limited to the above four indexes, and may also include other indexes.
as an implementation mode, manual marking can be adopted, and the basis of the manual marking is the expert opinion; the standard of expert opinion mainly depends on the subjective judgment of the expert.
The step of marking the historical news data with the grade tags comprises the steps of marking the historical news data of different event numbers with the grade tags according to an average reading index of news sources, a news type heat index, a news comment number index and a news reading index, determining the influence of the historical news data according to an influence calculation model, dividing the grade tags according to the influence, and adding the grade tags to the historical news data.
The influence may be divided in multiple levels, normalized, divided in multiple levels from 0-1, e.g., into 5 levels a-E; assuming that the influence is 0.3, the rank may be assigned to D according to a rule of equal division, or the rank may be assigned to e.g. a value of 20% of the influence rank in the historical news data as a high rank and a medium rank.
This embodiment gives a specific division into five classes a-E, and preferably, the classes may be divided as follows: 0.875-1 → A; 0.625-0.875 → B; 0.375-0.625 → C; 0.125-0.375 → D; 0-0.125 → E.
in one embodiment, the present invention employs a basic BP-based neural network algorithm; the weight distribution process of the BP neural network algorithm comprises the following steps:
firstly, normalization processing needs to be carried out on an index column:
X＝[x1,x2,…,xn],xi∈[0,1],i＝1,2,…,n
their corresponding decision results are:
Y＝[0,100]
if there is a mapping G between X and Y, then:
Y＝G(X)
each index is listed as [ X ]1,X2,…,Xn]The decision result Y' is used as the output of the BP network, a sample set formed by the X and the Y is trained by utilizing the learning algorithm of the BP network, and when the network is converged, the mapping relation H between the X and the Y can be realized, and the following conditions are met:
|H(X)-G(X)|<ε
wherein ε is an arbitrarily small positive number. Considering that the effect of the input of each neuron on that neuron is reflected in the magnitude of the connection weight, it can be expected that the effect of X on Y is also necessarily reflected in the magnitude of the connection weight.
Optionally, a three-layer BP network model shown in fig. 4 may then be established, including an input layer, a hidden layer, and an output layer; the input layer can input the normalized indexes, and the output layer can output the decision result.
Specifically, n units in the input layer correspond to n indexes, Y in the output layer represents the final decision result, and the hidden layer comprises r units, WihV represents the connection weight between the input layer unit i and the hidden layer unit hhjRepresenting the connection right between the hidden layer unit h and the output layer unit j, the output b of the hidden layer unit hhComprises the following steps:
wherein f () is Sigmoid function, thetahIs the threshold of the h-th cell.
Similarly, the output y of the j-th cell in the output layerjComprises the following steps:
wherein e isjIs the threshold of the j-th cell in the output layer.
From the above formula, x can be obtainediFor yjThe sensitivity of (a) is:
order to
Meanwhile, the formula can be used for obtaining:
wherein, deriving the Sigmoid function can obtain:
from the above formula, one can obtain:
likewise, x can be obtainedkFor yjThe sensitivity of (a) is:
and can be derived:
training of X and Y compositionTraining set as learning sample of BP neural network, and setting WihAnd WkhAre respectively xiAnd xkThe connection weight coefficient between the corresponding input units i and k and the hidden layer unit h if Wi1|>|Wk1|,|Wi2|>|Wk2|,…,|Wir|>|WkrI, then coefficient xiSensitivity S ofiRatio coefficient xkSensitivity S ofkIt is large. This is because when the network converges, there are generally:
f1(Ob1)|V1j|≈f1(Ob2)|V2j|≈…f1(Obr)|Vrj|
order to
|Wsi|＝|Wi1|+|Wi2|+…+|Wir|
Assume that the weight assignment of n indices is λ1,λ2,…,λnAnd is and
it is possible to obtain:
wherein, | Wsi|＝|Wi1|+|Wi2|+…+|WirI is the coefficient x at network convergenceiAnd the sum of absolute values of the connection weights between the corresponding input unit i and all the hidden layer units.
The trained influence calculation model (NIM) includes:
NI＝(1+exp(-(ωarNar+ωthQth+ωscNsc+ωsrNsr-θ)))-1；
in the formula, NIFinal news influence; n is a radical ofarAverage reading number for news source; qthIs the news type popularity; n is a radical ofscNumber of news reviews for a stable period; n is a radical of hydrogensrFor a stable periodReading the data through the smell; theta is a threshold parameter of the BP neural network output layer; omegaarRepresenting the weight of the BP neural network corresponding to the first index; omegathRepresenting the weight of the BP neural network corresponding to the second index; omegascRepresents the weight, omega, of the BP neural network corresponding to the third indexsrAnd representing the weight of the BP neural network corresponding to the fourth index. Omegaar、ωht、ωcs、ωrsTheta and theta are selected by relying on a BP neural network according to the condition of the data set, and the numerical values are not fixed in different data sets and different stages of calculation; the first index, the second index, the third index and the fourth index can sequentially correspond to a news source average reading index, a news type heat index, a news comment number index and a news reading index.
And 108, inputting the collected news data to be calculated into the trained influence calculation model, outputting the news influence to be calculated, and outputting the corresponding news influence according to the grade label.
FIG. 5 is a flow chart of the present invention for calculating influence, comprising:
1081. inputting news data to be calculated;
1082. and outputting news influence from the trained BP neural network.
In one embodiment, as shown in fig. 6, this embodiment provides a flowchart of a calculation method for calculating influence of historical news, including the following steps:
finding a period in which the user behavior tends to be stable in historical news data according to the growth rate;
in the period tending to be stable, predicting the number of missing comments and reading numbers by using the LSTM, and screening out news data after the user behavior is stable;
event division is carried out on news data by using cosine distance, and training data are formed;
sending the training data to a BP neural network for training;
and transmitting the news data to be calculated into the trained BP neural network, and outputting the calculated influence result.
In this embodiment, in order to verify the correctness and validity of the proposed method, experiments were performed on real news data of the last week of 7 months in 2019 of fox search net. Crawl the news of 7 blocks in total for politics, sports, entertainment, finance, science, military and education as in table 1, and specifically contain the following fields: title, link, news source, total news item at source, total number of reads at source, news release time, acquisition time, news content, tag, news type, number of reviews, and reading number. Meanwhile, the number of comments and the reading number of each news in one week are tracked, and the number of comments and the reading number of each stable day are found.
TABLE 1 News data sorting scenarios
In the news data of 9 days in total, the data of the first two days are used for parameter selection experiments and mining of days with stable user behaviors, and the data of the last seven days are used for final influence calculation experiments.
In the data acquisition process, data with missing partial fields can appear, and the data with missing news type fields is processed in a manual supplementing mode; the data which is missing from key fields such as title, news content, labels, comment number and reading number and the data which are repeatedly collected are removed by deletion; finally, consistency detection is carried out on each numerical value field, and abnormal data are deleted.
In the collected news data set, the news categories have seven categories as shown in table 2: the need for a quantitative process for politics, sports, entertainment, finance, science, military and education. And (4) carrying out quantitative processing on each news type by using a hot news situation on the Baidu search wind and cloud board. The type division conditions of hot news (268 pieces after duplication removal) of seven days in total from 7/16 th in 2019 to 7/22 th in 2019 are counted (for news related to multiple types, the statistics is carried out in an equally dividing mode), and the news types are quantified by utilizing the top ranking times of the news of each type. The following table specifically shows:
TABLE 2 statistics of popular times in each news category
The popularity of the news type is directly represented by the times of the top charts of various types.
In the verification of the reasonability of the indexes, the maximum value of the Pearson correlation coefficient between every two of the four indexes is 0.448672 and is lower than 0.5, so that the four indexes all meet the requirement of constructing an index system.
The experimental parameters were set as follows:
(1) in the analysis of the stability of the user behavior, a threshold value theta is setbeThe setting is 0.02, and the result shows that the fifth day is the stable user behavior day.
(2) In the event partitioning process, sim 'and eq' are calculated using cycle validation. step size of sim 'in the range of [0,0.7] is set to 0.02, step size of eq' in the range of [0,5] is 1, and cosine similarity measure method is adopted as text similarity measure method. The maximum value of 0.92 for the final accuracy is obtained when sim ' is set to 0.26 and eq ' is set to 0 and 1, and the final optimal threshold parameters are, in combination with stability considerations when eq ' is 0 and 1: sim '0.26, eq' 1.
(3) 4 input neurons and 1 output neuron can be determined according to the number of input indexes and the number of output results; through the verification of the circular training, the maximum training frequency is selected to be 15000, and the minimum error and the learning rate of a training target are 0.1; the number of hidden layers is according to an empirical formula n2＝2×n1+1, wherein n is2The number of hidden layer neurons is 9, and the number of input layer neurons is 9.
The model is trained and influence calculation is further carried out in a mode of combining a Baidu popular news list and manual evaluation, and in order to facilitate the construction of a training set, the influence is represented by five grades of A-E from large to small in sequence. The experimental results were evaluated using grade differences. On the basis of 100% accuracy, the accuracy is decreased by 25% every other level. The accuracy is calculated as follows:
acc＝[1-0.25×dis(res,res′)]×100％
where acc is the model accuracy, res is the actual influence level of a certain news item, res' is the influence level calculated by the model, and dis (i, j) is used to calculate the separation distance between the i level and the j level.
The evaluation method can effectively convert the difference situation between the calculation result and the actual result into numbers. And when the calculation result is completely consistent with the actual result, the acc value is 1, when the calculation result is not matched with the actual result, the acc is smaller when the difference is larger, and when the difference reaches four maximum levels, the acc is 0.
Finally, the comparative experiments were performed on five test sets each having a data size of 30, and the comparative model was MIQM [1]]And MINF 2]. In the MIQM model, four key indexes are combined in a linear weighting mode, finally, the setting of each parameter value is determined in a step verification mode, and the parameter setting aims to enable the mean value and the standard deviation of the grading result to be small at the same time. When the four weight parameters are respectively 0.05, 0 and 0.9, the mean value and the standard deviation of Euclidean distances of the scoring result set are simultaneously smaller, wherein the mean value is 0.0373, and the standard deviation is 0.0982. Therefore, when the MIQM model is used for news influence calculation, ω ar is 0.05, ωht＝0.05、ωcs＝0、ωrs0.9. In the MINF model, based on a linear structure, the news influence is calculated by using four indexes of release acquisition time difference, information source influence factor, transfer factor and reply number, and the weight parameter of the model is set to directly determine that two parameters are respectively 0.2 and 0.8 according to the twenty-eight law.
Table 3 and table 4 show ten news in the highest-influence class and ten news in the lowest-influence class in the calculation results of the present invention, respectively, and fig. 7 is a comparison graph of the accuracy of the results of the present invention with MIQM and MINF, in which the abscissa represents data grouping and the ordinate represents accuracy, which clearly shows that the present invention has a great improvement and progress in accuracy compared with the other two prior art.
Table 3 ten news in the top-ranked class
Table 4 scoring ten news items in the lowest class
The source files for MIQM and MINF are:
[1] sungguo catalpa, Chou, is Yan and Lihuakang, a microblog influence power model [ J ] based on linear weighting, university of Sichuan university, 2016,48(01), 78-84.
[2] Yangweie, Darutha, Chixia, an influence analysis method of network news based on information retrieval technology [ J ] software academic report 2009,20(09): 2397-.
In one embodiment, as shown in fig. 8, this embodiment provides a flowchart of a calculation method for calculating influence of current news, including the following steps:
predicting daily news comment number and news reading number by using the LSTM;
event division is carried out on news data by using cosine distance, and training data are formed;
sending the training data to a BP neural network for training;
and transmitting the news data to be calculated into the trained BP neural network, and outputting the calculated influence result.
In order to apply the method to influence calculation of real-time news, real news data of 4-month-23-year-2020 search fox network are crawled, and real-time influence calculation is carried out. Crawl the news of 3 plates in total for the time, finance and technology, and specifically contain the following fields: title, link, news source, total news item at source, total number of reads at source, news release time, acquisition time, news content, tag, news type, number of reviews, and reading number.
TABLE 5 News data sorting scenarios
The data preprocessing and the quantification of news types are consistent with the implementation of historical news, and the main differences are that: the historical news data can adopt the comment number and the reading number of each piece of news after the user behavior is stable because the comment number and the reading number of each piece of news are tracked, so that the accuracy performance is ensured, and the influence calculation of the real-time news cannot realize the tracking of the user behavior, so that the comment number and the reading number after the stability are predicted by adopting an LSTM model and further participate in the influence calculation as indexes.
Table 6 shows the calculation results of ten news items in the influence calculation of real-time news data according to the present invention.
TABLE 6 calculation results of influence of ten real-time news data
According to the embodiment, the influence of the real-time news can be effectively calculated, namely, the influence of the historical news and the influence of the real-time news can be calculated, and the method and the device can be widely applied to the fields of news public opinion discovery and mining, precise release of news advertisements, news performance evaluation and assessment and the like.
In one embodiment, fig. 9 is a block diagram of a news influence computing device based on a BP neural network according to an embodiment of the present invention; the device comprises the following modules:
the data acquisition module is used for acquiring historical news data and news data to be calculated;
specifically, historical news data and news data to be calculated can be acquired through news APP acquisition and also can be acquired through other networks in a crawler mode; generally speaking, by nature of news dissemination, the same type of news will get a greater focus over a period of time; therefore, in order to more accurately calculate the influence of the news data to be calculated, the selected historical news data is close to the news data to be calculated in time.
The index calibration module is used for determining each index of the influence calculation model, including a news source average reading index, a news type heat index, a news comment number index and a news reading index;
specifically, the news type, the average reading number of news sources, the reading number of news reads, and the number of news comments are calibrated as indexes from the historical news data collected by the data collection module.
The model building module is used for building an influence force calculation model;
specifically, an influence calculation model is constructed according to various indexes determined by the index calibration module.
The tag calibration module is used for performing tag calibration on the influence of the historical news data;
specifically, the influence score may be normalized and divided into a plurality of levels; and calling a calculation module, corresponding the influence values (scores) of the historical news data calculated by the influence calculation model to each grade, and calibrating the historical news data by adopting grade labels after corresponding.
The BP neural network module is used for training the influence calculation model by using historical news data for label calibration;
specifically, the BP neural network may adopt a basic network, and first normalize each index, and correspond a decision result output by each index to the normalized index; training according to a learning algorithm of the BP neural network until the training is completed to determine a trained influence calculation model (NIM):
NI＝(1+exp(-(ωarNar+ωthQth+ωscNsc+ωsrNsr-θ)))-1；
in the formula, NIThe final news influence; n is a radical ofarAverage reading number for news source; qthIs news type popularity; n is a radical of hydrogenscNumber of news reviews for a stable period; n is a radical ofsrReading for stable period news; theta is a threshold parameter of the BP neural network output layer; omegaarRepresenting the weight of the BP neural network corresponding to the first index; omegathRepresenting the weight of the BP neural network corresponding to the second index; omegascRepresenting the weight of the BP neural network corresponding to the third index; omegasrAnd representing the weight of the BP neural network corresponding to the fourth index.
And the calculation module is used for calculating the influence of the historical news data according to the influence calculation model or calculating the influence of the news data to be calculated according to the trained influence calculation model.
Fig. 10 is a block diagram of a computer device according to an embodiment of the present invention. As shown in fig. 10, the computer apparatus includes a processor, a nonvolatile storage medium, an internal memory, and a network interface connected through a system bus. The non-volatile storage medium of the computer device stores an operating system and computer-executable instructions, and the computer-executable instructions are used for implementing the news recommending method based on user behavior data detection provided by the embodiment of the invention. The processor is used to provide computing and control capabilities to support the operation of the entire computer device. The internal memory of the computer device provides an environment for the operating system and the computer-executable instructions of the non-volatile storage medium to run, and the network interface is used for network communication with other computer devices. The computer device may be a terminal such as a mobile phone, a tablet computer, a pc (personal computer), or a server. Those skilled in the art will appreciate that the architecture shown in fig. 10 is a block diagram of only a portion of the architecture associated with the subject application, and is not intended to limit the computing device to which the subject application may be applied, and that a computing device may in particular include more or less components than those shown, or combine certain components, or have a different arrangement of components.
Aiming at the defects in the current news influence calculation model, the invention provides a method and a device for calculating news influence based on a BP neural network, computer equipment and a storage medium, so as to solve the problem of low accuracy in the current model. According to the invention, on the aspect of selecting the indexes, the factors of the breadth and depth of news propagation, the quality of news sources, the field of news topics and time are fully considered. After the processes of finding out stable days of user behaviors, predicting missing values of the stable days and dividing events are carried out, the data set is transmitted into a news influence calculation model, and the influence of measuring news objectively and accurately is sought.
In the present invention, unless otherwise expressly stated or limited, the terms "mounted," "disposed," "connected," "fixed," "rotated," and the like are to be construed broadly, e.g., as meaning fixedly connected, detachably connected, or integrally formed; can be mechanically or electrically connected; the terms may be directly connected or indirectly connected through an intermediate, and may be communication between two elements or interaction relationship between two elements, unless otherwise specifically limited, and the specific meaning of the terms in the present invention will be understood by those skilled in the art according to specific situations.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (9)
1. A news influence calculation method based on a BP neural network is characterized by comprising the following steps:
collecting historical news data and determining an average reading index of a news source; the historical news data at least comprises a news type popularity index;
selecting historical news data with stable user behaviors according to the user behavior growth rate;
determining a news comment number index and a news reading number index related to user behaviors from stable historical news data;
dividing different historical news data according to the similarity of news contents and the number of news labels, and dividing a plurality of event numbers;
establishing an influence calculation model by adopting a BP neural network according to each index;
determining the influence of the historical news data through an influence calculation model according to each index from the historical news data with the same event number, and marking a grade label on the historical news data;
training the influence calculation model through a BP neural network based on historical news data labeled with a grade label;
and inputting the collected news data to be calculated into the trained influence calculation model, outputting the news influence to be calculated, and outputting the corresponding news influence according to the grade label.
2. The method of claim 1, wherein the selecting historical news data with stable user behavior according to the user behavior growth rate comprises calculating the user behavior growth rate of each period, and calculating an absolute value of a difference between the growth rates of each period and the adjacent periods; when the absolute value of the difference value of the growth rates of a plurality of continuous periods meets the growth rate threshold, regarding the periods as periods which tend to be stable according to the growth rate of the user behavior, and acquiring historical news data in the stable periods, wherein the calculation formula of the growth rate of the user behavior of each period is represented as:
wherein, the first and the second end of the pipe are connected with each other,
mean comment number growth rate, c, representing the kth periodkRepresents the average number of reviews for all news in the kth period,
represents the average read increase rate, r, of the k-th cyclekRepresenting the average reading of all news for the kth cycle.
3. The method for calculating news influence based on the BP neural network according to claim 1, wherein the step of determining the comment number index and the reading number index related to the user behavior from the stable historical news data comprises the steps of dividing the collected historical news data into complete news data of the user behavior as a training set; supplementing incomplete news data of user behaviors by adopting an LSTM; therefore, the number of news comments and the number of news readings after the user behavior is stable are determined, and the number of the stable news comments and the number of the stable news readings are used as indexes.
4. The method of claim 1, wherein the dividing of the different historical news data according to the similarity of news contents and the number of news tags and the dividing of the plurality of event numbers includes calculating the different historical news data, calculating correlation values of the different historical news data, dividing the historical news data with the correlation value equal to 1 into the same event, dividing the historical news data with the correlation value equal to 0 into different events, and expressing a calculation formula of the correlation values as:
wherein g (i, j) represents the correlation value of the ith news and the jth news; if the function value of g (i, j) is 1, the fact that i news and j news are divided into the same event is represented, and if the function value of g (i, j) is 0, the fact that i news and j news are divided into the same event is not represented; simijRepresenting the similarity between the ith news item and the jth news item, sim' is the set similarity threshold value, eqijIndicating the number of identical tags existing in the ith news and the jth news, and eq' is a threshold value for the number of identical tags set.
5. The method of claim 1, wherein the step of labeling historical news data with a rating label comprises the steps of determining the influence of the historical news data according to an influence calculation model and a news source average reading index, a news type popularity index, a news comment number index and a news reading index for different event numbers, and labeling the historical news data with a rating label according to the influence level.
6. The method of claim 1, wherein the trained impact computation model comprises:
NI＝(1+exp(-(ωarNar+ωthQth+ωscNsc+ωsrNsr-θ)))-1；
in the formula, NIThe final news influence; n is a radical ofarAverage reading number for news source; qthIs the news type popularity; n is a radical ofscNumber of news reviews for a stable period; n is a radical ofsrNews reading in stable periods; theta is a threshold parameter of the BP neural network output layer; omegaarRepresenting the weight of the BP neural network corresponding to the first index; omegathRepresenting the weight of the BP neural network corresponding to the second index; omegascRepresenting the weight of the BP neural network corresponding to the third index; omegasrRepresents a fourth index pairThe corresponding BP neural network weights.
7. A news influence calculation apparatus based on a BP neural network, comprising:
the data acquisition module is used for acquiring historical news data and news data to be calculated;
the index calibration module is used for determining each index of the influence calculation model, including a news source average reading index, a news type heat index, a news comment number index and a news reading index;
the model establishing module is used for establishing an influence calculation model by adopting a BP neural network according to the indexes determined by the index calibration module and the indexes;
the label calibration module is used for performing label calibration on the influence of the historical news data, namely dividing different historical news data according to the similarity of news contents and the number of news labels and dividing a plurality of event numbers; determining the influence of the historical news data through an influence calculation model according to each index from the historical news data with the same event number, and marking a grade label on the historical news data;
the BP neural network module is used for training the influence calculation model by using historical news data subjected to label calibration, namely the historical news data subjected to the level labeling;
and the calculation module is used for calculating the influence of the historical news data according to the influence calculation model, or calculating the influence of the news data to be calculated according to the trained influence calculation model, and outputting the corresponding news influence according to the grade label.
8. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method according to any of claims 1 to 6 when executing the program.
9. A computer-readable storage medium, on which a computer program is stored, the computer program being executable by a processor for implementing the method as claimed in any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010386342.7A CN111552882B (en) | 2020-05-09 | 2020-05-09 | News influence calculation method and device, computer equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010386342.7A CN111552882B (en) | 2020-05-09 | 2020-05-09 | News influence calculation method and device, computer equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111552882A CN111552882A (en) | 2020-08-18 |
| CN111552882B true CN111552882B (en) | 2022-07-01 |
Family
ID=72000511
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010386342.7A Active CN111552882B (en) | 2020-05-09 | 2020-05-09 | News influence calculation method and device, computer equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111552882B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112613296A (en) * | 2020-12-07 | 2021-04-06 | 深圳价值在线信息科技股份有限公司 | News importance degree acquisition method and device, terminal equipment and storage medium |
| CN113420946B (en) * | 2021-01-20 | 2024-02-09 | 广州麦媒信息科技有限公司 | News media evaluation method |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101409634A (en) * | 2007-10-10 | 2009-04-15 | 中国科学院自动化研究所 | Quantitative analysis tools and method for internet news influence based on information retrieval |
| CN103617279A (en) * | 2013-12-09 | 2014-03-05 | 南京邮电大学 | Method for achieving microblog information spreading influence assessment model on basis of Pagerank method |
| CN107153908A (en) * | 2017-03-24 | 2017-09-12 | 国家计算机网络与信息安全管理中心 | Mobile news App influence power ranking methods |
| CN107577688A (en) * | 2017-04-25 | 2018-01-12 | 上海市互联网信息办公室 | Original article influence power analysis system based on media information collection |
| CN107870957A (en) * | 2016-09-28 | 2018-04-03 | 郑州大学 | A kind of popular microblogging Forecasting Methodology based on information gain and BP neural network |
| CN108876058A (en) * | 2018-07-27 | 2018-11-23 | 南京航空航天大学 | A kind of media event influence force prediction method based on microblogging |
| CN109145090A (en) * | 2018-08-31 | 2019-01-04 | 东北大学 | A kind of microblog users emotion influence power analysis method based on deep learning |
| CN109299884A (en) * | 2018-10-19 | 2019-02-01 | 北京网智天元大数据科技有限公司 | A kind of influence power appraisal procedure and assessment device |
| CN109670032A (en) * | 2019-01-09 | 2019-04-23 | 合肥工业大学 | A kind of microblogging forward direction influence power seniority among brothers and sisters method neural network based |
| WO2019183191A1 (en) * | 2018-03-22 | 2019-09-26 | Michael Bronstein | Method of news evaluation in social media networks |
| CN110727798A (en) * | 2019-10-15 | 2020-01-24 | 东北大学 | Festival emotion analysis method based on naive Bayes classification |
-
2020
- 2020-05-09 CN CN202010386342.7A patent/CN111552882B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101409634A (en) * | 2007-10-10 | 2009-04-15 | 中国科学院自动化研究所 | Quantitative analysis tools and method for internet news influence based on information retrieval |
| CN103617279A (en) * | 2013-12-09 | 2014-03-05 | 南京邮电大学 | Method for achieving microblog information spreading influence assessment model on basis of Pagerank method |
| CN107870957A (en) * | 2016-09-28 | 2018-04-03 | 郑州大学 | A kind of popular microblogging Forecasting Methodology based on information gain and BP neural network |
| CN107153908A (en) * | 2017-03-24 | 2017-09-12 | 国家计算机网络与信息安全管理中心 | Mobile news App influence power ranking methods |
| CN107577688A (en) * | 2017-04-25 | 2018-01-12 | 上海市互联网信息办公室 | Original article influence power analysis system based on media information collection |
| WO2019183191A1 (en) * | 2018-03-22 | 2019-09-26 | Michael Bronstein | Method of news evaluation in social media networks |
| CN108876058A (en) * | 2018-07-27 | 2018-11-23 | 南京航空航天大学 | A kind of media event influence force prediction method based on microblogging |
| CN109145090A (en) * | 2018-08-31 | 2019-01-04 | 东北大学 | A kind of microblog users emotion influence power analysis method based on deep learning |
| CN109299884A (en) * | 2018-10-19 | 2019-02-01 | 北京网智天元大数据科技有限公司 | A kind of influence power appraisal procedure and assessment device |
| CN109670032A (en) * | 2019-01-09 | 2019-04-23 | 合肥工业大学 | A kind of microblogging forward direction influence power seniority among brothers and sisters method neural network based |
| CN110727798A (en) * | 2019-10-15 | 2020-01-24 | 东北大学 | Festival emotion analysis method based on naive Bayes classification |
Non-Patent Citations (5)
| Title |
|---|
| "Spread Influence Algorithm of News Website Based on PageRank";GuoWei Chen 等;《 2015 IEEE/ACIS 14th International Conference on Computer and Information Science (ICIS)》;20150727;1-4 * |
| "一种基于信息检索技术的网络新闻影响力分析方法";杨伟杰 等;《软件学报》;20090915;第20卷(第9期);2397-2406 * |
| "基于线性加权的微博影响力量化模型";孙国梓 等;《四川大学学报(工程科学版)》;20160111;第48卷(第1期);78-84 * |
| "基于聚类算法的微博影响力评价模型研究与实现";张剑金;《中国优秀硕士学位论文全文数据库 信息科技辑》;20150115(第01期);I138-839 * |
| "突发公共卫生事件微博影响力的预测研究";安璐 等;《情报理论与实践》;20170809;第40卷(第08期);76-81 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111552882A (en) | 2020-08-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kaliyar | Fake news detection using a deep neural network | |
| Liu et al. | An interactive method to improve crowdsourced annotations | |
| Yin et al. | Deep fusion of multimodal features for social media retweet time prediction | |
| Zhou et al. | Userrec: A user recommendation framework in social tagging systems | |
| CN104346425B (en) | A kind of method and system of the internet public feelings index system of stratification | |
| CN111831905B (en) | Recommendation method and device based on team scientific research influence and sustainability modeling | |
| Uddin et al. | Predicting the popularity of online news from content metadata | |
| CN111552882B (en) | News influence calculation method and device, computer equipment and storage medium | |
| CN112419029B (en) | Similar financial institution risk monitoring method, risk simulation system and storage medium | |
| CN109934469A (en) | Based on the heterologous power failure susceptibility method for early warning and device for intersecting regression analysis | |
| Kwon et al. | Disentangling user samples: A supervised machine learning approach to proxy-population mismatch in Twitter research | |
| CN113343126A (en) | Rumor detection method based on event and propagation structure | |
| Dong et al. | A two-step rumor detection model based on the supernetwork theory about Weibo | |
| CN116775882B (en) | Intelligent government affair message processing method and equipment | |
| CN110543601B (en) | Method and system for recommending context-aware interest points based on intelligent set | |
| Zhang et al. | SusRec: An Approach to Sustainable Developer Recommendation for Bug Resolution Using Multimodal Ensemble Learning | |
| Lei | Association Rule Mining Algorithm in College Students’ Quality Evaluation System | |
| Liao et al. | TIRR: A code reviewer recommendation algorithm with topic model and reviewer influence | |
| CN115310004A (en) | Graph nerve collaborative filtering recommendation method fusing project time sequence relation | |
| Guo et al. | Learning resource recommendation based on generalized matrix factorization and long short-term memory model | |
| Rezaeenour et al. | Developing a new hybrid intelligent approach for prediction online news popularity | |
| Tang et al. | Domain problem‐solving expert identification in community question answering | |
| Wang et al. | S-DeepTrust: A deep trust prediction method based on sentiment polarity perception | |
| De Silva | A Network Analysis Based Credibility Ranking Model to Combat Misinformation on Twitter | |
| Samizadeh et al. | Web mining based on word-centric search with clustering approach using MLP-PSO hybrid |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |