Fraud risk management method and system based on big data

Abstract

The invention relates to the technical field of data risk management, and discloses a fraud risk management method and system based on big data. The method comprises the following steps: collecting identification strategy contents accumulated in a counterfeit website identification process to construct an identification strategy sample set, performing discretization processing on the identification strategy contents in the identification strategy sample set by using an identification algorithm, and segmenting each identification strategy in the identification strategy sample set into strategy factors through discretization processing; the segmented strategy factors are processed, and the processed strategy factors are clustered at the same time; according to the website risk identification method and device, the strategy factors are clustered, risk scoring of the websites collected in real time is achieved in the mode that the clustered strategy factors and the website content data collected in real time are subjected to semantic relevancy matching, finally, the website fraud risk is judged in the mode that the website risk scoring threshold value is set, and the efficiency and accuracy of website risk identification are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data risk management, and in particular to a fraud risk management method and system based on big data. Background Art

[0002] Fake websites are characterized by rapid changes and short survival time. Different identification strategies need to be adopted for different types and periods of fake websites. A large number of identification strategies have been accumulated over many years of practice. Therefore, the main problem faced by the identification algorithm is that the identification strategy library continues to grow, resulting in a continuous decrease in identification efficiency.

[0003] At present, the active discovery and identification algorithm of fake websites adopts a single strategy independent judgment method. If there are N identification strategies, and the average time consumed by each identification strategy is M, then the total identification time is T0 = N*M. M is a constant, so T0 is linearly related to N. When N continues to grow, T0 will also continue to increase, and the identification efficiency will continue to decrease.

[0004] The prior art CN112949954A obtains sample data, cleans the sample data, removes errors, redundancy and noise, and unifies the dimensions of the data; at the same time, based on the initially established text framework, it obtains structured data of text information released by the sample company, analyzes the importance and importance ranking of each financial indicator, and formulates model evaluation standards to identify financial fraud based on the standards. However, it does not solve the problem that the identification of financial fraud is increasing and the identification efficiency is continuously reduced, and has great limitations. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] In view of the deficiencies in the prior art, the present invention provides a fraud risk management method and system based on big data, which has the advantages of real-time and high efficiency, and solves the problem of decreasing efficiency in identifying counterfeit websites.

[0007] (II) Technical solution

[0008] In order to solve the above-mentioned technical problem of the continuously decreasing efficiency of counterfeit website identification, the present invention provides the following technical solutions:

[0009] The present invention discloses a fraud risk management method based on big data, which specifically comprises the following steps:

[0010] S1. Collect the identification strategy content accumulated in the process of identifying fake websites to build an identification strategy sample set;

[0011] S2, using the recognition algorithm to discretize the recognition strategy content in the recognition strategy sample set, dividing each recognition strategy in the recognition strategy sample set into strategy factors through the discretization process, and saving the divided strategy factors to the recognition strategy sample set;

[0012] S3, performing data processing on the strategy factors in the identification strategy sample set to obtain processed strategy factors;

[0013] S4, clustering the processed strategy factors to obtain clustered strategy factors; the clustering results include: processed strategy factor content data, strategy factor weights, the number of clusters after clustering, and the number of strategy factors in each cluster;

[0014] S5. collecting website content data in real time, identifying the website content data collected in real time based on the clustered strategy factors, and performing risk scoring on the website content data collected in real time;

[0015] The present invention constructs an identification strategy sample set by collecting the identification strategy contents accumulated in the process of identifying counterfeit websites, and discretizes the contents of the identification strategies in the identification strategy sample set by using an identification algorithm based on the contents of the identification strategies, and divides each identification strategy in the identification strategy sample set into strategy factors; and processes the divided strategy factors, and clusters the processed strategy factors; and realizes risk scoring of real-time collected websites by matching the clustered strategy factors with the website content data collected in real time in terms of semantic relevance. Finally, the website fraud risk is judged by setting a website risk score threshold, thereby improving the efficiency and accuracy of website risk identification.

[0016] Preferably, the discretization of the identification strategy content in the identification strategy sample set using the identification algorithm comprises the following steps:

[0017] Extract candidate words in the content of the recognition strategy, and match the extracted candidate words in the entries of Wikipedia. By using the document connection information graph and classification structure graph of the Wikipedia knowledge base, calculate the semantic relevance of the candidate words, and construct a candidate word set in the current extraction and recognition strategy content based on the semantic relevance calculation results;

[0018] The semantic relevance calculation result includes the semantic relevance between the extracted candidate words and the Wikipedia entry;

[0019] The semantic relevance calculation is as follows:

[0020]

[0021] Among them, sim(x ij,y1) indicates that the semantic relevance value between the extracted candidate word and the Wikipedia entry is in the interval [-1,1], x ij represents the jth candidate word in the i-th recognition strategy content, y1 represents the first entry in Wikipedia, Lenx ij Dep represents the breadth-first traversal path information of the concept node of the j-th candidate word in the i-th recognition strategy content on the classification structure diagram, Leny1 represents the breadth-first traversal path information of the concept node of the first entry in Wikipedia on the classification structure diagram, pub Depx represents the nearest common root node between the candidate word and the Wikipedia entry in the classification structure diagram. ij Depy1 represents the depth information of the concept node of the first entry in Wikipedia on the classification structure diagram.

[0022] When the semantic relevance value between the extracted candidate word and the Wikipedia entry is in the interval [-1,0], it means that the extracted candidate word does not match the Wikipedia entry, and the candidate word in the recognition strategy content is re-extracted;

[0023] When the semantic relevance value between the extracted candidate word and the Wikipedia entry is in the interval of (0,1], it means that the extracted candidate word matches the Wikipedia entry, and the extracted candidate word is added to the candidate word set in the current extraction and recognition strategy content;

[0024] Taking each candidate word in the candidate word set in the current extraction recognition strategy content as a strategy factor of the current recognition strategy;

[0025] The discretization process includes extracting candidate words in the recognition strategy content and calculating semantic relevance;

[0026] Discretization is performed on each recognition strategy in the recognition strategy sample set, and the strategy factors of each recognition strategy in the recognition strategy sample set are summarized, and the summarized strategy factors are saved in the recognition strategy sample set.

[0027] The present invention extracts candidate words for each recognition strategy in the recognition strategy sample set, and segmented each recognition strategy in the recognition strategy sample set by calculating semantic relevance, and generates strategy factors, thereby reducing the redundancy of fraud risk identification and improving the efficiency of fraud risk identification.

[0028] Preferably, the data processing of the strategy factors in the identification strategy sample set includes the following steps:

[0029] S31, calculating the weight of each strategy factor;

[0030] S32, performing a deduplication operation on repeated strategy factors in the identification strategy sample set to obtain deduplicated strategy factors;

[0031] Traverse the strategy factors in the strategy sample set, and when duplicate strategy factors are detected during the traversal process, retain the first strategy factor and delete the duplicate strategy factors;

[0032] S33, arranging the deduplicated strategy factors according to the calculated weights, and summarizing them to obtain the processed strategy factors;

[0033] Set the one with the largest weight to be ranked first.

[0034] Preferably, the weight calculation formula of the strategy factor is as follows:

[0035] weight k =α×f k +β×loc k +γ×sen k

[0036] Among them, weight k represents the calculated k-th strategy factor weight, f k represents the frequency of the kth strategy factor, α represents the adjustment factor of the set frequency, loc k represents the regional position of the kth strategy factor, β represents the adjustment factor of the set regional position, sen k It represents the number of identification strategies contained in the kth strategy factor, and γ represents the adjustment factor of the number of identification strategies contained in the set strategy factor.

[0037] The present invention calculates the weight of each strategy factor by preprocessing each strategy factor after segmentation, removes duplicate strategy factors, and ensures the efficiency of fraud risk identification.

[0038] Preferably, clustering the processed strategy factors comprises the following steps:

[0039] Set the weight range of the strategy factor and use the bi-kmeans algorithm to cluster the strategy factor within the set weight range:

[0040] Set the number of clusters h in the clustering results and the number of strategy factors r in each cluster;

[0041] S41, treating the strategy factors within the set weight range as a cluster;

[0042] S42, judging whether the number of clusters is less than h, executing S43 when the number of clusters is less than h, and executing S44 when the number of clusters is equal to h;

[0043] S43, for each cluster, select a strategy factor in the cluster as the cluster center, calculate the similarity between other strategy factors in the cluster and the cluster center, divide the cluster based on the calculated similarity value, and execute S42 after the division is completed to determine whether to continue the division;

[0044] S44, determine whether the number of strategy factors within the cluster is less than r. When the number of strategy factors within the cluster is less than or equal to r, output the clustering result. When the number of strategy factors within the cluster is greater than h, set h=h+1, increase the number of clusters and execute S43, and divide the cluster again if the number of strategy factors within the cluster is greater than h.

[0045] Preferably, the calculating the similarity between other strategy factors in the cluster and the cluster center, and dividing the cluster based on the calculated similarity value comprises the following steps:

[0046] The similarity calculation is as follows:

[0047]

[0048] Among them, cos(A,B) represents the similarity value between other strategic factors in the cluster and the cluster center. The calculated similarity value ranges from [-1,1]. A represents a strategic factor in the cluster, and B represents the currently selected cluster center B.

[0049] Since the calculated similarity value range is in the interval [-1,1];

[0050] When the similarity value between other strategy factors in the cluster and the cluster center is in the interval [-1,0], it is determined that other strategy factors in the cluster are not similar to the cluster center;

[0051] When the similarity value between other strategy factors in the cluster and the cluster center is in the interval (0,1], it is determined that other strategy factors in the cluster are similar to the cluster center;

[0052] Based on the similar results of other strategy factors within the cluster and the cluster center, the cluster is divided into two.

[0053] The present invention achieves clustering of strategic factors by using a clustering algorithm, thereby further improving the efficiency of fraud risk identification.

[0054] Preferably, the step of identifying the website content data collected in real time based on the clustered strategy factors and performing risk scoring on the website content data collected in real time comprises the following steps:

[0055] S51, collecting website content data in real time, and extracting keywords from the website content data collected in real time;

[0056] S52, matching the keywords in the website content data collected in real time extracted with the strategy factors in the clustering results one by one based on the semantic relevance calculation method;

[0057] It is set that every time a match is successful, the weights corresponding to the successfully matched strategy factors are accumulated, and the accumulated weights are used as the risk score of the website content data collected in real time;

[0058] S53, setting a risk score threshold for website content data collected in real time;

[0059] When the accumulated risk score of the website content data collected in real time is greater than or equal to the set threshold, the current website is determined to be a fake website;

[0060] When the accumulated risk score of the website content data collected in real time is less than the set threshold, the current website is determined to be a normal website.

[0061] The present invention compares and matches the clustered strategy factors with the website content data collected in real time to achieve the identification of the websites collected in real time. At the same time, the fraud risk level of the websites is judged by scoring the risks of the websites collected in real time and setting scoring thresholds, thereby ensuring the accuracy of fraud risk identification.

[0062] The present invention also discloses a fraud risk management system based on big data, comprising: a database, a semantic recognition module, a data collection module, a data processing module, a clustering module and a risk scoring module;

[0063] The database is used to store various accumulated recognition strategies;

[0064] The semantic recognition module is used to extract various recognition strategies in the database, segment the various recognition strategies, and generate strategy factors;

[0065] The data processing module is used to process the strategy factors generated by segmentation and transmit the processed strategy factors to the clustering module;

[0066] The clustering module is used to cluster the received strategic factors and output the clustering results to be stored in the database;

[0067] The data collection module is used to collect website content data in real time;

[0068] The risk scoring module is used to score and display the websites collected in real time according to the clustered strategy factors.

[0069] (III) Beneficial effects

[0070] Compared with the prior art, the present invention provides a fraud risk management method and system based on big data, which has the following beneficial effects:

[0071] 1. The invention constructs an identification strategy sample set by collecting the identification strategy contents accumulated in the process of identifying counterfeit websites, and discretizes the contents of the identification strategies in the identification strategy sample set using an identification algorithm based on the contents of the identification strategies, and divides each identification strategy in the identification strategy sample set into strategy factors; and processes the divided strategy factors and clusters the processed strategy factors; and realizes risk scoring of real-time collected websites by matching the clustered strategy factors with the real-time collected website content data through semantic relevance. Finally, the website fraud risk is judged by setting a website risk score threshold, thereby improving the efficiency and accuracy of website risk identification.

[0072] 2. The invention extracts candidate words for each recognition strategy in the recognition strategy sample set, and segments each recognition strategy in the recognition strategy sample set by calculating semantic relevance, and generates strategy factors, thereby reducing the redundancy of fraud risk identification and improving the efficiency of fraud risk identification.

[0073] 3. The invention calculates the weight of each strategy factor and removes duplicate strategy factors by preprocessing each strategy factor after segmentation, thereby ensuring the efficiency of fraud risk identification.

[0074] 4. The invention realizes clustering of strategy factors by using clustering algorithm, thereby further improving the efficiency of fraud risk identification.

[0075] 5. The invention compares and matches the clustered strategy factors with the website content data collected in real time to achieve the identification of the websites collected in real time. At the same time, the invention determines the fraud risk level of the websites by scoring the risks of the websites collected in real time and setting scoring thresholds, thereby ensuring the accuracy of fraud risk identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 This is a schematic diagram of the process structure of the fraud risk management method for big data of the present invention. DETAILED DESCRIPTION

[0077] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0078] The present invention discloses a fraud risk management method based on big data, which specifically comprises the following steps:

[0079] S1. Collect the identification strategy content accumulated in the process of identifying fake websites to build an identification strategy sample set;

[0080] The sample set of the recognition strategy constructed is set to X = {x1, x2, ..., x n};

[0081] Among them, x1 represents the first recognition strategy in the recognition strategy sample set, x n represents the nth recognition strategy in the recognition strategy sample set;

[0082] S2, using the recognition algorithm to discretize the recognition strategy content in the recognition strategy sample set, dividing each recognition strategy in the recognition strategy sample set into strategy factors through the discretization process, and saving the divided strategy factors to the recognition strategy sample set;

[0083] Discretizing the content of the recognition strategy in the recognition strategy sample set using the recognition algorithm includes the following steps:

[0084] Extract candidate words in the content of the recognition strategy, and match the extracted candidate words in the entries of Wikipedia. By using the document connection information graph and classification structure graph of the Wikipedia knowledge base, calculate the semantic relevance of the candidate words, and construct a candidate word set in the current extraction and recognition strategy content based on the semantic relevance calculation results;

[0085] The semantic relevance calculation result includes the semantic relevance between the extracted candidate words and the Wikipedia entry;

[0086] The semantic relevance calculation is as follows:

[0087]

[0088] Among them, sim(x ij ,y1) indicates that the semantic relevance value between the extracted candidate word and the Wikipedia entry is in the interval [-1,1], x ij represents the jth candidate word in the i-th recognition strategy content, y1 represents the first entry in Wikipedia, Lenx ij Dep represents the breadth-first traversal path information of the concept node of the j-th candidate word in the i-th recognition strategy content on the classification structure diagram, Leny1 represents the breadth-first traversal path information of the concept node of the first entry in Wikipedia on the classification structure diagram, pub Depx represents the nearest common root node between the candidate word and the Wikipedia entry in the classification structure diagram. ijDepy1 represents the depth information of the concept node of the first entry in Wikipedia on the classification structure diagram.

[0089] When the semantic relevance value between the extracted candidate word and the Wikipedia entry is in the interval [-1,0], it means that the extracted candidate word does not match the Wikipedia entry, and the candidate word in the recognition strategy content is re-extracted;

[0090] When the semantic relevance value between the extracted candidate word and the Wikipedia entry is in the interval of (0,1], it means that the extracted candidate word matches the Wikipedia entry, and the extracted candidate word is added to the candidate word set in the current extraction and recognition strategy content;

[0091] Further, each candidate word in the candidate word set in the current extraction recognition strategy content is used as a strategy factor of the current recognition strategy;

[0092] Furthermore, the discretization process includes extracting candidate words in the recognition strategy content and calculating semantic relevance;

[0093] Further, each recognition strategy in the recognition strategy sample set is discretized, and the strategy factors of each recognition strategy in the recognition strategy sample set are summarized, and the summarized strategy factors are saved in the recognition strategy sample set;

[0094] S3, performing data processing on the strategy factors in the identification strategy sample set to obtain processed strategy factors;

[0095] Data processing of the strategy factors in the identification strategy sample set includes the following steps:

[0096] S31, calculating the weight of each strategy factor;

[0097] The weight calculation formula of the strategy factor is as follows:

[0098] weight k =α×f k +β×loc k +γ×sen k

[0099] Among them, weight k represents the calculated k-th strategy factor weight, f k represents the frequency of the kth strategy factor, α represents the adjustment factor of the set frequency, loc k represents the regional position of the kth strategy factor, β represents the adjustment factor of the set regional position, sen krepresents the number of identification strategies contained in the kth strategy factor, and γ represents the adjustment factor of the number of identification strategies contained in the set strategy factor;

[0100] S32, performing a deduplication operation on repeated strategy factors in the identification strategy sample set to obtain deduplicated strategy factors;

[0101] Traverse the strategy factors in the strategy sample set, and when duplicate strategy factors are detected during the traversal process, retain the first strategy factor and delete the duplicate strategy factors;

[0102] S33, arranging the deduplicated strategy factors according to the calculated weights, and summarizing them to obtain the processed strategy factors;

[0103] Set the one with the largest weight to be ranked first;

[0104] S4, clustering the processed strategy factors to obtain clustered strategy factors; the clustering results include: processed strategy factor content data, strategy factor weights, the number of clusters after clustering, and the number of strategy factors in each cluster;

[0105] Clustering the processed strategic factors includes the following steps:

[0106] Set the weight range of the strategy factor and use the bi-kmeans algorithm to cluster the strategy factor within the set weight range:

[0107] Set the number of clusters h in the clustering results and the number of strategy factors r in each cluster;

[0108] S41, treating the strategy factors within the set weight range as a cluster;

[0109] S42, judging whether the number of clusters is less than h, executing S43 when the number of clusters is less than h, and executing S44 when the number of clusters is equal to h;

[0110] S43, for each cluster, selecting a strategy factor in the cluster as the cluster center, calculating the similarity between other strategy factors in the cluster and the cluster center, and dividing the cluster based on the calculated similarity value;

[0111] The similarity calculation is as follows:

[0112]

[0113] Among them, cos(A,B) represents the similarity value between other strategic factors in the cluster and the cluster center. The calculated similarity value ranges from [-1,1]. A represents a strategic factor in the cluster, and B represents the currently selected cluster center B.

[0114] Furthermore, since the calculated similarity value range is in the interval [-1,1];

[0115] When the similarity value between other strategy factors in the cluster and the cluster center is in the interval [-1,0], it is determined that other strategy factors in the cluster are not similar to the cluster center;

[0116] When the similarity value between other strategy factors in the cluster and the cluster center is in the interval (0,1], it is determined that other strategy factors in the cluster are similar to the cluster center;

[0117] Based on the similar results of other strategy factors in the cluster and the cluster center, the cluster is divided into two;

[0118] Further, after the division is completed, S42 is executed to determine whether further division is required;

[0119] S44, judging whether the number of strategy factors in the cluster is less than r, outputting the clustering result when the number of strategy factors in the cluster is less than or equal to r, and setting h=h+1 when the number of strategy factors in the cluster is greater than h, increasing the number of clusters and executing S43, and re-dividing the clusters when the number of strategy factors in the cluster is greater than h;

[0120] S5. collecting website content data in real time, identifying the website content data collected in real time based on the clustered strategy factors, and performing risk scoring on the website content data collected in real time;

[0121] Identifying the website content data collected in real time based on the clustered strategy factors and scoring the website content data collected in real time includes the following steps:

[0122] S51, collecting website content data in real time, and extracting keywords from the website content data collected in real time;

[0123] S52, matching the keywords in the website content data collected in real time extracted with the strategy factors in the clustering results one by one based on the semantic relevance calculation method;

[0124] It is set that every time a match is successful, the weights corresponding to the successfully matched strategy factors are accumulated, and the accumulated weights are used as the risk score of the website content data collected in real time;

[0125] S53, setting a risk score threshold for website content data collected in real time;

[0126] When the accumulated risk score of the website content data collected in real time is greater than or equal to the set threshold, the current website is determined to be a fake website;

[0127] When the accumulated risk score of the website content data collected in real time is less than the set threshold, the current website is determined to be a normal website;

[0128] The present invention also discloses a fraud risk management system based on big data, comprising: a database, a semantic recognition module, a data collection module, a data processing module, a clustering module and a risk scoring module;

[0129] The database is used to store various accumulated recognition strategies;

[0130] The semantic recognition module is used to extract various recognition strategies in the database, segment the various recognition strategies, and generate strategy factors;

[0131] The data processing module is used to process the strategy factors generated by segmentation and transmit the processed strategy factors to the clustering module;

[0132] The clustering module is used to cluster the received strategic factors and output the clustering results to be stored in the database;

[0133] The data collection module is used to collect website content data in real time;

[0134] The risk scoring module is used to score and display the websites collected in real time according to the clustered strategy factors.

[0135] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A fraud risk management method based on big data, characterized in that: The following steps are involved: S1. Collect the identification strategy content accumulated in the process of identifying fake websites to build an identification strategy sample set; S2, using the recognition algorithm to discretize the recognition strategy content in the recognition strategy sample set, dividing each recognition strategy in the recognition strategy sample set into strategy factors through the discretization process, and saving the divided strategy factors to the recognition strategy sample set; S3, performing data processing on the strategy factors in the identification strategy sample set to obtain processed strategy factors; S4, clustering the processed strategic factors to obtain clustered strategic factors; The clustering results include: processed strategic factor content data, strategic factor weights, the number of clusters after clustering, and the number of strategic factors in each cluster; S5. Collect website content data in real time, identify the website content data collected in real time based on the clustered strategy factors, and perform risk scoring on the website content data collected in real time.

2. The fraud risk management method based on big data according to claim 1, characterized in that: The discretization of the identification strategy content in the identification strategy sample set using the identification algorithm comprises the following steps: Extract candidate words in the content of the recognition strategy, and match the extracted candidate words in the entries of Wikipedia. By using the document connection information graph and classification structure graph of the Wikipedia knowledge base, calculate the semantic relevance of the candidate words, and construct a candidate word set in the current extraction and recognition strategy content based on the semantic relevance calculation results; The semantic relevance calculation result includes the semantic relevance between the extracted candidate words and the Wikipedia entry; The semantic relevance calculation is as follows: Among them, sim(x ij ,y1) indicates that the semantic relevance value between the extracted candidate word and the Wikipedia entry is in the interval [-1,1], x ij represents the jth candidate word in the i-th recognition strategy content, y1 represents the first entry in Wikipedia, Lenx ij Dep represents the breadth-first traversal path information of the concept node of the j-th candidate word in the i-th recognition strategy content on the classification structure diagram, Leny1 represents the breadth-first traversal path information of the concept node of the first entry in Wikipedia on the classification structure diagram, pub Depx represents the nearest common root node between the candidate word and the Wikipedia entry in the classification structure diagram. ij Depy1 represents the depth information of the concept node of the first entry in Wikipedia on the classification structure diagram. When the semantic relevance value between the extracted candidate word and the Wikipedia entry is in the interval [-1,0], it means that the extracted candidate word does not match the Wikipedia entry, and the candidate word in the recognition strategy content is re-extracted; When the semantic relevance value between the extracted candidate word and the Wikipedia entry is in the interval of (0,1], it means that the extracted candidate word matches the Wikipedia entry, and the extracted candidate word is added to the candidate word set in the current extraction and recognition strategy content; Taking each candidate word in the candidate word set in the current extraction recognition strategy content as a strategy factor of the current recognition strategy; The discretization process includes extracting candidate words in the recognition strategy content and calculating semantic relevance; Discretization is performed on each recognition strategy in the recognition strategy sample set, and the strategy factors of each recognition strategy in the recognition strategy sample set are summarized, and the summarized strategy factors are saved in the recognition strategy sample set.

3. The fraud risk management method based on big data according to claim 2, characterized in that: The data processing of the strategy factors in the identification strategy sample set comprises the following steps: S31, calculating the weight of each strategy factor; S32, performing a deduplication operation on repeated strategy factors in the identification strategy sample set to obtain deduplicated strategy factors; Traverse the strategy factors in the strategy sample set, and when duplicate strategy factors are detected during the traversal process, retain the first strategy factor and delete the duplicate strategy factors; S33, arranging the deduplicated strategy factors according to the calculated weights, and summarizing them to obtain the processed strategy factors; Set the one with the largest weight to be ranked first.

4. The fraud risk management method based on big data according to claim 3 is characterized in that: The weight calculation formula of the strategy factor is as follows: weight k =α×f k +β×loc k +γ×sen k Among them, weight k represents the calculated k-th strategy factor weight, f k represents the frequency of the kth strategy factor, α represents the adjustment factor of the set frequency, loc k represents the regional position of the kth strategy factor, β represents the adjustment factor of the set regional position, sen k It represents the number of identification strategies contained in the kth strategy factor, and γ represents the adjustment factor of the number of identification strategies contained in the set strategy factor.

5. The fraud risk management method based on big data according to claim 4 is characterized in that: The clustering of the processed strategy factors comprises the following steps: Set the weight range of the strategy factor and use the bi-kmeans algorithm to cluster the strategy factor within the set weight range: Set the number of clusters h in the clustering results and the number of strategy factors r in each cluster; S41, treating the strategy factors within the set weight range as a cluster; S42, judging whether the number of clusters is less than h, executing S43 when the number of clusters is less than h, and executing S44 when the number of clusters is equal to h; S43, for each cluster, select a strategy factor in the cluster as the cluster center, calculate the similarity between other strategy factors in the cluster and the cluster center, divide the cluster based on the calculated similarity value, and execute S42 after the division is completed to determine whether to continue the division; S44, determine whether the number of strategy factors within the cluster is less than r. When the number of strategy factors within the cluster is less than or equal to r, output the clustering result. When the number of strategy factors within the cluster is greater than h, set h=h+1, increase the number of clusters and execute S43, and divide the cluster again if the number of strategy factors within the cluster is greater than h.

6. The fraud risk management method based on big data according to claim 5, characterized in that: The calculating of the similarity between other strategy factors in the cluster and the cluster center and dividing the cluster based on the calculated similarity value comprises the following steps: The similarity calculation is as follows: Among them, cos(A,B) represents the similarity value between other strategic factors in the cluster and the cluster center. The calculated similarity value ranges from [-1,1]. A represents a strategic factor in the cluster, and B represents the currently selected cluster center B. Since the calculated similarity value range is in the interval [-1,1]; When the similarity value between other strategy factors in the cluster and the cluster center is in the interval [-1,0], it is determined that other strategy factors in the cluster are not similar to the cluster center; When the similarity value between other strategy factors in the cluster and the cluster center is in the interval (0,1], it is determined that other strategy factors in the cluster are similar to the cluster center; Based on the similar results of other strategy factors within the cluster and the cluster center, the cluster is divided into two.

7. The fraud risk management method based on big data according to claim 1, characterized in that: The method of identifying the website content data collected in real time based on the clustered strategy factors and performing risk scoring on the website content data collected in real time includes the following steps: S51, collecting website content data in real time, and extracting keywords from the website content data collected in real time; S52, matching the keywords in the website content data collected in real time extracted with the strategy factors in the clustering results one by one based on the semantic relevance calculation method; It is set that every time a match is successful, the weights corresponding to the successfully matched strategy factors are accumulated, and the accumulated weights are used as the risk score of the website content data collected in real time; S53, setting a risk score threshold for website content data collected in real time; When the accumulated risk score of the website content data collected in real time is greater than or equal to the set threshold, the current website is determined to be a fake website; When the accumulated risk score of the website content data collected in real time is less than the set threshold, the current website is determined to be a normal website.

8. A fraud risk management system based on big data that implements the fraud risk management method based on big data as described in any one of claims 1 to 7, characterized in that: It includes database, semantic recognition module, data collection module, data processing module and clustering module; The database is used to store various accumulated recognition strategies; The semantic recognition module is used to extract various recognition strategies in the database, segment the various recognition strategies, and generate strategy factors; The data processing module is used to process the strategy factors generated by segmentation and transmit the processed strategy factors to the clustering module; The clustering module is used to cluster the received strategic factors and output the clustering results to be stored in the database; The data collection module is used to collect website content data in real time.

9. A fraud risk management system based on big data according to claim 8, characterized in that: It also includes a risk scoring module, which is used to score and display the websites collected in real time according to the clustered strategy factors.

Images