Intelligent manuscript issuing system and method

Abstract

The invention provides an intelligent manuscript issuing system and method, and belongs to the field of intelligent pushing. The problem of low manuscript pushing efficiency is solved; the method specifically comprises the following steps: a manuscript acquisition module acquires a to-be-pushed manuscript; the manuscript processing module is aligned with the pushed manuscripts for primary classification, and preferential manuscripts are extracted; performing secondary classification on the pushed manuscripts to obtain interested manuscripts; adding a picture to the interested manuscript according to the content in the interested manuscript to obtain a manuscript to be pushed; the manuscript promotion module is combined with the manuscript to be pushed to calculate the additional supplement amount of the manuscript to be pushed; acquiring page views of different types of manuscripts of each user, calculating acceptance values of the different types of manuscripts of each user, and pushing the manuscripts to be pushed; the manuscript updating module updates the original manuscript data and pushes the manuscript; according to the method, the manuscripts are acquired, screened, classified and pushed, so that the user can read the high-quality manuscripts.

Description

Technical Field

[0001] The present invention discloses an intelligent manuscript publishing system and method, and relates to the field of intelligent push. Background Art

[0002] The existing intelligent publishing systems and methods have the following deficiencies:

[0003] Technology dependence and talent shortage: The existing publishing system is highly dependent on advanced technology and algorithms, which requires media companies to continuously invest resources in technology research and development and optimization; currently, there is a large demand for professional talents in the field of artificial intelligence in China, but the supply is insufficient, which may lead to companies facing a shortage of technical talents in the short term.

[0004] Limited scope of application: The existing publishing system is mainly suitable for the push of manuscripts with weak logic and fixed forms in the fields of finance, sports, emergencies, etc. It is impossible to accurately push manuscripts with diverse forms and contents, especially those focusing on current affairs, science and technology, popular science, culture, education, health, economy, etc., resulting in a poor user experience.

[0005] Unstable content quality: Although the existing publishing system can push relevant content based on user interests and behavior data, the quality of the pushed content is uneven. This is mainly due to the limited algorithm and data processing capabilities of the system. It can only match based on keywords and tags, but cannot deeply understand the content of the article, resulting in a deviation between the pushed content and the actual needs of users, or the pushed content is too superficial and lacks depth. Summary of the invention

[0006] In view of the deficiencies in the prior art, the present invention aims to provide an intelligent manuscript publishing system and method, aiming to solve the problem of low manuscript push efficiency.

[0007] In order to achieve the above object, the present invention is implemented through the following technical solutions: an intelligent manuscript publishing system includes:

[0008] Manuscript acquisition module: used to obtain original manuscripts and submission information of original manuscripts, remove manuscripts and submission information with bad content, and obtain manuscripts and submission information that are ready to be pushed;

[0009] Manuscript processing module: including manuscript classification submodule and manuscript illustration submodule;

[0010] Manuscript classification submodule: used to build a news recognition model, and use the news recognition model to classify the manuscripts to be pushed, extract the news broadcast manuscripts as the priority manuscripts;

[0011] Change the training object of the news recognition model and build an interest recognition model; perform secondary classification on the articles to be pushed according to the interest recognition model, add interest tags to non-news broadcast articles, and obtain articles of interest;

[0012] Manuscript picture submodule: Use natural language processing technology to identify the content in the manuscript of interest, and add pictures to the manuscript of interest according to the content in the manuscript of interest, so as to obtain the manuscript to be pushed;

[0013] Manuscript promotion module: including manuscript statistics submodule and manuscript push submodule;

[0014] Manuscript statistics submodule: used to count the historical views of different types of manuscripts to be pushed, and calculate the additional supplementary amount of manuscripts to be pushed in combination with the manuscripts to be pushed;

[0015] Collect the information of prospective contributors with continuous submission records as manuscript source information; obtain the number of views of news articles and articles of interest corresponding to the manuscript source information in the past three months to determine the high-quality manuscript source information;

[0016] Manuscript push submodule: used to obtain the number of all users and push priority manuscripts to all users; obtain the number of views of different types of manuscripts of each user, calculate the acceptance value of each user for different types of manuscripts, and recommend manuscripts to be pushed to users based on the user's acceptance value;

[0017] Manuscript update module: used to obtain manuscripts with high-quality source information and manuscripts with non-high-quality source information in real time, update original manuscript data, and push manuscripts.

[0018] Furthermore, the workflow of the manuscript classification submodule is as follows:

[0019] Process B1: Count the number of quasi-submission information, recorded as si; count the number of quasi-pushed manuscripts corresponding to each quasi-submission information, and get do1, do2~do si ;

[0020] Among them, do1 represents the number of manuscripts that are expected to be pushed corresponding to the first submission information; do2 represents the number of manuscripts that are expected to be pushed corresponding to the second submission information; and so on. si Indicates the number of manuscripts that are expected to be pushed corresponding to the si-th expected submission information;

[0021] Process B2: Get news articles as training articles a1 and non-news articles as training articles a2, and use the logistic regression model to build a news recognition model;

[0022] Process B3: Use the joblib library to load the quasi-pushing manuscripts corresponding to the 1st to sith quasi-submission information into the news recognition model, classify the quasi-pushing manuscripts once, and extract the news broadcast manuscripts corresponding to each quasi-submission information as the priority manuscripts;

[0023] Count the number of priority manuscripts corresponding to each prospective submission information, and get: ne1, ne2~ne si ;

[0024] Among them, ne1 represents the number of priority manuscripts corresponding to the first quasi-submission information; ne2 represents the number of priority manuscripts corresponding to the second quasi-submission information; and so on, ne si Indicates the number of priority manuscripts corresponding to the si-th prospective submission information;

[0025] Process B4: Treat the manuscripts that are not priority manuscripts as manuscripts to be classified, and define interest tags for the manuscripts to be classified. The interest tags include: technology, life, shopping, entertainment and others;

[0026] Change the training object of the news recognition model and build an interest recognition model; perform secondary classification on the manuscripts to be classified according to the interest recognition model, and count the number of scientific and technological manuscripts in the manuscripts to be classified corresponding to the 1st to sith quasi-submission information, and obtain: sc1, sc2~sc si ;

[0027] The number of life manuscripts is: li1, li2~li si ;

[0028] The number of shopping manuscripts is: sh1, sh2~sh si ;

[0029] The number of entertainment articles: re1, re2~re si ;

[0030] The number of other manuscripts is: ot1, ot2~ot si ;

[0031] Among them, sc1, li1, sh1, re1 and ot1 respectively represent the number of scientific manuscripts, life manuscripts, shopping manuscripts, entertainment manuscripts and other manuscripts corresponding to the first quasi-submission information;

[0032] sc2, li2, sh2, re2, and ot2 respectively represent the number of scientific manuscripts, life manuscripts, shopping manuscripts, entertainment manuscripts, and other manuscripts corresponding to the second quasi-submission information;

[0033] By analogy, sc si ,li si ,sh si 、re siandot si , respectively represent the number of scientific manuscripts, life manuscripts, shopping manuscripts, entertainment manuscripts and other manuscripts corresponding to the si-th prospective submission information;

[0034] Process B5: Send the scientific manuscripts, life manuscripts, shopping manuscripts, entertainment manuscripts and other manuscripts corresponding to the 1st to sith quasi-submission information as interest manuscripts to the manuscript illustration submodule for illustration, obtain the manuscripts to be pushed, and enter the manuscript statistics submodule.

[0035] Furthermore, the specific process of process B2 is as follows:

[0036] Process B21: Import the pandas library and the scikit-learn library;

[0037] Create a list yes_texts to store training manuscript a1; create a list no_texts to store training manuscript a2;

[0038] Combine news and non-news article text as feature X, that is:

[0039] X=news_texts+non_news_texts;

[0040] Define the label y, mark news articles as 1 and non-news articles as 0, that is:

[0041] y=[1]*len(yes_texts)+[0]*len(no_texts);

[0042] Process B22: Use the load_data function to load the feature X and label y as the combined data set; divide the combined data set into a training set and a test set;

[0043] The text of the training set is denoted as X_train, and the label is denoted as y_train;

[0044] The text of the test set is denoted as X_test, and the label is denoted as y_test;

[0045] The number of training sets and test sets satisfies: training set > test set;

[0046] Process B23: Use the manuscripts in the training set as training texts and the manuscripts in the test set as test texts; initialize the TF-IDF vectorizer and perform TF-IDF feature extraction on the training set X_train;

[0047] Use the fit_transform method to learn the vocabulary in the training text, calculate the TF-IDF value of each word in each training text, and get the sparse matrix X_train_tfidf;

[0048] Each row in the sparse matrix X_train_tfidf represents a training text, each column represents a word, and the values ​​in the matrix correspond to the TF-IDF values ​​of the word in different training texts;

[0049] Use the transform method to extract TF-IDF features from the test set text to obtain the training feature X_test_tfidf;

[0050] Process B24: Initialize the logistic regression model model and train the model with the training set data, that is:

[0051] model.fit(X_train_tfidf,y_train);

[0052] Use the trained model to predict the test set, that is:

[0053] y_pred=model.predict(X_test_tfidf);

[0054] Use the accuracy_score() method to calculate the accuracy of the model on the test set; and use the classification_report() method to calculate the precision until the accuracy and precision between any article with label y_train=1 in the training set and any article with label y_test=1 in the test set are close to 100%; the accuracy and precision between the three types of articles, namely, the article with label y_train=1 and the article with label y_test=0, the article with label y_train=0 and the article with label y_test=1, and the article with label y_train=0 and the article with label y_test=0, are all 0%, completing the training of the news recognition model;

[0055] Load the joblib library to save the news recognition model and load manuscripts for the news recognition model.

[0056] Furthermore, the workflow of the manuscript statistics submodule is as follows:

[0057] Process C1: Count the views of scientific manuscripts in the past three months and get: hsc3, hsc2, and hsc1;

[0058] The number of views of life articles is: hli3, hli2 and hli1;

[0059] The number of views of shopping articles is: hsh3, hsh2 and hsh1;

[0060] The number of views of entertainment articles is: hre3, hre2 and hre1;

[0061] The views of other manuscripts are: hot3, hot2 and hot1;

[0062] Among them, hsc3, hli3, hsh3, hre3 and hot3 represent the views of science and technology articles, life articles, shopping articles, entertainment articles and other articles in the third month;

[0063] hsc2, hli2, hsh2, hre2 and hot2 represent the views of science and technology articles, life articles, shopping articles, entertainment articles and other articles in the second month;

[0064] hsc1, hli1, hsh1, hre1 and hot1 represent the number of views of science and technology articles, life articles, shopping articles, entertainment articles and other articles in the first month;

[0065] Process C2: Calculate the growth rate of scientific manuscripts from the third month to the second month, denoted as usc (3-2) ;usc (3-2) The calculation formula is: usc (3-2) =(hsc2-hsc3) / hsc3;

[0066] The growth rate of life articles from the third month to the second month, denoted as uli (3-2) uli (3-2) The calculation formula is: (3-2) =(hli2-hli3) / hli3;

[0067] The growth rate of shopping articles from the third month to the second month, denoted as ush (3-2) ;ush (3-2) The calculation formula is: (3-2) =(hsh2-hsh3) / hsh3;

[0068] The growth rate of entertainment articles from the third month to the second month, denoted as ure (3-2) ure (3-2) The calculation formula is:ure (3-2) =(hre2-hre3) / hre3;

[0069] The growth rate of other manuscripts from the third month to the second month is denoted as uot (3-2) ;uot (3-2) The calculation formula is: uot (3-2) =(hot2-hot3) / hot3.

[0070] Furthermore, the subsequent process of process C2 is as follows:

[0071] Process C3: Calculate the growth rate of scientific manuscripts from the second month to the first month, denoted as usc (2-1) ;usc (2-1) The calculation formula is: usc (2-1) =(hsc1-hsc2) / hsc2;

[0072] The growth rate of life articles from the second month to the first month, denoted as uli (2-1) uli (2-1) The calculation formula is: (2-1) =(hli1-hli2) / hli2;

[0073] The growth rate of shopping articles from the second month to the first month, denoted as ush (2-1) ;ush (2-1) The calculation formula is: (2-1) =(hsh1-hsh2) / hsh2;

[0074] The growth rate of entertainment articles from the second month to the first month, denoted as ure (2-1) ure (2-1) The calculation formula is:ure (2-1) =(hre1-hre2) / hre2;

[0075] The growth rate of other manuscripts from the second month to the first month, denoted as uot (2-1) ;uot (2-1) The calculation formula is: uot (2-1) =(hot1-hot2) / hot2;

[0076] Process C4: Calculation of sc1~sc si The sum of is denoted asc; li1, li2~li si The sum of is written as ali; sh1, sh2~sh si The sum of re1, re2 and re si The sum of is written as are; ot1, ot2~ot si The sum of is written as aot;

[0077] With {usc (3-2) , usc (2-1) , hsc2, hsc1} is group a1, with {uli (3-2) ,uli (2-1) , hli2, hli1} is group a2, with {ush (3-2) ,ush (2-1) , hsh2, hsh1} is group a3, with {ure (3-2) ,ure (2-1), hre2, hre1} is group a4, with {uot (3-2) ,uot (2-1) , hot2, hot1} is group a5;

[0078] Using the state transfer matrix, the expected number of scientific manuscripts submitted is calculated based on group a1, recorded as Esc; the expected number of scientific manuscripts submitted is calculated based on group a2, recorded as Eli; the expected number of scientific manuscripts submitted is calculated based on group a3, recorded as Esh; the expected number of scientific manuscripts submitted is calculated based on group a4, recorded as Ere; the expected number of scientific manuscripts submitted is calculated based on group a5, recorded as Eot;

[0079] Process C5: Compare asc and Esc. If asc≤Esc, then Δsc=(Esc-asc); if asc>Esc, then Δsc=0; where Δsc represents the number of additional supplementary manuscripts of scientific and technological manuscripts;

[0080] Compare ali and Eli. If ali≤Eli, then Δli=(Eli-ali); if ali>Eli, then Δli=0; where Δli represents the number of additional supplementary manuscripts for life manuscripts;

[0081] Compare ash and Esh. If ash≤Esh, then Δsh=(Esh-ash); if ash>Esh, then Δsh=0; where Δsh represents the number of additional supplementary manuscripts for shopping manuscripts;

[0082] Compare are and Ere. If are ≤ Ere, then Δre = (Ere - are); if are > Ere, then Δre = 0; where Δre represents the number of additional supplementary articles for entertainment articles.

[0083] Compare aot and Eot. If aot≤Eot, then Δot=(Eot-aot); if aot>Eot, then Δot=0; where Δot represents the number of additional supplementary manuscripts for other manuscripts;

[0084] According to Δsc, Δli, Δsh, Δre and Δot, supplement the corresponding types of manuscripts;

[0085] Process C6: Collect the information of prospective manuscripts with continuous submission records in the past three months as manuscript source information;

[0086] The number of manuscript source information is counted, denoted as ni; the relationship between ni and si satisfies: ni≤si; si represents the number of quasi-submission information;

[0087] Obtain the number of news article views and interest article views corresponding to the manuscript source information in the past three months, determine the high-quality manuscript source information, and enter the manuscript push sub-module.

[0088] Furthermore, the specific process of process C6 is as follows:

[0089] Process C61: Obtain the news article page views corresponding to the source information in the past three months and determine the high-quality source information of the news articles;

[0090] Get the news article views corresponding to the source information in the past three months, and get ne(1,3), ne(1,2), ne(1,1)~ne(ni,1);

[0091] Among them, ne(1,3) represents the number of news article views in the third month corresponding to the first source information; ne(1,2) represents the number of news article views in the second month corresponding to the first source information; ne(1,1) represents the number of news article views in the first month corresponding to the first source information;

[0092] By analogy, ne(ni,1) represents the number of news article views in the first month corresponding to the ni-th source information;

[0093] Process C62: Calculate the sum of the number of news article views in the third month corresponding to the first to the nith source information, and record it as ane(-, 3); (i.e., the sum of ne(1,3), ne(2,3) to ne(ni,3))

[0094] The sum of the number of news article views in the second month is recorded as ane(-, 2); (i.e. the sum of ne(1,2), ne(2,2) to ne(ni,2))

[0095] The sum of the number of news article views in the first month is recorded as ane(-, 1); (i.e. the sum of ne(1,1), ne(2,1) to ne(ni,1))

[0096] Process C63: Calculate the continuous change coefficient of the number of views of the first source information news article, recorded as up (1) ; up (1) The calculation formula is:

[0097]

[0098] Calculate the continuous change coefficient of the second source news article pageviews, recorded as up (2) ; up (2) The calculation formula is:

[0099]

[0100] Similarly, calculate the continuous change coefficient of the number of views of the ni-th source information news article, recorded as up (ni) ; up (ni) The calculation formula is:

[0101]

[0102] Process C64: Obtain the number of priority manuscripts corresponding to the manuscript source information from the 1st to the nith, and obtain: nne1, nne2~nne ni ;nne1,nne2~nne ni With ne1, ne2~ne si The relationship satisfies:

[0103] Among them, nne1 represents the number of priority manuscripts corresponding to the first manuscript source information; nne2 represents the number of priority manuscripts corresponding to the second manuscript source information; and so on, nne ni Indicates the number of priority manuscripts corresponding to the ni-th manuscript source information;

[0104] Process C65: Calculate the news priority coefficient of the first source information, denoted as fne (1) ;fne (1) The calculation formula is: fne (1) =up (1) ×nne1;

[0105] Calculate the news priority coefficient of the second source information, denoted as fne (2) ;fne (2) The calculation formula is: fne (2) =up (2) ×nne2;

[0106] Similarly, calculate the news priority coefficient of the ni-th source information, denoted as fne (ni) ;fne (ni) The calculation formula is: fne (ni) =up (ni) ×nne ni ;

[0107] Process C66: Extraction of fne (1) ~fne (ni) The largest news priority coefficient is recorded as nem; the source information corresponding to nem is used as the high-quality source information of the news article;

[0108] Process C67: Repeat the same process of determining high-quality source information of news articles to determine the high-quality source information corresponding to each article of interest.

[0109] Furthermore, the workflow of the manuscript push submodule is as follows:

[0110] Process D1: Get the number of all users, denoted as us; push priority manuscripts to all users;

[0111] Get the number of views of each user's scientific articles in the past three months, and get: usc(1,3), usc(1,2), usc(1,1)~usc(us,1);

[0112] Among them, usc(1,1) represents the number of views of the scientific articles of the first user in the third month; usc(1,2) represents the number of views of the scientific articles of the first user in the second month; usc(1,1) represents the number of views of the scientific articles of the first user in the first month;

[0113] By analogy, usc(us,1) represents the number of views of scientific articles by the us-th user in the first month;

[0114] Get the number of views of each user's life articles in the past three months, and get: uli(1,3), uli(1,2), uli(1,1)~uli(us,1);

[0115] Among them, uli(1,1) represents the number of views of the life manuscript of the first user in the third month; uli(1,2) represents the number of views of the life manuscript of the first user in the second month; uli(1,1) represents the number of views of the life manuscript of the first user in the first month;

[0116] By analogy, uli(us,1) represents the number of views of the life articles of the us-th user in the first month;

[0117] Get the number of views of shopping articles of each user in the past three months, and get: ush(1,3), ush(1,2), ush(1,1)~ush(us,1);

[0118] Among them, ush(1,3) represents the number of views of the shopping manuscript of the first user in the third month; ush(1,2) represents the number of views of the shopping manuscript of the first user in the second month; ush(1,1) represents the number of views of the shopping manuscript of the first user in the first month;

[0119] By analogy, ush(us,1) represents the number of views of the shopping articles of the us-th user in the first month;

[0120] Get the number of views of entertainment articles of each user in the past three months, and get: ure(1,3), ure(1,2), ure(1,1)~ure(us,1);

[0121] Among them, ure(1,3) represents the number of page views of the entertainment articles of the first user in the third month; ure(1,2) represents the number of page views of the entertainment articles of the first user in the second month; ure(1,1) represents the number of page views of the entertainment articles of the first user in the first month;

[0122] By analogy, ure(us,1) represents the number of entertainment article views by the us-th user in the first month;

[0123] Get the page views of other articles of each user in the past three months, and get: uot(1,3), uot(1,2), uot(1,1)~uot(us,1);

[0124] Among them, uot(1,3) represents the number of views of other articles of the first user in the third month; uot(1,2) represents the number of views of other articles of the first user in the second month; uot(1,1) represents the number of views of other articles of the first user in the first month;

[0125] By analogy, uot(us,1) represents the number of views of other articles by the us-th user in the first month;

[0126] It should be noted that the "user" in the manuscript push submodule means: a person or unit that accepts the manuscript push service of the present invention;

[0127] Process D2: Calculate the average number of scientific article views of all users in the past three months, recorded as msc;

[0128] The average number of views of life articles in the past three months is recorded as mli;

[0129] The average number of shopping article views in the past three months, denoted as msh;

[0130] The average number of entertainment article views in the past three months, recorded as mre;

[0131] The average number of views of other manuscripts in the past three months is recorded as mot;

[0132] Process D3: Take the number of views of the first user's science and technology articles, life articles, shopping articles, entertainment articles and other articles in the past three months as sample data, and construct a sample matrix, recorded as matrix o; the mathematical expression of matrix o is:

[0133]

[0134] Calculate the acceptance value of the first user for different types of manuscripts according to the matrix o, and push the manuscripts to be pushed to the first user;

[0135] Process D4: Repeat the same process of pushing the manuscript to be pushed for the first user, and push the manuscript to be pushed for the second to us users.

[0136] Furthermore, the specific process of the process D3 is as follows:

[0137] Process D31: Define calculation formula d1-1: X (1,j)=1-[usc(1,j) / (msc / 3)];

[0138] Calculation formula d1-2: Y (1,j) =1-[uli(1,j) / (mli / 3)];

[0139] Calculation formula d1-3: Z (1,j) =1-[ush(1,j) / (msh / 3)];

[0140] Calculation formula d1-4: P (1,j) =1-[ure(1,j) / (mre / 3)];

[0141] Calculation formula d1-5: Q (1,j) =1-[uot(1,j) / (mot / 3)];

[0142] Among them, X (1,j) It represents the deviation of the reading volume of scientific articles of the first user in the jth month relative to the average reading volume of scientific articles of all users;

[0143] Y (1,j) It represents the deviation of the reading volume of the first user's life articles in the jth month relative to the average reading volume of all users' life articles;

[0144] Z (1,j) It represents the deviation of the number of shopping articles read by the first user in the jth month relative to the average number of shopping articles read by all users;

[0145] P (1,j) It represents the deviation of the entertainment article reading volume of the first user in the jth month relative to the average entertainment article reading volume of all users;

[0146] Q (1,j) It represents the deviation of the reading volume of other articles of the first user in the jth month relative to the average reading volume of other articles of all users;

[0147] The value range of j is: 1 to 3;

[0148] Process D32: According to the calculation formulas d1-1 to d1-5 in process D31, the offsets of the parameters in the matrix o are calculated to obtain the matrix o1; the mathematical expression of the matrix o1 is:

[0149]

[0150] Calculate X (1,1) , X (1,2) and X (1,3) The average value of (1,1) , Y (1,2) and Y (1,3) The average value is denoted as aY; Z(1,1) , Z (1,2) and Z (1,3) The average value is denoted as aZ; P (1,1) , P (1,2) and P (1,3) The average value is denoted as aP; Q (1,1) , Q (1,2) and Q (1,3) The average value of is denoted as aQ;

[0151] Process D33: Calculate X (1,1) , X (1,2) and X (1,3) The standard deviation of is denoted as SX; the calculation formula of SX is:

[0152]

[0153] Calculate Y (1,1) , Y (1,2) and Y (1,3) The standard deviation of is denoted as SY. The calculation formula of SY is:

[0154]

[0155] Calculate Z (1,1) , Z (1,2) and Z (1,3) The standard deviation of is denoted as SZ; the calculation formula of SZ is:

[0156]

[0157] Calculate P (1,1) , P (1,2) and P (1,3) The standard deviation of is denoted as SP; the calculation formula of SP is:

[0158]

[0159] Calculate Q (1,1) , Q (1,2) and Q (1,3) The standard deviation of is denoted as SQ; the calculation formula of SQ is:

[0160]

[0161] Process D34: Define calculation formula d2-1: dX (1,j) =(X (1,j) -aX) / SX;

[0162] Calculation formula d2-2: dY (1,j) =(Y (1,j) -aY) / SY;

[0163] Calculation formula d2-3: dZ(1,j) =(Z (1,j) -aZ) / SZ;

[0164] Calculation formula d2-4: dP (1,j) =(P (1,j) -aP) / SP;

[0165] Calculation formula d2-5: dQ (1,j) =(Q (1,j) -aQ) / SQ;

[0166] Among them, dX (1，j) Indicates X (1，j) The value of the second normalization, dY (1，j) Indicates Y (1，j) The value of the second normalization, dZ (1，j) Indicates Z (1，j) The value of the second normalization, dP (1，j) Indicates P (1，j) The value of the second normalization, dQ (1，j) Express Q (1，j) The value for secondary normalization; the value range of j is: 1 to 3.

[0167] Furthermore, the subsequent process of the process D34 is as follows:

[0168] Process D35: According to the calculation formulas d2-1 to d2-5 in process D34, the quadratic normalized values ​​of the parameters in the matrix o1 are calculated to obtain the matrix o2; the mathematical expression of the matrix o2 is:

[0169]

[0170] Calculate dX (1,1) 、dX (1,2) and dX (1,3) The average value, denoted as dX (1,-) ; Calculate dY (1,1) ,dY (1,2) and dY (1,3) The average value of (1,-) ; Calculate dZ (1,1) ,dZ (1,2) and dZ (1,3) The average value of (1,-) ; Calculate dP (1,1) , dP (1,2) and dP (1,3) The average value is denoted as dP (1,-) ; Calculate dQ (1,1) , dQ (1,2) and dQ (1,3) The average value of (1,-) ;

[0171] Calculate the average value of the parameters in the first row of matrix o2, denoted as dD (-，1) ;

[0172] Calculate the average value of the parameters in the second row of matrix o2, denoted as dD (-，2) ;

[0173] Calculate the average value of the parameters in the third row of matrix o2, denoted as dD (-，3) ;

[0174] Process D36: Define calculation formula d3-1:

[0175] Among them, rX (1，j) Denotes dX (1，j) The covariance of (-，j) Represents the average value of the jth row in matrix o2; the value range of j is: 1 to 3;

[0176] Calculation formula d3-2:

[0177] Among them, rY (1，j) indicates dY (1，j) The covariance of

[0178] Calculation formula d3-3:

[0179] Among them, rZ (1，j) Denotes dZ (1，j) The covariance of

[0180] Calculation formula d3-4:

[0181] Among them, rP (1，j) Indicates dP (1，j) The covariance of

[0182] Calculation formula d3-5:

[0183] Among them, rQ (1，j) Indicates dQ (1，j) The covariance of

[0184] Process D37: According to the calculation formulas d3-1 to d3-5 in process D36, the covariance of each parameter in the matrix o2 is calculated to obtain the matrix o3; the mathematical expression of the matrix o3 is:

[0185]

[0186] Split the matrix o3 to obtain matrices o4-1 to o4-5;

[0187] The mathematical expression of matrix o4-1 is:

[0188]

[0189] The mathematical expression of matrix o4-2 is:

[0190]

[0191] The mathematical expression of matrix o4-3 is:

[0192]

[0193] The mathematical expression of matrix o4-4 is:

[0194]

[0195] The mathematical expression of matrix o4-5 is:

[0196]

[0197] Process D38: Use Intel MKL library functions to define and calculate weight values ​​λ1, λ2, λ3, λ4, and λ54 and offsets μ1, μ2, μ3, μ4, and μ3, so that matrices o4-1 to o4-5 satisfy the following mathematical relationship:

[0198]

[0199] Process D39: Calculate the acceptance value of the first user's scientific manuscript and record it as bsc, bsc = λ1 / μ1;

[0200] Calculate the acceptance value of the first user's life manuscript and record it as bli, bli = λ2 / μ2;

[0201] Calculate the acceptance value of the first user's shopping manuscript and record it as bsh, bsh = λ3 / μ3;

[0202] Calculate the acceptance value of the first user's entertainment manuscript and record it as bre, bre = λ4 / μ4;

[0203] Calculate the acceptance value of other manuscripts of the first user and record it as bot, bot = λ5 / μ5;

[0204] Arrange bsc, bli, bsh, bre and bot in descending order to obtain the interest sequence of the first user, and push the articles to be pushed to the first user in the order of the interest sequence.

[0205] An intelligent publishing method includes:

[0206] Step S1: obtaining original manuscripts and submission information of original manuscripts, eliminating manuscripts and submission information with bad content, and obtaining manuscripts to be pushed and submission information to be submitted;

[0207] Step S2: construct a news recognition model, and use the news recognition model to classify the manuscripts to be pushed, extract the news broadcast manuscripts as the priority manuscripts;

[0208] Change the training object of the news recognition model and build an interest recognition model; perform secondary classification on the articles to be pushed according to the interest recognition model, add interest tags to non-news broadcast articles, and obtain articles of interest;

[0209] Using natural language processing technology, the content in the manuscripts of interest is identified, and pictures are added to the manuscripts of interest according to the content in the manuscripts of interest, so as to obtain the manuscripts to be pushed;

[0210] Step S3: Count the historical views of different types of manuscripts to be pushed, and calculate the additional supplementary amount of manuscripts to be pushed in combination with the manuscripts to be pushed;

[0211] Collect the information of prospective contributors with continuous submission records as manuscript source information; obtain the number of views of news articles and articles of interest corresponding to the manuscript source information in the past three months to determine the high-quality manuscript source information;

[0212] Obtain the number of all users and push priority manuscripts to all users; obtain the number of views of different types of manuscripts for each user, calculate the acceptance value of each user for different types of manuscripts, and recommend manuscripts to be pushed to users based on the user's acceptance value;

[0213] Step S4: Obtain manuscripts with high-quality manuscript source information and manuscripts with non-high-quality manuscript source information in real time, update original manuscript data, and push manuscripts.

[0214] Compared with the prior art, the present invention has the following beneficial effects:

[0215] Precise delivery: The present invention has powerful data analysis capabilities and is designed for multiple fields including but not limited to news, technology, life, shopping, etc. In the process of screening manuscripts for users, it also analyzes user needs and the acceptance of different manuscripts based on the reading habits of different users, and selects different media resources for users to achieve precise delivery and wide coverage.

[0216] Improve production efficiency: The present invention can process massive amounts of data in a short period of time, collect, screen, organize, illustrate and push manuscripts, simplify the complicated manuscript processing links, and greatly improve the dissemination speed of traditional media; at the same time, compared with traditional manual creation, the present invention can save labor costs and reduce the operating costs of enterprises.

[0217] Data monitoring and feedback: The present invention has data monitoring and feedback functions. Media workers can understand the publishing effect of manuscripts and the reading tendency of users in a certain period of time in real time, and screen high-quality manuscript sources for media workers. This processing method helps media workers adjust and optimize subsequent manuscript content. BRIEF DESCRIPTION OF THE DRAWINGS

[0218] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0219] Figure 1 It is a schematic diagram of the system of the present invention;

[0220] Figure 2 It is a schematic diagram of the method of the present invention;

[0221] Figure 3 This is a schematic diagram of the logic design of the present invention. DETAILED DESCRIPTION

[0222] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0223] Embodiment 1

[0224] See also Figure 1 and Figure 3 , an intelligent manuscript publishing system includes: a manuscript acquisition module, a manuscript processing module, a manuscript promotion module, a manuscript update module, a database and a server; wherein the manuscript acquisition module, the manuscript processing module, the manuscript promotion module and the manuscript update module are connected to the database and the server respectively;

[0225] Among them, the database and server are used for data transmission and data calculation between the manuscript acquisition module, the manuscript processing module, the manuscript promotion module and the manuscript update module;

[0226] Manuscript acquisition module: used to obtain original manuscripts and submission information of original manuscripts, remove manuscripts and submission information with bad content, and obtain manuscripts and submission information that are ready to be pushed;

[0227] Submission information means: the identity information of the original manuscript contributor or the submitting organization;

[0228] It should be noted that the "original manuscript" in the present invention means: the manuscript pushed by the present invention (an intelligent manuscript publishing system and method); "submitting unit" means: a media organization;

[0229] Process A: The workflow of the manuscript acquisition module is as follows:

[0230] Process A1: Obtain the original manuscript and the submission information of the original manuscript;

[0231] Use the Worktile tool to remove manuscripts with bad content from the original manuscripts and obtain manuscripts that are ready to be pushed;

[0232] Mark submissions with inappropriate content as offenders and report to the police;

[0233] It should be noted that "bad content" in the present invention refers to content that is harmful to public security and social morality;

[0234] Process A2: Summarize the submission information of all manuscripts to be pushed as the quasi-submission information;

[0235] Count the number of quasi-submission information, quasi-submission information - summarize the manuscripts in the format of quasi-pushed manuscripts, obtain the manuscripts to be assigned, and enter the manuscript processing module.

[0236] Manuscript processing module: including manuscript classification submodule and manuscript illustration submodule;

[0237] Manuscript classification submodule: used to build a news recognition model, and use the news recognition model to classify the manuscripts to be pushed, extract the news broadcast manuscripts as the priority manuscripts;

[0238] Change the training object of the news recognition model and build an interest recognition model; perform secondary classification on the articles to be pushed according to the interest recognition model, add interest tags to non-news broadcast articles, and obtain articles of interest;

[0239] Process B: The workflow of the manuscript classification submodule is as follows:

[0240] Process B1: Count the number of quasi-submission information, recorded as si; count the number of quasi-pushed manuscripts corresponding to each quasi-submission information, and get do1, do2~do si ;

[0241] Among them, do1 represents the number of manuscripts that are expected to be pushed corresponding to the first submission information; do2 represents the number of manuscripts that are expected to be pushed corresponding to the second submission information; and so on. si Indicates the number of manuscripts that are expected to be pushed corresponding to the si-th expected submission information;

[0242] Process B2: Obtain (existing online) news articles as training articles a1, (existing online) non-news articles as training articles a2, and use the logistic regression model to build a news recognition model;

[0243] Process B21: Import the pandas library and the scikit-learn library;

[0244] Create a list yes_texts to store training manuscript a1; create a list no_texts to store training manuscript a2;

[0245] Combine news and non-news article text as feature X, that is:

[0246] X=news_texts+non_news_texts;

[0247] Define the label y, mark news articles as 1 and non-news articles as 0, that is:

[0248] y=[1]*len(yes_texts)+[0]*len(no_texts);

[0249] Process B22: Use the load_data function to load the feature X and label y as the combined data set; divide the combined data set into a training set and a test set;

[0250] The text of the training set is denoted as X_train, and the label is denoted as y_train;

[0251] The text of the test set is denoted as X_test, and the label is denoted as y_test;

[0252] The number of training sets and test sets satisfies: training set > test set;

[0253] Process B23: Use the manuscripts in the training set as training texts and the manuscripts in the test set as test texts; initialize the TF-IDF vectorizer and perform TF-IDF feature extraction on the training set X_train;

[0254] Use the fit_transform method to learn the vocabulary in the training text (that is, all non-repeated words), calculate the TF-IDF value of each word in each training text, and obtain the sparse matrix X_train_tfidf;

[0255] Each row in the sparse matrix X_train_tfidf represents a training text, each column represents a word, and the values ​​in the matrix correspond to the TF-IDF values ​​of the word in different training texts;

[0256] Use the transform method to extract TF-IDF features from the test set text to obtain the training feature X_test_tfidf; (It should be noted that the transform method is used here instead of fit_transform to ensure that the vocabulary used in the test set is consistent with the vocabulary used in the training set; the purpose of this is to avoid introducing new words that do not appear in the training set into the test set, and to ensure that the evaluation of the model on the test set is fair)

[0257] Process B24: Initialize the logistic regression model model and train the model with the training set data, that is:

[0258] model.fit(X_train_tfidf,y_train);

[0259] Use the trained model to predict the test set, that is:

[0260] y_pred=model.predict(X_test_tfidf);

[0261] Use the accuracy_score() method to calculate the accuracy of the model on the test set; and use the classification_report() method to calculate the precision until the accuracy and precision between any article with label y_train=1 in the training set and any article with label y_test=1 in the test set are close to 100%; the accuracy and precision between the three types of articles, namely, the article with label y_train=1 and the article with label y_test=0, the article with label y_train=0 and the article with label y_test=1, and the article with label y_train=0 and the article with label y_test=0, are all 0%, completing the training of the news recognition model;

[0262] Load the joblib library to save the (trained) news recognition model and load manuscripts for the news recognition model;

[0263] Process B3: Use the joblib library to load the quasi-pushing manuscripts corresponding to the 1st to sith quasi-submission information into the news recognition model, classify the quasi-pushing manuscripts once, and extract the news broadcast manuscripts corresponding to each quasi-submission information as the priority manuscripts;

[0264] Count the number of priority manuscripts (i.e. news broadcast manuscripts) corresponding to each quasi-submission information, and get: ne1, ne2~ne si ;

[0265] Among them, ne1 represents the number of priority manuscripts corresponding to the first quasi-submission information; ne2 represents the number of priority manuscripts corresponding to the second quasi-submission information; and so on, ne si Indicates the number of priority manuscripts corresponding to the si-th prospective submission information;

[0266] Process B4: Treat the manuscripts that are not priority manuscripts as manuscripts to be classified, and define interest tags for the manuscripts to be classified. The interest tags include: technology, life, shopping, entertainment and others;

[0267] The training object of the news recognition model is changed to build an interest recognition model; the manuscripts to be classified are secondary classified according to the interest recognition model, and interest tags are added to the manuscripts to be classified to obtain interest manuscripts; interest manuscripts include: science and technology manuscripts, life manuscripts, shopping manuscripts, entertainment manuscripts and other manuscripts;

[0268] It should be noted that "other" refers to articles "not related to science and technology, life, shopping, or entertainment";

[0269] Process B41: Obtain scientific and technological articles and non-scientific and technological articles (existing on the Internet), replace the training article a1 in process B2 with scientific and technological articles (existing on the Internet), and replace the training article a2 with non-scientific and technological articles (existing on the Internet), repeat the same process of building a news recognition model (i.e., process B21 and process B24), build a science and technology-interest recognition model, use the science and technology-interest recognition model to extract scientific and technological articles from the articles to be classified, and add scientific and technological labels;

[0270] Process B42: Obtain (existing on the Internet) life manuscripts and non-life manuscripts, shopping manuscripts and non-shopping manuscripts, and entertainment manuscripts and non-entertainment manuscripts, repeat the same process of building a technology-interest recognition model (as in process B41), and build a life-interest recognition model, a shopping-interest recognition model, and an entertainment-interest recognition model;

[0271] Use the life-interest recognition model, shopping-interest recognition model and entertainment-interest recognition model to extract life manuscripts, shopping manuscripts and entertainment manuscripts from the manuscripts to be classified, and add life, shopping and entertainment tags in sequence;

[0272] Process B43: Count the number of scientific manuscripts in the manuscripts to be classified corresponding to the 1st to sith quasi-submission information, and obtain: sc1, sc2~sc si ;

[0273] The number of life manuscripts is: li1, li2~li si ;

[0274] The number of shopping manuscripts is: sh1, sh2~sh si ;

[0275] The number of entertainment articles: re1, re2~re si ;

[0276] Among them, sc1, li1, sh1 and re1 respectively represent the number of scientific manuscripts, life manuscripts, shopping manuscripts and entertainment manuscripts corresponding to the first quasi-submission information;

[0277] sc2, li2, sh2, and re2 represent the number of scientific manuscripts, life manuscripts, shopping manuscripts, and entertainment manuscripts corresponding to the second quasi-submission information, respectively;

[0278] By analogy, sc si ,li si ,sh si and si , respectively represent the number of scientific manuscripts, life manuscripts, shopping manuscripts and entertainment manuscripts corresponding to the si-th prospective submission information;

[0279] Process B44: Calculate the number of other manuscripts ot1 corresponding to the first quasi-submission information, and the calculation formula of ot1 is: ot1=do1-(ne1+sc1+li1+sh1+re1);

[0280] Calculate the number of other manuscripts ot2 corresponding to the second quasi-submission information. The calculation formula of ot2 is: ot2 = do2 - (ne2 + sc2 + li2 + sh2 + re2);

[0281] Similarly, calculate the number of other manuscripts ot corresponding to the si-th quasi-submission information si ,ot si The calculation formula is: si =do si -(ne si +sc si +li si +sh si +re si );

[0282] The number of other manuscripts corresponding to the 1st to sith quasi-submission information is: ot1, ot2~ot si ;

[0283] Manuscript picture submodule: used to use natural language processing technology to identify the content in the manuscript of interest, and add pictures to the manuscript of interest according to the content in the manuscript of interest, so as to obtain the manuscript to be pushed;

[0284] Summarize the manuscripts to be pushed and enter the manuscript promotion module; the tags and quantity of the manuscripts to be pushed are the same as the tags and quantity of the manuscripts of interest

[0285] Manuscript promotion module: including manuscript statistics submodule and manuscript push submodule;

[0286] Manuscript Statistics Submodule: used to count the historical views of different types of non-news articles (in the past three months), and calculate the additional supplementary amount of articles to be pushed in combination with the articles to be pushed;

[0287] Collect statistics on prospective contributors with continuous contribution records (in the past three months) as manuscript source information; obtain the number of views of news articles and articles of interest corresponding to the manuscript source information in the past three months to determine high-quality manuscript source information;

[0288] Process C: The workflow of the manuscript statistics submodule is as follows:

[0289] Process C1: Count the views of scientific manuscripts in the past three months and get: hsc3, hsc2, and hsc1;

[0290] The number of views of life articles is: hli3, hli2 and hli1;

[0291] The number of views of shopping articles is: hsh3, hsh2 and hsh1;

[0292] The number of views of entertainment articles is: hre3, hre2 and hre1;

[0293] The views of other manuscripts are: hot3, hot2 and hot1;

[0294] Among them, hsc3, hli3, hsh3, hre3 and hot3 represent the views of science and technology articles, life articles, shopping articles, entertainment articles and other articles in the third month;

[0295] hsc2, hli2, hsh2, hre2 and hot2 represent the views of science and technology articles, life articles, shopping articles, entertainment articles and other articles in the second month;

[0296] hsc1, hli1, hsh1, hre1 and hot1 represent the number of views of science and technology articles, life articles, shopping articles, entertainment articles and other articles in the first month;

[0297] Process C2: Calculate the growth rate of scientific manuscripts from the third month to the second month, denoted as usc (3-2) ;usc (3-2) The calculation formula is: usc (3-2) =(hsc2-hsc3) / hsc3;

[0298] The growth rate of life articles from the third month to the second month, denoted as uli (3-2) uli (3-2) The calculation formula is: (3-2) =(hli2-hli3) / hli3;

[0299] The growth rate of shopping articles from the third month to the second month, denoted as ush (3-2) ;ush (3-2) The calculation formula is: (3-2)=(hsh2-hsh3) / hsh3;

[0300] The growth rate of entertainment articles from the third month to the second month, denoted as ure (3-2) ure (3-2) The calculation formula is:ure (3-2) =(hre2-hre3) / hre3;

[0301] The growth rate of other manuscripts from the third month to the second month is denoted as uot (3-2) ;uot (3-2) The calculation formula is: uot (3-2) =(hot2-hot3) / hot3;

[0302] Process C3: Calculate the growth rate of scientific manuscripts from the second month to the first month, denoted as usc (2-1) ;usc (2-1) The calculation formula is: usc (2-1) =(hsc1-hsc2) / hsc2;

[0303] The growth rate of life articles from the second month to the first month, denoted as uli (2-1) uli (2-1) The calculation formula is: (2-1) =(hli1-hli2) / hli2;

[0304] The growth rate of shopping articles from the second month to the first month, denoted as ush (2-1) ;ush (2-1) The calculation formula is: (2-1) =(hsh1-hsh2) / hsh2;

[0305] The growth rate of entertainment articles from the second month to the first month, denoted as ure (2-1) ure (2-1) The calculation formula is:ure (2-1) =(hre1-hre2) / hre2;

[0306] The growth rate of other manuscripts from the second month to the first month, denoted as uot (2-1) ;uot (2-1) The calculation formula is: uot (2-1) =(hot1-hot2) / hot2;

[0307] Process C4: Calculation of sc1~sc si The sum of is denoted asc; li1, li2~li si The sum of is written as ali; sh1, sh2~sh si The sum of re1, re2 and resi The sum of is written as are; ot1, ot2~ot si The sum of is written as aot;

[0308] With {usc (3-2) , usc (2-1) , hsc2, hsc1} is group a1, with {uli (3-2) ,uli (2-1) , hli2, hli1} is group a2, with {ush (3-2) ,ush (2-1) , hsh2, hsh1} is group a3, with {ure (3-2) ,ure (2-1) , hre2, hre1} is group a4, with {uot (3-2) ,uot (2-1) , hot2, hot1} is group a5;

[0309] Using the state transfer matrix, the expected number of scientific manuscripts submitted is calculated based on group a1, recorded as Esc; the expected number of scientific manuscripts submitted is calculated based on group a2, recorded as Eli; the expected number of scientific manuscripts submitted is calculated based on group a3, recorded as Esh; the expected number of scientific manuscripts submitted is calculated based on group a4, recorded as Ere; the expected number of scientific manuscripts submitted is calculated based on group a5, recorded as Eot;

[0310] Process C5: Compare asc and Esc. If asc≤Esc, then Δsc=(Esc-asc); if asc>Esc, then Δsc=0; where Δsc represents the number of additional supplementary manuscripts of scientific and technological manuscripts;

[0311] Compare ali and Eli. If ali≤Eli, then Δli=(Eli-ali); if ali>Eli, then Δli=0; where Δli represents the number of additional supplementary manuscripts for life manuscripts;

[0312] Compare ash and Esh. If ash≤Esh, then Δsh=(Esh-ash); if ash>Esh, then Δsh=0; where Δsh represents the number of additional supplementary manuscripts for shopping manuscripts;

[0313] Compare are and Ere. If are ≤ Ere, then Δre = (Ere - are); if are > Ere, then Δre = 0; where Δre represents the number of additional supplementary articles for entertainment articles.

[0314] Compare aot and Eot. If aot≤Eot, then Δot=(Eot-aot); if aot>Eot, then Δot=0; where Δot represents the number of additional supplementary manuscripts for other manuscripts;

[0315] According to Δsc, Δli, Δsh, Δre and Δot, add corresponding types of manuscripts (i.e. science and technology manuscripts, life manuscripts, shopping manuscripts, entertainment manuscripts and other manuscripts);

[0316] Process C6: Collect the information of prospective manuscripts with continuous submission records in the past three months as manuscript source information;

[0317] The number of manuscript source information is counted, denoted as ni; the relationship between ni and si satisfies: ni≤si; si represents the number of quasi-submission information;

[0318] Obtain the number of news article views and interest article views corresponding to the manuscript source information in the past three months to determine the high-quality manuscript source information;

[0319] Process C61: Obtain the news article page views corresponding to the source information in the past three months and determine the high-quality source information of the news articles;

[0320] Get the news article views corresponding to the source information in the past three months, and get ne(1,3), ne(1,2), ne(1,1)~ne(ni,1);

[0321] Among them, ne(1,3) represents the number of news article views in the third month corresponding to the first source information; ne(1,2) represents the number of news article views in the second month corresponding to the first source information; ne(1,1) represents the number of news article views in the first month corresponding to the first source information;

[0322] By analogy, ne(ni,1) represents the number of news article views in the first month corresponding to the ni-th source information;

[0323] Process C62: Calculate the sum of the number of news article views in the third month corresponding to the first to the nith source information, and record it as ane(-, 3); (i.e., the sum of ne(1,3), ne(2,3) to ne(ni,3))

[0324] The sum of the number of news article views in the second month is recorded as ane(-, 2); (i.e. the sum of ne(1,2), ne(2,2) to ne(ni,2))

[0325] The sum of the number of news article views in the first month is recorded as ane(-, 1); (i.e. the sum of ne(1,1), ne(2,1) to ne(ni,1))

[0326] Process C63: Calculate the continuous change coefficient of the number of views of the first source information news article, recorded as up (1) ; up (1) The calculation formula is:

[0327]

[0328] Calculate the continuous change coefficient of the second source news article pageviews, recorded as up (2) ; up (2) The calculation formula is:

[0329]

[0330] Similarly, calculate the continuous change coefficient of the number of views of the ni-th source information news article, recorded as up (ni) ; up (ni) The calculation formula is:

[0331]

[0332] Process C64: Obtain the number of priority manuscripts corresponding to the manuscript source information from the 1st to the nith, and obtain: nne1, nne2~nne ni ;nne1,nne2~nne ni With ne1, ne2~ne si The relationship satisfies:

[0333] Among them, nne1 represents the number of priority manuscripts corresponding to the first manuscript source information; nne2 represents the number of priority manuscripts corresponding to the second manuscript source information; and so on, nne ni Indicates the number of priority manuscripts corresponding to the ni-th manuscript source information;

[0334] Process C65: Calculate the news priority coefficient of the first source information, denoted as fne (1) ;fne (1) The calculation formula is: fne (1) =up (1) ×nne1;

[0335] Calculate the news priority coefficient of the second source information, denoted as fne (2) ;fne (2) The calculation formula is: fne (2) =up (2) ×nne2;

[0336] Similarly, calculate the news priority coefficient of the ni-th source information, denoted as fne (ni) ;fne (ni) The calculation formula is: fne (ni) =up (ni) ×nne ni ;

[0337] Process C66: Extraction of fne (1) ~fne(ni) The largest news priority coefficient is recorded as nem; the source information corresponding to nem is used as the high-quality source information of the news article;

[0338] Process C67: Repeat the same process of determining the high-quality source information of news articles (i.e., process C61 to process C66) to determine the high-quality source information corresponding to each article of interest;

[0339] Manuscript push submodule: used to obtain the number of all users and push priority manuscripts to all users; obtain the number of views of different types of manuscripts (i.e. non-news manuscripts) of each user (in the past three months), calculate the acceptance value of each user for different types of manuscripts (i.e. non-news manuscripts), and recommend manuscripts to be pushed to users based on their acceptance values;

[0340] Process D: The workflow of the manuscript push submodule is as follows:

[0341] Process D1: Get the number of all users, denoted as us; push priority manuscripts to all users;

[0342] Get the number of views of each user's scientific articles in the past three months, and get: usc(1,3), usc(1,2), usc(1,1)~usc(us,1);

[0343] Among them, usc(1,1) represents the number of views of the scientific articles of the first user in the third month; usc(1,2) represents the number of views of the scientific articles of the first user in the second month; usc(1,1) represents the number of views of the scientific articles of the first user in the first month;

[0344] By analogy, usc(us,1) represents the number of views of scientific articles by the us-th user in the first month;

[0345] Get the number of views of each user's life articles in the past three months, and get: uli(1,3), uli(1,2), uli(1,1)~uli(us,1);

[0346] Among them, uli(1,1) represents the number of views of the life manuscript of the first user in the third month; uli(1,2) represents the number of views of the life manuscript of the first user in the second month; uli(1,1) represents the number of views of the life manuscript of the first user in the first month;

[0347] By analogy, uli(us,1) represents the number of views of the life articles of the us-th user in the first month;

[0348] Get the number of views of shopping articles of each user in the past three months, and get: ush(1,3), ush(1,2), ush(1,1)~ush(us,1);

[0349] Among them, ush(1,3) represents the number of views of the shopping manuscript of the first user in the third month; ush(1,2) represents the number of views of the shopping manuscript of the first user in the second month; ush(1,1) represents the number of views of the shopping manuscript of the first user in the first month;

[0350] By analogy, ush(us,1) represents the number of views of the shopping articles of the us-th user in the first month;

[0351] Get the number of views of entertainment articles of each user in the past three months, and get: ure(1,3), ure(1,2), ure(1,1)~ure(us,1);

[0352] Among them, ure(1,3) represents the number of page views of the entertainment articles of the first user in the third month; ure(1,2) represents the number of page views of the entertainment articles of the first user in the second month; ure(1,1) represents the number of page views of the entertainment articles of the first user in the first month;

[0353] By analogy, ure(us,1) represents the number of entertainment article views by the us-th user in the first month;

[0354] Get the page views of other articles of each user in the past three months, and get: uot(1,3), uot(1,2), uot(1,1)~uot(us,1);

[0355] Among them, uot(1,3) represents the number of views of other articles of the first user in the third month; uot(1,2) represents the number of views of other articles of the first user in the second month; uot(1,1) represents the number of views of other articles of the first user in the first month;

[0356] By analogy, uot(us,1) represents the number of views of other articles by the us-th user in the first month;

[0357] It should be noted that the "user" in the manuscript push submodule refers to: a person or unit that accepts the manuscript push service of the present invention (a smart manuscript publishing system and method);

[0358] Process D2: Calculate the average number of scientific article views of all users in the past three months (i.e., the average of usc(1,3), usc(1,2), usc(1,1) to usc(us,1)), recorded as msc;

[0359] The average number of views of life articles in the past three months (i.e., the average of uli(1,3), uli(1,2), and uli(1,1) to uli(us,1)), denoted as mli;

[0360] The average number of shopping article views in the past three months (i.e., the average of ush(1,3), ush(1,2), ush(1,1) to ush(us,1)), denoted as msh;

[0361] The average number of entertainment article views in the past three months (i.e., the average of ure(1,3), ure(1,2), ure(1,1) to ure(us,1)), denoted as mre;

[0362] The average number of views of other manuscripts in the past three months (i.e., the average of uot(1,3), uot(1,2), uot(1,1) to uot(us,1)), denoted as mot;

[0363] Process D3: Take the number of views of the first user's science and technology articles, life articles, shopping articles, entertainment articles and other articles in the past three months as sample data, and construct a sample matrix, recorded as matrix o; the mathematical expression of matrix o is:

[0364]

[0365] Calculate the acceptance value of the first user for different types of manuscripts (i.e., non-news manuscripts) based on the matrix o, and push the manuscripts to be pushed to the first user;

[0366] Process D31: Define calculation formula d1-1: X (1,j) =1-[usc(1,j) / (msc / 3)];

[0367] Calculation formula d1-2: Y (1,j) =1-[uli(1,j) / (mli / 3)];

[0368] Calculation formula d1-3: Z (1,j) =1-[ush(1,j) / (msh / 3)];

[0369] Calculation formula d1-4: P (1,j) =1-[ure(1,j) / (mre / 3)];

[0370] Calculation formula d1-5: Q (1,j) =1-[uot(1,j) / (mot / 3)];

[0371] Among them, X (1,j) It represents the deviation of the reading volume of scientific articles of the first user in the jth month relative to the average reading volume of scientific articles of all users;

[0372] Y (1,j) It represents the deviation of the reading volume of the first user's life articles in the jth month relative to the average reading volume of all users' life articles;

[0373] Z (1,j) It represents the deviation of the number of shopping articles read by the first user in the jth month relative to the average number of shopping articles read by all users;

[0374] P (1,j) It represents the deviation of the entertainment article reading volume of the first user in the jth month relative to the average entertainment article reading volume of all users;

[0375] Q (1,j) It represents the deviation of the reading volume of other articles of the first user in the jth month relative to the average reading volume of other articles of all users;

[0376] The value range of j is: 1 to 3;

[0377] Process D32: According to the calculation formulas d1-1 to d1-5 in process D31, the offsets of the parameters in the matrix o are calculated to obtain the matrix o1; the mathematical expression of the matrix o1 is:

[0378]

[0379] Calculate X (1,1) , X (1,2) and X (1,3) The average value of (1,1) , Y (1,2) and Y (1,3) The average value is denoted as aY; Z (1,1) , Z (1,2) and Z (1,3) The average value is denoted as aZ; P (1,1) , P (1,2) and P (1,3) The average value is denoted as aP; Q (1,1) , Q (1,2) and Q (1,3) The average value of is denoted as aQ;

[0380] Process D33: Calculate X (1,1) , X (1,2) and X (1,3) The standard deviation of is denoted as SX; the calculation formula of SX is:

[0381]

[0382] Calculate Y (1,1) , Y (1,2) and Y (1,3) The standard deviation of is denoted as SY. The calculation formula of SY is:

[0383]

[0384] Calculate Z (1,1) , Z (1,2) and Z(1,3) The standard deviation of is denoted as SZ; the calculation formula of SZ is:

[0385]

[0386] Calculate P (1,1) , P (1,2) and P (1,3) The standard deviation of is denoted as SP; the calculation formula of SP is:

[0387]

[0388] Calculate Q (1,1) , Q (1,2) and Q (1,3) The standard deviation of is denoted as SQ; the calculation formula of SQ is:

[0389]

[0390] Process D34: Define calculation formula d2-1: dX (1,j) =(X (1,j) -aX) / SX;

[0391] Calculation formula d2-2: dY (1,j) =(Y (1,j) -aY) / SY;

[0392] Calculation formula d2-3: dZ (1,j) =(Z (1,j) -aZ) / SZ;

[0393] Calculation formula d2-4: dP (1,j) =(P (1,j) -aP) / SP;

[0394] Calculation formula d2-5: dQ (1,j) =(Q (1,j) -aQ) / SQ;

[0395] Among them, dX (1，j) Indicates X (1，j) The value of the second normalization, dY (1，j) Indicates Y (1，j) The value of the second normalization, dZ (1，j) Indicates Z (1，j) The value of the second normalization, dP (1，j) Indicates P (1，j) The value of the second normalization, dQ (1，j) Express Q (1，j) The value for secondary normalization; the value range of j is: 1 to 3;

[0396] Process D35: According to the calculation formulas d2-1 to d2-5 in process D34, the quadratic normalized values ​​of the parameters in the matrix o1 are calculated to obtain the matrix o2; the mathematical expression of the matrix o2 is:

[0397]

[0398] Calculate dX (1,1) 、dX (1,2) and dX (1,3) The average value, denoted as dX (1,-) ; Calculate dY (1,1) ,dY (1,2) and dY (1,3) The average value of (1,-) ; Calculate dZ (1,1) ,dZ (1,2) and dZ (1,3) The average value of (1,-) ; Calculate dP (1,1) , dP (1,2) and dP (1,3) The average value is denoted as dP (1,-) ; Calculate dQ (1,1) , dQ (1,2) and dQ (1,3) The average value of (1,-) ;

[0399] Calculate the average value of the parameters in the first row of matrix o2 (i.e. dX (1,1) ,dY (1,1) ,dZ (1,1) , dP (1,1) and dQ (1,1) The average value of (-，1) ;

[0400] Calculate the average value of the parameters in the second row of matrix o2 (i.e. dX (1,2) ,dY (1,2) ,dZ (1,2) , dP (1,2) and dQ (1,2) The average value of (-，2) ;

[0401] Calculate the average value of the parameters in the third row of matrix o2 (i.e. dX (1,3) ,dY (1,3) ,dZ (1,3) , dP (1,3) and dQ (1,3) The average value of (-，3) ;

[0402] Process D36: Define calculation formula d3-1:

[0403] Among them, rX (1，j) Denotes dX (1，j) The covariance of (-，j) Represents the average value of the jth row in matrix o2; the value range of j is: 1 to 3;

[0404] Calculation formula d3-2:

[0405] Among them, rY (1，j) indicates dY (1，j) The covariance of

[0406] Calculation formula d3-3:

[0407] Among them, rZ (1，j) Denotes dZ (1，j) The covariance of

[0408] Calculation formula d3-4:

[0409] Among them, rP (1，j) Indicates dP (1，j) The covariance of

[0410] Calculation formula d3-5:

[0411] Among them, rQ (1，j) Indicates dQ (1，j) The covariance of

[0412] Process D37: According to the calculation formulas d3-1 to d3-5 in process D36, the covariance of each parameter in the matrix o2 is calculated to obtain the matrix o3; the mathematical expression of the matrix o3 is:

[0413]

[0414] Split the matrix o3 to obtain matrices o4-1 to o4-5;

[0415] The mathematical expression of matrix o4-1 is:

[0416]

[0417] The mathematical expression of matrix o4-2 is:

[0418]

[0419] The mathematical expression of matrix o4-3 is:

[0420]

[0421] The mathematical expression of matrix o4-4 is:

[0422]

[0423] The mathematical expression of matrix o4-5 is:

[0424]

[0425] Process D38: Use Intel MKL library functions to define and calculate weight values ​​λ1, λ2, λ3, λ4, and λ54 and offsets μ1, μ2, μ3, μ4, and μ3, so that matrices o4-1 to o4-5 satisfy the following mathematical relationship:

[0426]

[0427] Process D39: Calculate the acceptance value of the first user's scientific manuscript and record it as bsc, bsc = λ1 / μ1;

[0428] Calculate the acceptance value of the first user's life manuscript and record it as bli, bli = λ2 / μ2;

[0429] Calculate the acceptance value of the first user's shopping manuscript and record it as bsh, bsh = λ3 / μ3;

[0430] Calculate the acceptance value of the first user's entertainment manuscript and record it as bre, bre = λ4 / μ4;

[0431] Calculate the acceptance value of other manuscripts of the first user and record it as bot, bot = λ5 / μ5;

[0432] Arrange bsc, bli, bsh, bre and bot in descending order to obtain the interest sequence of the first user, and push the manuscripts to be pushed to the first user in the order of the interest sequence;

[0433] Process D4: Repeat the same process of pushing the manuscript to be pushed for the first user (ie, process D3), and push the manuscript to be pushed for the second to us-th users.

[0434] Manuscript update module: used to obtain manuscripts with high-quality source information and manuscripts with non-high-quality source information in real time, update original manuscript data, and push manuscripts.

[0435] It should be noted that the present invention only lists four specific types of labels, namely "technology, life, shopping, and entertainment". Media workers or relevant technical personnel can define or design more labels on their own when using the present invention to improve the benefits of the present invention to different user groups.

[0436] Embodiment 2

[0437] See also Figure 2 , an intelligent publishing method includes:

[0438] Step S1: obtaining original manuscripts and submission information of original manuscripts, eliminating manuscripts and submission information with bad content, and obtaining manuscripts to be pushed and submission information to be submitted;

[0439] Step S2: construct a news recognition model, and use the news recognition model to classify the manuscripts to be pushed, extract the news broadcast manuscripts as the priority manuscripts;

[0440] Change the training object of the news recognition model and build an interest recognition model; perform secondary classification on the articles to be pushed according to the interest recognition model, add interest tags to non-news broadcast articles, and obtain articles of interest;

[0441] Using natural language processing technology, the content in the manuscripts of interest is identified, and pictures are added to the manuscripts of interest according to the content in the manuscripts of interest, so as to obtain the manuscripts to be pushed;

[0442] Step S3: Count the historical page views of different types of non-news articles (in the past three months), and calculate the additional supplementary amount of articles to be pushed in combination with the articles to be pushed;

[0443] Collect statistics on prospective contributors with continuous contribution records (in the past three months) as manuscript source information; obtain the number of views of news articles and articles of interest corresponding to the manuscript source information in the past three months to determine high-quality manuscript source information;

[0444] Obtain the number of all users and push priority articles to all users; obtain the number of views of different types of articles (i.e., non-news articles) of each user (in the past three months), calculate the acceptance value of each user for different types of articles (i.e., non-news articles), and recommend articles to be pushed to users based on their acceptance values;

[0445] Step S4: Obtain manuscripts with high-quality manuscript source information and manuscripts with non-high-quality manuscript source information in real time, update original manuscript data, and push manuscripts.

[0446] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technical personnel in this field according to actual conditions. For example, if there are weight coefficients and proportional coefficients, their set sizes are to quantify each parameter to obtain a specific value, which is convenient for subsequent comparison. Regarding the size of the weight coefficient and the proportional coefficient, it is sufficient as long as it does not affect the proportional relationship between the parameter and the quantized value.

[0447] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. An intelligent manuscript publishing system, characterized in that: The manuscript publishing system comprises: Manuscript acquisition module: used to obtain original manuscripts and submission information of original manuscripts, remove manuscripts and submission information with bad content, and obtain manuscripts and submission information that are ready to be pushed; Manuscript processing module: including manuscript classification submodule and manuscript illustration submodule; Manuscript classification submodule: used to build a news recognition model, and use the news recognition model to classify the manuscripts to be pushed, extract the news broadcast manuscripts as the priority manuscripts; Change the training object of the news recognition model and build an interest recognition model; perform secondary classification on the articles to be pushed according to the interest recognition model, add interest tags to non-news broadcast articles, and obtain articles of interest; Manuscript picture submodule: Use natural language processing technology to identify the content in the manuscript of interest, and add pictures to the manuscript of interest according to the content in the manuscript of interest, so as to obtain the manuscript to be pushed; Manuscript promotion module: including manuscript statistics submodule and manuscript push submodule; Manuscript statistics submodule: used to count the historical views of different types of manuscripts to be pushed, and calculate the additional supplementary amount of manuscripts to be pushed in combination with the manuscripts to be pushed; Collect the information of prospective contributors with continuous submission records as manuscript source information; obtain the number of views of news articles and articles of interest corresponding to the manuscript source information in the past three months to determine the high-quality manuscript source information; Manuscript push submodule: used to obtain the number of all users and push priority manuscripts to all users; obtain the number of views of different types of manuscripts of each user, calculate the acceptance value of each user for different types of manuscripts, and recommend manuscripts to be pushed to users based on the user's acceptance value; Manuscript update module: used to obtain manuscripts with high-quality source information and manuscripts with non-high-quality source information in real time, update original manuscript data, and push manuscripts.

2. The intelligent manuscript publishing system according to claim 1, characterized in that: The workflow of the manuscript classification submodule is as follows: Process B1: Count the number of quasi-submission information, recorded as si; count the number of quasi-pushed manuscripts corresponding to each quasi-submission information, and get do1, do2~do si ; Among them, do1 represents the number of manuscripts that are expected to be pushed corresponding to the first submission information; do2 represents the number of manuscripts that are expected to be pushed corresponding to the second submission information; and so on. si Indicates the number of manuscripts that are expected to be pushed corresponding to the si-th expected submission information; Process B2: Get news articles as training articles a1 and non-news articles as training articles a2, and use the logistic regression model to build a news recognition model; Process B3: Use the joblib library to load the quasi-pushing manuscripts corresponding to the 1st to sith quasi-submission information into the news recognition model, classify the quasi-pushing manuscripts once, and extract the news broadcast manuscripts corresponding to each quasi-submission information as the priority manuscripts; Count the number of priority manuscripts corresponding to each prospective submission information, and get: ne1, ne2~ne si ; Among them, ne1 represents the number of priority manuscripts corresponding to the first quasi-submission information; ne2 represents the number of priority manuscripts corresponding to the second quasi-submission information; and so on, ne si Indicates the number of priority manuscripts corresponding to the si-th prospective submission information; Process B4: Treat the manuscripts that are not priority manuscripts as manuscripts to be classified, and define interest tags for the manuscripts to be classified. The interest tags include: technology, life, shopping, entertainment and others; Change the training object of the news recognition model and build an interest recognition model; perform secondary classification on the manuscripts to be classified according to the interest recognition model, and count the number of scientific and technological manuscripts in the manuscripts to be classified corresponding to the 1st to sith quasi-submission information, and obtain: sc1, sc2~sc si ; The number of life manuscripts is: li1, li2~li si ; The number of shopping articles is: sh1, sh2~sh si ; The number of entertainment articles is: re1, re2~re si ; The number of other manuscripts is: ot1, ot2~ot si ; Among them, sc1, l i1, sh1, re1 and ot1 respectively represent the number of scientific manuscripts, life manuscripts, shopping manuscripts, entertainment manuscripts and other manuscripts corresponding to the first quasi-submission information; sc2, l i2, sh2, re2, and ot2 respectively represent the number of scientific manuscripts, life manuscripts, shopping manuscripts, entertainment manuscripts, and other manuscripts corresponding to the second quasi-submission information; By analogy, sc si ,li si ,sh si 、re si andot si , respectively represent the number of scientific manuscripts, life manuscripts, shopping manuscripts, entertainment manuscripts and other manuscripts corresponding to the si-th prospective submission information; Process B5: Send the scientific manuscripts, life manuscripts, shopping manuscripts, entertainment manuscripts and other manuscripts corresponding to the 1st to sith quasi-submission information as interest manuscripts to the manuscript illustration submodule for illustration, obtain the manuscripts to be pushed, and enter the manuscript statistics submodule.

3. The intelligent manuscript publishing system according to claim 2, characterized in that: The specific process of process B2 is as follows: Process B21: Import the pandas library and the scikit-learn library; Create a list yes_texts to store training manuscript a1; create a list no_texts to store training manuscript a2; Combine news and non-news article text as feature X, that is: X=news_texts+non_news_texts; Define the label y, mark news articles as 1 and non-news articles as 0, that is: y=[1]*len(yes_texts)+[0]*len(no_texts); Process B22: Use the load_data function to load the feature X and label y as the combined data set; divide the combined data set into a training set and a test set; The text of the training set is denoted as X_train, and the label is denoted as y_train; The text of the test set is denoted as X_test, and the label is denoted as y_test; The number of training sets and test sets satisfies: training set > test set; Process B23: Use the manuscripts in the training set as training texts and the manuscripts in the test set as test texts; initialize the TF-IDF vectorizer and perform TF-IDF feature extraction on the training set X_train; Use the fit_transform method to learn the vocabulary in the training text, calculate the TF-IDF value of each word in each training text, and get the sparse matrix X_train_tfidf; In which, each row in the sparse matrix X_train_tfidf represents a training text, each column represents a word, and the values ​​in the matrix correspond to the TF-IDF values ​​of the words in different training texts; Use the transform method to extract TF-IDF features from the test set text to obtain the training feature X_test_tfidf; Process B24: Initialize the logistic regression model model and train the model with the training set data, that is: model.fit(X_train_tfidf,y_train); Use the trained model to predict the test set, that is: y_pred=model.predict(X_test_tfidf); Use the accuracy_score() method to calculate the accuracy of the model on the test set; and use the classification_report() method to calculate the precision until the accuracy and precision between any article with label y_train=1 in the training set and any article with label y_test=1 in the test set are close to 100%; the accuracy and precision between the three types of articles, namely, the article with label y_train=1 and the article with label y_test=0, the article with label y_train=0 and the article with label y_test=1, and the article with label y_train=0 and the article with label y_test=0, are all 0%, completing the training of the news recognition model; Load the joblib library to save the news recognition model and load manuscripts for the news recognition model.

4. The intelligent manuscript publishing system according to claim 2, characterized in that: The workflow of the manuscript statistics submodule is as follows: Process C1: Count the views of scientific manuscripts in the past three months and get: hsc3, hsc2 and hsc1; The number of views of life articles is: hli3, hli2 and hli1; The number of views of shopping articles is: hsh3, hsh2 and hsh1; The number of views of entertainment articles is: hre3, hre2 and hre1; The views of other manuscripts are: hot3, hot2 and hot1; Among them, hsc3, hl i3, hsh3, hre3 and hot3 represent the views of science and technology articles, life articles, shopping articles, entertainment articles and other articles in the third month; hsc2, hli2, hsh2, hre2 and hot2 represent the views of science and technology articles, life articles, shopping articles, entertainment articles and other articles in the second month; hsc1, hli1, hsh1, hre1 and hot1 represent the number of views of science and technology articles, life articles, shopping articles, entertainment articles and other articles in the first month; Process C2: Calculate the growth rate of scientific manuscripts from the third month to the second month, denoted as usc (3-2) ;usc (3-2) The calculation formula is: usc (3-2) =(hsc2-hsc3) / hsc3; The growth rate of life articles from the third month to the second month, denoted as uli (3-2) uli (3-2) The calculation formula is: (3-2) =(hli2-hli3) / hli3; The growth rate of shopping articles from the third month to the second month, denoted as ush (3-2) ;ush (3-2) The calculation formula is: (3-2) =(hsh2-hsh3) / hsh3; The growth rate of entertainment articles from the third month to the second month, denoted as ure (3-2) ure (3-2) The calculation formula is:ure (3-2) =(hre2-hre3) / hre3; The growth rate of other manuscripts from the third month to the second month is denoted as uot (3-2) ;uot (3-2) The calculation formula is: uot (3-2) =(hot2-hot3) / hot3.

5. The intelligent manuscript publishing system according to claim 4, characterized in that: The subsequent process of process C2 is as follows: Process C3: Calculate the growth rate of scientific manuscripts from the second month to the first month, denoted as usc (2-1) ;usc (2-1) The calculation formula is: usc (2-1) =(hsc1-hsc2) / hsc2; The growth rate of life articles from the second month to the first month, denoted as uli (2-1) uli (2-1) The calculation formula is: (2-1) =(hli1-hli2) / hli2; The growth rate of shopping articles from the second month to the first month, denoted as ush (2-1) ;ush (2-1) The calculation formula is: (2-1) =(hsh1-hsh2) / hsh2; The growth rate of entertainment articles from the second month to the first month, denoted as ure (2-1) ure (2-1) The calculation formula is:ure (2-1) =(hre1-hre2) / hre2; The growth rate of other manuscripts from the second month to the first month, denoted as uot (2-1) ;uot (2-1) The calculation formula is: uot (2-1) =(hot1-hot2) / hot2; Process C4: Calculation of sc1~sc si The sum of is recorded as asc; li1~li si The sum of is written as ali; sh1~sh si The sum of re1 and re2 is denoted as ash. si The sum of is written as are; ot1~ot si The sum of is written as aot; With {usc (3-2) , usc (2-1) , hsc2, hsc1} is group a1, with {uli (3-2) ,uli (2-1) , hli2, hli1} is group a2, with {ush (3-2) ,ush (2-1) , hsh2, hsh1} is group a3, with {ure (3-2) ,ure (2-1) , hre2, hre1} is group a4, with {uot (3-2) ,uot (2-1) , hot2, hot1} is group a5; Using the state transition matrix, the expected number of scientific manuscript submissions is calculated based on group a1, denoted as Esc; The expected number of scientific manuscripts submitted based on group a2 is calculated as El i; the expected number of scientific manuscripts submitted based on group a3 is calculated as Esh; the expected number of scientific manuscripts submitted based on group a4 is calculated as Ere; the expected number of scientific manuscripts submitted based on group a5 is calculated as Eot; Process C5: Compare asc and Esc. If asc≤Esc, then Δsc=(Esc-asc); if asc>Esc, then Δsc=0; where Δsc represents the number of additional supplementary manuscripts of scientific and technological manuscripts; Compare ali and Eli. If ali≤Eli, then Δli=(Eli-ali); if ali>Eli, then Δli=0; where Δli represents the number of additional supplementary manuscripts for life manuscripts; Compare ash and Esh. If ash≤Esh, then Δsh=(Esh-ash); if ash>Esh, then Δsh=0; where Δsh represents the number of additional supplementary manuscripts for shopping manuscripts; Compare are and Ere. If are ≤ Ere, then Δre = (Ere - are); if are > Ere, then Δre = 0; where Δre represents the number of additional supplementary articles for entertainment articles. Compare aot and Eot. If aot≤Eot, then Δot=(Eot-aot); if aot>Eot, then Δot=0; where Δot represents the number of additional supplementary manuscripts for other manuscripts; According to Δsc, Δl i, Δsh, Δre and Δot, supplement the corresponding types of manuscripts; Process C6: Collect the information of prospective manuscripts with continuous submission records in the past three months as manuscript source information; The number of manuscript source information is counted, denoted as ni; the relationship between ni and si satisfies: ni≤si; si represents the number of quasi-submission information; Obtain the number of news article views and interest article views corresponding to the manuscript source information in the past three months, determine the high-quality manuscript source information, and enter the manuscript push sub-module.

6. The intelligent manuscript publishing system according to claim 5, characterized in that: The specific process of process C6 is as follows: Process C61: Obtain the news article page views corresponding to the source information in the past three months and determine the high-quality source information of the news articles; Get the news article views corresponding to the source information in the past three months, and get ne(1,3), ne(1,2), ne(1,1)~ne(ni,1); Among them, ne(1,3) represents the number of news article views in the third month corresponding to the first source information; ne(1,2) represents the number of news article views in the second month corresponding to the first source information; ne(1,1) represents the number of news article views in the first month corresponding to the first source information; By analogy, ne(ni,1) represents the number of news article views in the first month corresponding to the ni-th source information; Process C62: Calculate the sum of the number of news article views in the third month corresponding to the first to the nith source information, record it as ane(-, 3); the sum of the number of news article views in the second month is record it as ane(-, 2); the sum of the number of news article views in the first month is record it as ane(-, 1); Process C63: Calculate the continuous change coefficient of the number of views of the first source information news article, recorded as up (1) ; up (1) The calculation formula is: Calculate the continuous change coefficient of the second source news article pageviews, recorded as up (2) ; up (2) The calculation formula is: Similarly, calculate the continuous change coefficient of the number of views of the ni-th source information news article, recorded as up (ni) ; up (ni) The calculation formula is: Process C64: Obtain the number of priority manuscripts corresponding to the manuscript source information from the 1st to the nith, and obtain: nne1, nne2~nne ni ;nne1,nne2~nne ni With ne1, ne2~ne si The relationship satisfies: Among them, nne1 represents the number of priority manuscripts corresponding to the first manuscript source information; nne2 represents the number of priority manuscripts corresponding to the second manuscript source information; and so on, nne ni Indicates the number of priority manuscripts corresponding to the ni-th manuscript source information; Process C65: Calculate the news priority coefficient of the first source information, denoted as fne (1) ;fne (1) The calculation formula is: fne (1) =up (1) ×nne1; Calculate the news priority coefficient of the second source information, denoted as fne (2) ;fne (2) The calculation formula is: fne (2) =up (2) ×nne2; Similarly, calculate the news priority coefficient of the ni-th source information, denoted as fne (ni) ;fne (ni) The calculation formula is: fne (ni) =up (ni) ×nne ni ; Process C66: Extraction of fne (1) ~fne (ni) The largest news priority coefficient is recorded as nem; the source information corresponding to nem is used as the high-quality source information of the news article; Process C67: Repeat the same process of determining high-quality source information of news articles to determine the high-quality source information corresponding to each article of interest.

7. The intelligent manuscript publishing system according to claim 5, characterized in that: The workflow of the manuscript push submodule is as follows: Process D1: Get the number of all users, denoted as us; push priority manuscripts to all users; Get the number of views of each user's scientific articles in the past three months, and get: usc(1,3), usc(1,2), usc(1,1)~usc(us,1); Among them, usc(1,1) represents the number of views of the scientific articles of the first user in the third month; usc(1,2) represents the number of views of the scientific articles of the first user in the second month; usc(1,1) represents the number of views of the scientific articles of the first user in the first month; By analogy, usc(us,1) represents the number of views of scientific articles by the us-th user in the first month; Get the number of views of each user's life articles in the past three months, and get: uli(1,3), uli(1,2), uli(1,1)~uli(us,1); Among them, uli(1,1) represents the number of views of the life manuscript of the first user in the third month; uli(1,2) represents the number of views of the life manuscript of the first user in the second month; uli(1,1) represents the number of views of the life manuscript of the first user in the first month; By analogy, uli(us,1) represents the number of views of the life articles of the us-th user in the first month; Get the number of views of shopping articles of each user in the past three months, and get: ush(1,3), ush(1,2), ush(1,1)~ush(us,1); Among them, ush(1,3) represents the number of views of the shopping manuscript of the first user in the third month; ush(1,2) represents the number of views of the shopping manuscript of the first user in the second month; ush(1,1) represents the number of views of the shopping manuscript of the first user in the first month; By analogy, ush(us,1) represents the number of views of the shopping articles of the us-th user in the first month; Get the number of views of entertainment articles of each user in the past three months, and get: ure(1,3), ure(1,2), ure(1,1)~ure(us,1); Among them, ure(1,3) represents the number of page views of the entertainment articles of the first user in the third month; ure(1,2) represents the number of page views of the entertainment articles of the first user in the second month; ure(1,1) represents the number of page views of the entertainment articles of the first user in the first month; By analogy, ure(us,1) represents the number of entertainment article views by the us-th user in the first month; Get the page views of other articles of each user in the past three months, and get: uot(1,3), uot(1,2), uot(1,1)~uot(us,1); Among them, uot(1,3) represents the number of views of other articles of the first user in the third month; uot(1,2) represents the number of views of other articles of the first user in the second month; uot(1,1) represents the number of views of other articles of the first user in the first month; Similarly, uot(us,1) represents the number of views of other articles by the us-th user in the first month; Process D2: Calculate the average number of scientific article views of all users in the past three months, recorded as msc; The average number of views of life articles in the past three months is recorded as mli; The average number of shopping article views in the past three months, denoted as msh; The average number of entertainment article views in the past three months, recorded as mre; The average number of views of other manuscripts in the past three months is recorded as mot; Process D3: Take the number of views of the first user's science and technology articles, life articles, shopping articles, entertainment articles and other articles in the past three months as sample data, and construct a sample matrix, recorded as matrix o; the mathematical expression of matrix o is: Calculate the acceptance value of the first user for different types of manuscripts according to the matrix o, and push the manuscripts to be pushed to the first user; Process D4: Repeat the same process of pushing the manuscript to be pushed for the first user, and push the manuscript to be pushed for the second to us users.

8. The intelligent manuscript publishing system according to claim 7, characterized in that: The specific process of process D3 is as follows: Process D31: Define calculation formula d1-1: X (1,j) =1-[usc(1,j) / (msc / 3)]; Calculation formula d1-2: Y (1,j) =1-[uli(1,j) / (mli / 3)]; Calculation formula d1-3: Z (1,j) =1-[ush(1,j) / (msh / 3)]; Calculation formula d1-4: P (1,j) =1-[ure(1,j) / (mre / 3)]; Calculation formula d1-5: Q (1,j) =1-[uot(1,j) / (mot / 3)]; Among them, X (1,j) It represents the deviation of the reading volume of scientific articles of the first user in the jth month relative to the average reading volume of scientific articles of all users; Y (1,j) It represents the deviation of the reading volume of the first user's life articles in the jth month relative to the average reading volume of all users' life articles; Z (1,j) It represents the deviation of the number of shopping articles read by the first user in the jth month relative to the average number of shopping articles read by all users; P (1,j) It represents the deviation of the entertainment article reading volume of the first user in the jth month relative to the average entertainment article reading volume of all users; Q (1,j) It represents the deviation of the reading volume of other articles of the first user in the jth month relative to the average reading volume of other articles of all users; The value range of j is: 1 to 3; Process D32: According to the calculation formulas d1-1 to d1-5 in process D31, the offsets of the parameters in the matrix o are calculated to obtain the matrix o1; the mathematical expression of the matrix o1 is: Calculate X (1,1) , X (1,2) and X (1,3) The average value of (1,1) , Y (1,2) and Y (1,3) The average value is denoted as aY; Z (1,1) , Z (1,2) and Z (1,3) The average value is denoted as aZ; P (1,1) , P (1,2) and P (1,3) The average value is denoted as aP; Q (1,1) , Q (1,2) and Q (1,3) The average value of is denoted as aQ; Process D33: Calculate X (1,1) , X (1,2) and X (1,3) The standard deviation of is denoted as SX; the calculation formula of SX is: Calculate Y (1,1) , Y (1,2) and Y (1,3) The standard deviation of is denoted as SY. The calculation formula of SY is: Calculate Z (1,1) , Z (1,2) and Z (1,3) The standard deviation of is denoted as SZ; the calculation formula of SZ is: Calculate P (1,1) , P (1,2) and P (1,3) The standard deviation of is denoted as SP; the calculation formula of SP is: Calculate Q (1,1) , Q (1,2) and Q (1,3) The standard deviation of is denoted as SQ; the calculation formula of SQ is: Process D34: Define calculation formula d2-1: dX (1,j) =(X (1,j) -aX) / SX; Calculation formula d2-2: dY (1,j) =(Y (1,j) -aY) / SY; Calculation formula d2-3: dZ (1,j) =(Z (1,j) -aZ) / SZ; Calculation formula d2-4: dP (1,j) =(P (1,j) -aP) / SP; Calculation formula d2-5: dQ (1,j) =(Q (1,j) -aQ) / SQ; Among them, dX (1，j) Indicates X (1，j) The value of the second normalization, dY (1，j) Indicates Y (1，j) The value of the second normalization, dZ (1，j) Indicates Z (1，j) The value of the second normalization, dP (1，j) Indicates P (1，j) The value of the second normalization, dQ (1，j) Express Q (1，j) The value for secondary normalization; the value range of j is: 1 to 3.

9. The intelligent manuscript publishing system according to claim 8, characterized in that: The subsequent process of the process D34 is as follows: Process D35: According to the calculation formulas d2-1 to d2-5 in process D34, the quadratic normalized values ​​of the parameters in the matrix o1 are calculated to obtain the matrix o2; the mathematical expression of the matrix o2 is: Calculate dX (1,1) 、dX (1,2) and dX (1,3) The average value, denoted as dX (1,-) ; Calculate dY (1,1) ,dY (1,2) and dY (1,3) The average value of (1,-) ; Calculate dZ (1,1) ,dZ (1,2) and dZ (1,3) The average value of (1,-) ; Calculate dP (1,1) , dP (1,2) and dP (1,3) The average value is denoted as dP (1,-) ; Calculate dQ (1,1) , dQ (1,2) and dQ (1,3) The average value of (1,-) ; Calculate the average value of the parameters in the first row of matrix o2, denoted as dD (-，1) ; Calculate the average value of the parameters in the second row of matrix o2, denoted as dD (-，2) ; Calculate the average value of the parameters in the third row of matrix o2, denoted as dD (-，3) ; Process D36: Define calculation formula d3-1: Among them, rX (1，j) Denotes dX (1，j) The covariance of (-，j) Represents the average value of the jth row in matrix o2; the value range of j is: 1 to 3; Calculation formula d3-2: Among them, rY (1，j) indicates dY (1，j) The covariance of Calculation formula d3-3: Among them, rZ (1，j) Denotes dZ (1，j) The covariance of Calculation formula d3-4: Among them, rP (1，j) Indicates dP (1，j) The covariance of Calculation formula d3-5: Among them, rQ (1，j) Indicates dQ (1，j) The covariance of Process D37: According to the calculation formulas d3-1 to d3-5 in process D36, the covariance of each parameter in the matrix o2 is calculated to obtain the matrix o3; the mathematical expression of the matrix o3 is: Split the matrix o3 to obtain matrices o4-1 to o4-5; The mathematical expression of matrix o4-1 is: The mathematical expression of matrix o4-2 is: The mathematical expression of matrix o4-3 is: The mathematical expression of matrix o4-4 is: The mathematical expression of matrix o4-5 is: Process D38: Use Intel MKL library functions to define and calculate weight values ​​λ1, λ2, λ3, λ4, and λ54 and offsets μ1, μ2, μ3, μ4, and μ3, so that matrices o4-1 to o4-5 satisfy the following mathematical relationship: Process D39: Calculate the acceptance value of the first user's scientific manuscript and record it as bsc, bsc = λ1 / μ1; Calculate the acceptance value of the first user's life manuscript and record it as bli, bli = λ2 / μ2; Calculate the acceptance value of the first user's shopping manuscript and record it as bsh, bsh = λ3 / μ3; Calculate the acceptance value of the first user's entertainment manuscript and record it as bre, bre = λ4 / μ4; Calculate the acceptance value of other manuscripts of the first user and record it as bot, bot = λ5 / μ5; Arrange bsc, bl i, bsh, bre and bot in descending order to obtain the interest sequence of the first user, and push the articles to be pushed to the first user in the order of the interest sequence.

10. An intelligent manuscript publishing method, applicable to an intelligent manuscript publishing system according to any one of claims 1 to 9, characterized in that: The publication method includes: Step S1: obtaining original manuscripts and submission information of original manuscripts, eliminating manuscripts and submission information with bad content, and obtaining manuscripts to be pushed and submission information to be submitted; Step S2: construct a news recognition model, and use the news recognition model to classify the manuscripts to be pushed, extract the news broadcast manuscripts as the priority manuscripts; Change the training object of the news recognition model and build an interest recognition model; perform secondary classification on the articles to be pushed according to the interest recognition model, add interest tags to non-news broadcast articles, and obtain articles of interest; Using natural language processing technology, the content in the manuscripts of interest is identified, and pictures are added to the manuscripts of interest according to the content in the manuscripts of interest, so as to obtain the manuscripts to be pushed; Step S3: Count the historical views of different types of manuscripts to be pushed, and calculate the additional supplementary amount of manuscripts to be pushed in combination with the manuscripts to be pushed; Collect the information of prospective contributors with continuous submission records as manuscript source information; obtain the number of views of news articles and articles of interest corresponding to the manuscript source information in the past three months to determine the high-quality manuscript source information; Obtain the number of all users and push priority manuscripts to all users; obtain the number of views of different types of manuscripts for each user, calculate the acceptance value of each user for different types of manuscripts, and recommend manuscripts to be pushed to users based on the user's acceptance value; Step S4: Obtain manuscripts with high-quality manuscript source information and manuscripts with non-high-quality manuscript source information in real time, update original manuscript data, and push manuscripts.

Images