CN103927398A - Microblog hype group discovering method based on maximum frequent item set mining - Google Patents

Abstract

The invention relates to a method for discovering microblog hype groups based on the mining of the largest frequent itemset, which effectively solves the problem of discovering microblog hype groups and preventing false and malicious hype. The microblog public open platform obtains a set of accounts participating in hype microblog communication; takes a single microblog as a transaction and accounts participating in microblog communication as items, and builds a hype microblog transaction database; treats and detects transactions in the microblog group corresponding to the transaction database. For each transaction, find the largest frequent item set contained in all transactions, calculate the overlap rate between each largest frequent item set, merge small item sets into large item sets, reduce the number of intersections, and when taking intersections between transactions, The binary search method is used to judge whether a certain item is included in the transaction, to improve the efficiency of mining the largest frequent item set, and to discover groups of microblog hype.

Description

Microblog Hype Group Discovery Method Based on Maximum Frequent Itemset Mining

technical field

The invention relates to the field of microblog public opinion monitoring, in particular to a method for discovering microblog hype groups based on maximum frequent item set mining.

Background technique

As a new form of social media, Weibo has the functions of blogging, media and instant messaging. The immediacy, grassroots, mobility, and interactivity of Weibo itself make it a natural carrier for the spread of Internet public opinion. In online public opinion, Weibo not only becomes the center and channel of public opinion dissemination, but also participates in the formation, development and guidance of public opinion.

Weibo dissemination is a double-edged sword: on the one hand, Weibo provides a platform for rapid response to information disclosure in some social events, which makes up for the shortcomings of traditional media and other network tools to a certain extent; , Weibo is different from traditional news media in that its news releases are repetitive and its authenticity cannot be guaranteed. It may be used as a carrier of rumors, a fuse of dissatisfaction, and even cause serious damage to national security and social stability. s consequence. Online disinformation begins with its creators and spreads to its disseminators.

Hewlett-Packard's social computing research team said in its latest report that Sina Weibo has an extremely serious problem of topic hype, and half of the Weibo reposted around hot topics are sent by hype users. The study found that the number of artificially manipulated false retweets in the spread of hot topics is extremely large, with 1% of spammers creating 49% of retweets. Since August 2013, government departments have intensified their efforts to guide online public opinion. According to the results of investigations on the Internet promoters companies such as "Qin Huohuo" and "Li 2 Demolition 4", there are a large number of malicious people on the Internet. The push-hand team of the organization, together with a small number of "opinion leaders" to organize the Internet "Water Army", has been fabricating fake news on the Internet for a long time, deliberately distorting facts, creating troubles, confusing right and wrong, and seriously disrupting the order of Internet public opinion. Their behavior has been controlled by national public opinion Highly concerned, relevant persons were also detained in accordance with the law on suspicion of crimes.

Therefore, facing emerging media and aiming at all kinds of hidden public opinion incitement behaviors, carry out the identification of hyped microblogs, analyze the characteristics of their dissemination groups, collect identification evidence of false push behaviors, and identify hype hotspots artificially created by others. It has important theoretical value and practical significance to improve the government's ability to supervise public opinion and maintain social harmony and stability.

With the explosive development of microblog, research on microblog accounts has attracted widespread interest from scholars at home and abroad, and some research results have been published in important conferences such as WWW and KDD in recent years. The current research on microblog accounts can be roughly divided into the following three categories: 1) feature analysis, including account attribute characteristics and behavior characteristics, etc.; 2) influence analysis, including influence evaluation system construction and measurement methods, etc.; 3) account interaction Relationship network analysis, including the basic attributes, generation and evolution of account relationship networks.

However, there are relatively few domestic and foreign literatures on the study of hype groups. The main relevant literatures include the identification of spammer accounts, sockpuppet accounts, and zombie accounts. Junk accounts refer to accounts that often post spam information. Z.Yi et al. analyzed the characteristics of junk accounts from multiple perspectives, and used machine learning to automatically identify junk accounts. Chao Yang et al. deeply analyzed the social relationship among spam accounts, and proposed a method to discover spam accounts according to the intimacy between accounts. Vest accounts refer to fake accounts that register multiple accounts for posting, forwarding, and commenting. Xueling Zheng et al. proposed a method to identify vest accounts by using text content and similarity matching. Zombie accounts refer to accounts registered maliciously for the purpose of buying and selling fans. Fang Ming et al. proposed an intelligent classification method based on the feature extraction of Weibo registered account names, which has a high accuracy rate. However, these methods do not solve how to discover Weibo hype groups and prevent false hype. The biggest difference between hype accounts and the above-mentioned types of accounts is that hype accounts focus on their "hype" behavior, and accounts participating in hype are scattered and directly related. Not obvious, more concealed and organized, and more difficult to find.

Group hype is similar to ordinary Weibo. The behaviors of posting, reposting, and commenting of the spreading crowd are isolated on the surface, but unconventional malicious spreading is often not the behavior of a single person, but an organized group behavior. However, this group behavior Is hidden, difficult to detect. Therefore, how to discover microblog hype groups and prevent false and malicious hype from causing adverse effects on society and unnecessary economic losses is a technical problem that must be seriously solved.

Contents of the invention

In view of the above situation, in order to overcome the defects of the prior art, the purpose of the present invention is to provide a method for discovering microblog hype groups based on maximum frequent itemset mining, which can effectively solve the discovery of microblog hype groups and prevent false and malicious hype .

The technical solution solved by the present invention is that the microblog speculation account discovery method based on the largest frequent itemset mining includes the following steps:

(1) Sample collection of hype microblogs: take the relevance of hype microblogs as clues, and obtain a collection of accounts participating in the spread of hype microblogs based on crawler technology or the public open platform of microblogs;

(2) Construction of transactional database: taking a single microblog as a transaction and accounts participating in the spread of microblogs as items, construct a hype microblog transactional database;

(3) Maximum frequent itemset mining: For each transaction in the transaction database corresponding to the microblog group to be detected, use the iterative intersection method to find out the maximum frequent itemsets contained in all transactions, and obtain several maximum frequent itemsets;

Due to the hype that each transaction in the microblog transaction database contains tens of thousands of items, directly mining the largest frequent item set in the original transaction database will affect the efficiency of algorithm execution, and use the binary search method to quickly eliminate non-frequent items in transactions project, find out the candidate set of the largest frequent item set, and reduce the size of the transaction database;

(4) Maximum frequent itemset merging: For each maximum frequent itemset, calculate the overlap rate between itemsets, merge the largest frequent itemsets, try to merge smaller itemsets into larger itemsets, and ensure After merging, the accounts in the item set still have certain relevance; by reducing the size of the transaction database, reducing the number of intersections, and when taking the intersection between transactions, use the binary search method to determine whether a certain item is included in the transaction, so as to improve the efficiency of mining the largest frequent itemset , so as to discover Weibo hype groups.

The method of the invention is simple and easy to operate, can accurately discover malicious microblog hype groups, prevent adverse effects on society and unnecessary economic losses, and has practical application value.

Description of drawings

Fig. 1 is a flow chart diagram of the present invention.

Fig. 2 is a schematic diagram of the hype microblog transaction database of the present invention.

Fig. 3 is a screenshot of the hype microblog transaction database of the present invention.

Fig. 4 is a comparison diagram of execution time of the algorithm of the present invention on the Mushroom dataset.

Fig. 5 is a comparison diagram of the execution time of the algorithm of the present invention on the hype microblog data set.

Fig. 6 is a graph showing the change of the number of itemsets in the MFS of the present invention.

Fig. 7 is a diagram of the maximum length variation of itemsets in the MFS of the present invention.

Detailed ways

The specific implementation manners of the present invention will be described in detail below in conjunction with the accompanying drawings.

As shown in Fig. 1, the present invention includes hype microblog transaction library, maximum frequent item set mining and maximum frequent item set merging part, hype microblog transaction library construction module is mainly responsible for collecting data and preprocessing, constructing transaction database D; The frequent itemset mining module first screens the candidate maximum frequent itemsets based on the binary search method, and then mines the maximum frequent itemsets MFS from the transaction database D based on the iterative intersection method; the maximum frequent itemsets merge module mainly merges the MFSs to maximize It is possible to restore the real hype group. The specific steps are:

1) Collect hype Weibo samples

The collection of hype microblog samples is the first step to realize the present invention. The selection of microblog samples should be relevant. We should learn from the existing mature discrimination methods or expert systems. There are two ways to collect hype microblog samples: one method is to choose crawler technology to download web pages from microblog web pages, analyze the page structure and extract the information of microblog dissemination accounts; One method is to call the Weibo public open platform and call the API function provided by the Weibo official to obtain the information of the Weibo dissemination account. In order to facilitate the discovery of hype groups, the following principles should also be followed when selecting hype Weibo samples:

a. Select popular microblogs with a relatively high number of retweets;

b. Weibo release time span < 180 days;

According to the algorithmic analysis conditions of the hype account to be mined, the content of the sample collection should include the microblog identification number, the microblog account identification number, and the basic information of the microblog account;

2) Build a transactional database

Transform the discovery of hype groups into the largest frequent itemset mining in data mining. Based on the collection of hype microblog samples, the corresponding transactions of hype microblogs and the items in the corresponding transactions of accounts participating in microblog forwarding are constructed to build a transaction database. as shown in picture 2;

3) Screening of candidate maximum frequent itemsets based on binary search

Due to the hype that each transaction in the microblog transaction database contains tens of thousands of items, directly mining the largest frequent item set in the original transaction database will affect the efficiency of algorithm execution. Based on the binary search method, it can quickly eliminate the items in the transaction. For non-frequent items, find out the candidate set of the largest frequent itemset, reduce the size of the transaction database, given the transaction database D, the minimum support number S, and filter the candidate maximum frequent itemset, the method is:

(1) Sort the transactions in the transaction library D from the largest to the smallest according to the number of items

(2) Record frequent item collection , the set of infrequent items ;Start from i=1, traverse each transaction T _i in D in order (1≤i≤|D|), for each item u in transaction T _i :

a) If u∈FI, keep u;

b) If u∈NFI, remove u from T _i ;

c) if , go to the next step to judge whether u is a frequent item;

(3) Start traversing the remaining transactions from j=i+1, and use the binary search method to judge whether u is contained in T _j , i<j≤|D|, and the termination condition is:

a) When the number of transactions containing u reaches S, it means that u is a frequent item, and u is added to FI;

b) When the sum of the number of remaining transactions and the number of transactions including u is less than S, it means that u is an infrequent item, and u is removed from T _i . If the number of transactions containing u is greater than 1 at this time, it means that u also appears in transactions other than T _i , then add u to NFI;

(4) After eliminating the non-frequent items in all affairs in D, the reduced transaction library _D1 can be obtained;

4) Mining of maximum frequent itemsets based on iterative intersection:

Mining the maximum frequent itemset by iteratively taking the intersection of transactions, given the reduced transaction database D ₁ , the minimum support number S, the method of mining the maximum frequent itemset is as follows:

(1) Sort the transactions in the transaction database D ₁ according to the number of items from large to small, so as to find the largest frequent item set as soon as possible, in order to reduce the size of the transaction database, merge the repeated transactions in the transaction database, and count the number of transactions ;

(2) In order to reduce the number of intersections, for a transaction T _i , 1≤i≤|D ₁ |-S+1, starting from i=1, first find the transaction set that contains any item in T _i , T _j |T _j contains at least one item in Ti; j>i), T _i takes the intersection with T _j in turn, moves the intersection of the two into the new transaction database D ₂ , and removes T _j at the same time, ;

(3) For the transaction T in the new transaction database D ₂ , if T is obtained by taking the intersection of not less than S transactions, then move T into the maximum frequent candidate item set set MFCS, and remove the child of T in D ₂ affairs;

(4) If the number of remaining transactions in the new transaction library D ₂ is less than S, then end the processing of the transaction library D ₂ and return to the upper-level transaction library; otherwise, start the process from the first step to D ₂ ;

(5) When the number of remaining transactions in the transaction database D ₁ is less than S, that is, i>|D ₁ |-S+1, end the processing of the current transaction database D ₁ ;

(6) Merge the itemsets in the MFCS and eliminate the non-maximal frequent itemsets at the same time, and the final result is the maximum frequent itemsets set MFS sought;

5) Maximum frequent itemsets merge:

Due to the limitation of the minimum support number, the maximum frequent itemsets in MFS are small in size, and there are a large number of overlapping items between some itemsets. The account groups represented by these itemsets are likely to belong to the same hype group. In order to solve this The problem is to use the overlap rate to reflect the similarity between two item sets, set the item set X ₁ , X ₂ ∈ MFS, and record the overlap rate of X ₁ and X ₂ as:

$\mathrm{<span\; class="notranslate">\; ORate</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&cap;</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span>}{\mathrm{<span\; class="notranslate">\; Min</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; ,</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; )</span>}$

In the above formula, |X ₁ ∩X ₂ | represents the number of overlapping items between X ₁ and X ₂ , Min(|X ₁ |,|X ₂ |) represents the number of items in a smaller itemset, and itemsets are merged The method is:

(1) Sort the maximum frequent itemsets in the MFS from the largest to the smallest according to the number of items;

(2) Traversing each maximum frequent itemset in MFS, starting from i=1, for , if ORate(X _i ,X _j )≥minOR, i<j≤|MFS|, then add the union of Xi and X _j to the new set MMFS _, and remove X _j at the same time;

(3) Repeat the above two steps for the itemsets in MMFS;

(4) When the overlapping rate of any two itemsets in MMFS is less than minOR, end.

The method of the present invention is simple, easy to operate, and through actual trial, it is shown that the method is stable and reliable, and has practical application value. The relevant information is as follows:

1) Dataset

Taking Sina Weibo as the research platform, and taking 81 microblogs suspected of hype as the research object, the number of accounts that actually participated in their forwarding was 380,726 (excluding accounts that participated in multiple forwardings), and the average number of items per transaction was 6,286. Most of these microblogs belong to the category of advertising and marketing, and there may be multiple hype groups participating in the dissemination process. Use the crawler program to crawl all the account IDs (UID) participating in these Weibo forwarding, and store them in the transaction database. The format of some data is shown in Figure 3.

In order to verify the efficiency of the algorithm described in the present invention (hereinafter referred to as IIA) applied to the mining of maximum frequent itemsets, a performance test is performed on a classic Mushroom data set and compared with known methods. The dataset contains 8,124 records, each record has 23 items, and records 23 attributes of mushrooms.

2) Performance evaluation

At first the performance of the method of the present invention is evaluated, the experimental environment is 4G internal memory, 2.0GHz dual-core Duo T5800CPU, Windows732 bit operating system, realize this algorithm with Java, and compare with classic MAFIA algorithm and DFMFI algorithm respectively.

Figure 4 shows the execution of the three algorithms under different support degrees in the Mushroom dataset. It can be seen that the efficiency of this method is significantly higher than the other two algorithms, and the execution efficiency has advantages even when the minimum support degree is very low. Figure 5 shows the implementation of the three algorithms on the hype microblog dataset. It can be seen that this method has the highest execution efficiency.

3) Parameter threshold selection

Figure 6 and Figure 7 show the results of the largest frequent itemsets found from the hype microblog data set under different minimum support numbers. Figures 6 and 7 show the number of itemsets in the largest frequent itemset and the largest The length varies with the minimum number of supports. In combination with the research background of the present invention, it can be found that the larger the setting of minSup (minimum number of supports), the greater the suspicion of speculation of the account group found, but the group size and number will also decrease thereupon; otherwise, the smaller the setting of minSup, The smaller the suspected account group speculation is, the larger the group size and number will be. To this end, it is necessary to set a reasonable threshold for minSup to discover groups with a certain scale and high suspicion of speculation.

On the other hand, when merging the itemsets in the largest frequent itemset, the setting of minOR will also directly affect the size of the merged itemsets. Through the continuous analysis of the data, the minOR is set to 50%, that is, when more than half of the items in the two item sets are the same, they will be merged.

In order to further determine the value of minSup, Table 1 lists the results of merging the largest frequent itemsets when minSup=3, 4, and 5 respectively, sorted by the length of the merged itemsets, and only the first 8 itemsets are listed here (Suspected hype groups). It can be seen from the table that when minSup=3 and 5, except for the first item set, the scale of other itemsets is very small; when minSup=4, the size of the itemset does not change sharply, and The scale is appropriate, indicating that the value is relatively reasonable. .

Table 1 Merging results of maximum frequent itemsets under different support numbers

serial number minSup=3 minSup=4 minSup=5 1 14,863 2,623 963 2 311 1,755 65 3 156 688 29 4 77 410 19 5 59 129 9

6 56 98 9 7 55 82 7 8 55 54 5

4) Accuracy analysis

In order to verify the accuracy of the hype group discovery algorithm described in the present invention, that is, the proportion of actual hype accounts in the discovered hype groups, the accuracy of the results is comprehensively verified by combining the existing hype account identification method based on multi-feature analysis and the manual labeling method. Assuming that the hype group to be verified is H, first use the existing hype account identification method based on multi-feature analysis to identify each account, and the obtained hype account set is recorded as H ₁ ; then, use the method of manual labeling to identify the remaining accounts Discrimination is carried out, and the obtained hype account set is denoted as H ₂ , and the calculation formula for the accuracy rate of hype group H is:

$\mathrm{<span\; class="notranslate">\; Precision</span>}<span\; class="notranslate">\; =</span>\frac{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span>}{<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; |</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; 100</span>}<span\; class="notranslate">\; \%</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

In the above formula, |H| represents the total number of accounts in H, and |H ₁ |+|H ₂ | represents the actual number of speculation accounts in H. In Table 1, some groups whose minSup=4 and whose group size (ie item set length) is greater than 100 are verified, and the specific results are shown in Table 2.

Table 2 The accuracy rate of hype group discovery (minSup=4)

serial number |H ₁ | |H ₂ | |H| Precision 1 2,016 451 2,623 94.1% 2 1,465 163 1,755 92.8% 3 571 78 688 94.3% 4 354 33 410 94.4% 5 109 10 129 92.2%

It can be seen from Table 2 that for each hype group found by this method, the proportion of actual hype accounts is higher than 90%, indicating that this method can identify more hidden hype accounts (namely H ₂ ), while These accounts are often hype accounts that occasionally participate in the hype but have a huge influence. It can be seen that the present invention has practical application value and huge economic and social benefits.

Claims (3)

1. A microblog hype group discovery method based on maximum frequent itemset mining, is characterized in that, comprises the steps: (1) Sample collection of hype microblogs: take the relevance of hype microblogs as clues, and obtain a collection of accounts participating in the spread of hype microblogs based on crawler technology or the public open platform of microblogs; (2) Construction of transactional database: taking a single microblog as a transaction and accounts participating in the spread of microblogs as items, construct a hype microblog transactional database; (3) Maximum frequent itemset mining: For each transaction in the transaction database corresponding to the microblog group to be detected, use the iterative intersection method to find out the maximum frequent itemsets contained in all transactions, and obtain several maximum frequent itemsets; Due to the hype that each transaction in the microblog transaction database contains tens of thousands of items, directly mining the largest frequent item set in the original transaction database will affect the efficiency of algorithm execution, and use the binary search method to quickly eliminate non-frequent items in transactions project, find out the candidate set of the largest frequent item set, and reduce the size of the transaction database; (4) Maximum frequent itemset merging: For each maximum frequent itemset, calculate the overlap rate between itemsets, merge the largest frequent itemsets, try to merge smaller itemsets into larger itemsets, and ensure After merging, the accounts in the item set still have certain relevance; by reducing the size of the transaction database, reducing the number of intersections, and when taking the intersection between transactions, use the binary search method to determine whether a certain item is included in the transaction, so as to improve the efficiency of mining the largest frequent itemset , so as to discover Weibo hype groups. 2. the microblog hype group discovery method based on maximum frequent itemset mining according to claim 1, is characterized in that, comprises hype microblog transaction library, maximum frequent item set mining and maximum frequent item set merging part, hype microblog The transaction database building module is mainly responsible for collecting data and preprocessing to construct the transaction database D; the maximum frequent itemset mining module first screens the candidate maximum frequent itemsets based on the binary search method, and then mines the maximum frequent itemsets from the transaction database D based on the iterative intersection method. Itemset MFS; the maximum frequent itemset merging module mainly merges MFS to restore the real hype group. The specific steps are: 1) Collect hype Weibo samples The collection of hype microblog samples is the first step to realize the present invention. The selection of microblog samples should be relevant. We should learn from the existing mature discrimination methods or expert systems. There are two ways to collect hype microblog samples: one method is to choose crawler technology to download web pages from microblog web pages, analyze the page structure and extract the information of microblog dissemination accounts; One method is to call the public open platform of Weibo, and call the API function provided by the Weibo official to obtain the information of the Weibo dissemination account; According to the algorithmic analysis conditions of the hype account to be mined, the content of the sample collection should include the microblog identification number, the microblog account identification number, and the basic information of the microblog account; 2) Build a transactional database Transform the discovery of hype groups into the largest frequent itemset mining in data mining. Based on the collection of hype microblog samples, the corresponding transactions of hype microblogs and the items in the corresponding transactions of accounts participating in microblog forwarding are constructed to build a transaction database. as shown in picture 2; 3) Screening of candidate maximum frequent itemsets based on binary search Due to the hype that each transaction in the microblog transaction database contains tens of thousands of items, directly mining the largest frequent item set in the original transaction database will affect the efficiency of algorithm execution. Based on the binary search method, it can quickly eliminate the items in the transaction. For non-frequent items, find out the candidate set of the largest frequent itemset, reduce the size of the transaction database, given the transaction database D, the minimum support number S, and filter the candidate maximum frequent itemset, the method is: (1) Sort the transactions in the transaction library D from the largest to the smallest according to the number of items (2) Record frequent item collection , the set of infrequent items ;Start from i=1, traverse each transaction T _i in D in order (1≤i≤|D|), for each item u in transaction T _i : a) If u∈FI, keep u; b) If u∈NFI, remove u from T _i ; c) if , go to the next step to judge whether u is a frequent item; (3) Start traversing the remaining transactions from j=i+1, and use the binary search method to judge whether u is contained in T _j , i<j≤|D|, and the termination condition is: a) When the number of transactions containing u reaches S, it means that u is a frequent item, and u is added to FI; b) When the sum of the number of remaining transactions and the number of transactions containing u is less than S, it means that u is an infrequent item, and u is removed from T _i . If the number of transactions containing u is greater than 1, it means that u Also appears in a transaction other than T _i , then add u to NFI; (4) After eliminating the non-frequent items in all affairs in D, the reduced transaction library _D1 can be obtained; 4) Mining of maximum frequent itemsets based on iterative intersection: Mining the maximum frequent itemset by iteratively taking the intersection of transactions, given the reduced transaction database D ₁ , the minimum support number S, the method of mining the maximum frequent itemset is as follows: (1) Sort the transactions in the transaction database D ₁ according to the number of items from large to small, so as to find the largest frequent item set as soon as possible, in order to reduce the size of the transaction database, merge the repeated transactions in the transaction database, and count the number of transactions ; (2) In order to reduce the number of intersections, for a transaction T _i , 1≤i≤|D ₁ |-S+1, starting from i=1, first find the transaction set that contains any item in T _i , T _j |T _j contains at least one item in T _i ; j>i), T _i takes the intersection with T _j in turn, moves the intersection of the two into the new transaction library D ₂ , and removes T _j at the same time, ; (3) For the transaction T in the new transaction database D ₂ , if T is obtained by taking the intersection of not less than S transactions, then move T into the maximum frequent candidate item set set MFCS, and remove the child of T in D ₂ affairs; (4) If the number of remaining transactions in the new transaction library D ₂ is less than S, then end the processing of the transaction library D ₂ and return to the upper-level transaction library; otherwise, start the process from the first step to D ₂ ; (5) When the number of remaining transactions in the transaction database D ₁ is less than S, that is, i>|D ₁ |-S+1, end the processing of the current transaction database D ₁ ; (6) Merge the itemsets in the MFCS and eliminate the non-maximal frequent itemsets at the same time, and the final result is the maximum frequent itemsets set MFS sought; 5) Maximum frequent itemsets merge: Due to the limitation of the minimum support number, the maximum frequent itemsets in MFS are small in size, and there are a large number of overlapping items between some itemsets. The account groups represented by these itemsets are likely to belong to the same hype group. In order to solve this The problem is to use the overlap rate to reflect the similarity between two item sets, set the item set X ₁ , X ₂ ∈ MFS, and record the overlap rate of X ₁ and X ₂ as: $\mathrm{<span\; class="notranslate">\; ORate</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&cap;</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span>}{\mathrm{<span\; class="notranslate">\; Min</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; ,</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; )</span>}$ In the above formula, |X ₁ ∩X ₂ | represents the number of overlapping items between X ₁ and X ₂ , Min(|X ₁ |,|X ₂ |) represents the number of items in a smaller itemset, and itemsets are merged The method is: (1) Sort the maximum frequent itemsets in the MFS from the largest to the smallest according to the number of items; (2) Traversing each maximum frequent itemset in MFS, starting from i=1, for , if ORate(X _i ,X _j )≥minOR, i<j≤|MFS|, then add the union of Xi and X _j to the new set MMFS _, and remove X _j at the same time; (3) Repeat the above two steps for the itemsets in MMFS; (4) When the overlapping rate of any two itemsets in MMFS is less than minOR, end. 3. the microblog hype group discovery method based on maximum frequent itemset mining according to claim 2 is characterized in that, in the described step 1), the collection of hype microblog samples should meet the following conditions: a. Select popular microblogs with a relatively high number of retweets; b. The time span of Weibo releases is <180 days; to facilitate the discovery of hype groups.