CN113343073A

CN113343073A - Big data and artificial intelligence based information fraud identification method and big data system

Info

Publication number: CN113343073A
Application number: CN202110651532.1A
Authority: CN
Inventors: 张丽艳; 张仕融; 张士光; 张洪艳
Original assignee: Guangzhou Xingsheng Communication Technology Co ltd
Current assignee: Guangzhou Xingsheng Communication Technology Co ltd
Priority date: 2021-06-11
Filing date: 2021-06-11
Publication date: 2021-09-03

Abstract

The embodiment of the disclosure provides an information fraud identification method and a big data system based on big data and artificial intelligence, carrying out fraud intention recognition on each first fraud discrimination vector, thus obtaining candidate fraud intentions after carrying out preliminary updating on the preset fraud intentions, a third example discrimination vector with better fraud recognition reference value can be obtained by fusing different discrimination vectors, and carrying out fraud intention identification on the third example discrimination vector according to the candidate fraud intention, thus obtaining the target fraud intention after secondary updating of the candidate fraud intention, wherein the target fraud intention can more accurately reflect the actual fraud intention of the target fraud trigger data, therefore, the identification effectiveness of the target fraud trigger data is improved, and the reference value of the pushing of the fraud reference content to the digital service equipment is improved, so that the fraud reference content can more effectively reflect the actual fraud intention of a fraud source.

Description

Big data and artificial intelligence based information fraud identification method and big data system

Technical Field

The disclosure relates to the technical field of big data fraud identification, and exemplarily relates to an information fraud identification method and a big data system based on big data and artificial intelligence.

Background

With the development of internet technology, the network information security problem is increased, and different levels of capital or other losses are brought to internet companies and individual users. At present, the internet risk events mainly comprise two types of stealing events and fraudulent events. Wherein, the fraud events can be personal fraud, merchant fraud, trojan horse, phishing, etc. Based on this, with the rapid development of big data and artificial intelligence technologies, in order to improve the accuracy of identifying the cheating intention of the cheating event, the related technologies usually perform feature matching based on some preset feature rules, however, this scheme needs to manually determine a large number of feature rules, easily generates omissions, and has extremely low efficiency, and once the identification is omitted, the timeliness and accuracy of pushing the cheating reference content will be seriously affected, resulting in a significant information security problem.

Disclosure of Invention

In order to overcome at least the above disadvantages in the prior art, the present disclosure is directed to providing an information fraud identification method and a big data system based on big data and artificial intelligence.

In a first aspect, the present disclosure provides an information fraud identification method based on big data and artificial intelligence, which is applied to a big data system, where the big data system is in communication connection with a plurality of digital service devices, and the method includes:

acquiring fraud event big data to be identified of the digital service equipment;

performing judgment vector extraction on the big data of the fraud event to be identified according to a fraud behavior identification model meeting the model convergence requirement to obtain a first fraud judgment vector of a plurality of fraud evaluation categories; each first fraud discrimination vector comprises a corresponding preset fraud intention;

carrying out fraud intention identification on each first fraud judgment vector to obtain candidate fraud intentions corresponding to target fraud trigger data on the big data of the fraud event to be identified;

obtaining a second fraud discrimination vector corresponding to the first fraud discrimination vector according to the first fraud discrimination vector, and performing vector weight fusion on the first fraud discrimination vector and the corresponding second fraud discrimination vector to obtain a corresponding third fraud discrimination vector;

and carrying out fraud intention identification on the third fraud discrimination vector according to the candidate fraud intention to obtain a target fraud intention corresponding to target fraud trigger data on the big data of the fraud event to be identified, and carrying out fraud reference content push on the digital service equipment according to the target fraud intention.

In a second aspect, an embodiment of the present disclosure further provides an information fraud identification system based on big data and artificial intelligence, where the information fraud identification system based on big data and artificial intelligence includes a big data system and a plurality of digital service devices communicatively connected to the big data system;

the big data system is used for:

According to any one of the above aspects, in the embodiments provided by the present disclosure, fraud intention recognition is performed on each first fraud discrimination vector, so as to obtain a candidate fraud intention after a preset fraud intention is primarily updated, a third example discrimination vector with a better fraud recognition reference value can be obtained by fusing different discrimination vectors, fraud intention recognition is performed on the third example discrimination vector according to the candidate fraud intention, so as to obtain a target fraud intention after a candidate fraud intention is secondarily updated, and the target fraud intention can more accurately reflect an actual fraud intention of target fraud trigger data, so as to improve the recognition effectiveness of the target fraud trigger data, so as to improve the reference value for carrying out fraud reference content push on the digital service device, so that the fraud reference content can more effectively reflect the actual fraud intention of a fraud source, and moreover, the preset feature rules for feature matching do not need to be manually set, and the efficiency of fraud intention identification and fraud reference content push is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings that need to be called in the embodiments are briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present disclosure, and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.

FIG. 1 is a schematic diagram of an application scenario of an information fraud identification system based on big data and artificial intelligence provided in an embodiment of the present disclosure;

FIG. 2 is a schematic flow chart of an information fraud identification method based on big data and artificial intelligence provided by an embodiment of the present disclosure;

FIG. 3 is a functional block diagram of an information fraud identification apparatus based on big data and artificial intelligence provided by an embodiment of the present disclosure;

fig. 4 is a schematic block diagram of a big data system for implementing the big data and artificial intelligence based information fraud identification method according to the embodiment of the present disclosure.

Detailed Description

The present disclosure is described in detail below with reference to the drawings, and the specific operation methods in the method embodiments can also be applied to the device embodiments or the system embodiments.

Fig. 1 is a schematic view of an application scenario of an information fraud identification system 10 based on big data and artificial intelligence according to an embodiment of the present disclosure. Big data and artificial intelligence based information fraud identification system 10 may include big data system 100 and a digital business apparatus 200 communicatively coupled to big data system 100. The big data and artificial intelligence based information fraud identification system 10 shown in FIG. 1 is only one possible example, and in other possible embodiments, the big data and artificial intelligence based information fraud identification system 10 may also include only at least some of the components shown in FIG. 1 or may also include other components.

In an embodiment that can be implemented independently, the big data system 100 and the digital service device 200 in the big data and artificial intelligence based information fraud recognition system 10 can cooperate to execute the big data and artificial intelligence based information fraud recognition method described in the following method embodiment, and the detailed description of the method embodiment can be referred to in the following steps of the big data system 100 and the digital service device 200.

In order to solve the technical problems in the background art, the big data and artificial intelligence based information fraud recognition method provided by the present embodiment may be executed by the big data system 100 shown in fig. 1, and the details of the big data and artificial intelligence based information fraud recognition method are described below.

Step S110, obtaining the big data of the example fraud event, and inputting the big data of the example fraud event into the initial fraud behavior recognition model, where the big data of the example fraud event includes an example fraud intention corresponding to the target fraud trigger data.

The example fraud big data refers to fraud big data used for AI training, and target fraud trigger data can be contained in the example fraud big data. The target fraud trigger data may specifically be independent service fraud data, such as credit service fraud data, privacy service fraud data, security service fraud data, and the like, or may also be specific service fraud data, such as fraud trigger data stolen during privacy upload, and the like. Fraud intent refers to intent classification information that is intent classified for target fraud trigger data. Fraud intent is typically represented by a classification data location to which the target fraud trigger data corresponds. An example fraud intent refers to a predetermined, accurately determined fraud intent for use as an actual fraud intent. The example fraud event big data comprises an example fraud intention corresponding to the target fraud trigger data, namely the example fraud event big data is fraud event big data of which the position data of the source where the target fraud trigger data is accurately determined in advance. The fraud identification model is an AI model for detecting targeted fraud trigger data in the fraud event big data.

In a separate embodiment, the big data system 100 can obtain the example fraud big data locally or from other terminals or cloud service platforms. After the big data system 100 acquires the big data of the example fraudulent event, the big data of the example fraudulent event is input into an initial fraudulent behavior recognition model, and AI training is performed on the fraudulent behavior recognition model through the big data of the example fraudulent event, so that a target fraudulent behavior recognition model is obtained.

Step S120, carrying out discrimination vector extraction on the example fraud event big data to obtain first example discrimination vectors of a plurality of fraud evaluation categories, wherein each first example discrimination vector comprises a corresponding preset fraud intention.

The judgment vector extraction is to map the big data of the fraud event to a feature interval of a preset judgment vector to obtain the big data feature of the fraud event, which can represent the essence of the big data of the fraud event and has a certain identification degree. The preset fraud intention refers to a preset fraud intention with a fixed fraud assessment category. The preset fraud intention may be fraud intents of a plurality of different fraud assessment categories, and is not particularly limited.

In an embodiment that can be implemented independently, after the big data system 100 inputs the big data of the example fraud event into the fraud recognition model, the big data of the example fraud event can be mapped by the fraud recognition model according to the fraud assessment categories, so as to extract the discrimination vectors of the big data of the example fraud event, obtain first example discrimination vectors of a plurality of fraud assessment categories, and mark and add various preset fraud intentions on each first example discrimination vector.

In an embodiment that can be implemented independently, the big data system 100 can mark each unit of the discriminant vector with various predetermined fraud intents. The big data system 100 may also select a part of the discrimination vector units from the discrimination vector as target discrimination vector units, and mark and add various preset fraud intentions on the target discrimination vector units, wherein the target discrimination vector units may be determined according to intention association information of the preset fraud intentions, and the goal is to associate and bind at least one preset fraud intention with each discrimination vector unit.

In an independently implementable embodiment, the fraud identification model includes a plurality of discrimination vector extraction layers, each of which is cascaded, and different discrimination vector extraction layers are used for performing discrimination vector extraction for different fraud assessment categories. Performing discrimination vector extraction on the sample fraud event big data to obtain a first sample discrimination vector of a plurality of fraud assessment categories, wherein the discrimination vector extraction comprises the following steps: and inputting the current first example discrimination vector output by the current discrimination vector extraction layer into a next discrimination vector extraction layer to obtain a first example discrimination vector associated with the fraud assessment category of the current first example discrimination vector.

In an independently implementable embodiment, the fraud identification model includes a plurality of discrimination vector extraction layers, each of which is cascaded, and different discrimination vector extraction layers are used for performing discrimination vector extraction for different fraud assessment categories. After the big data system 100 inputs the big data of the example fraud event into the fraud behavior recognition model, the big data of the example fraud event can be subjected to discriminant vector extraction through the first discriminant vector extraction layer to obtain a first example discriminant vector 1, the first example discriminant vector 1 is input into the second discriminant vector extraction layer to obtain a first example discriminant vector 2, the first example discriminant vector 2 is input into the third discriminant vector extraction layer to obtain a first example discriminant vector 3, and so on, and each first example discriminant vector is obtained according to the output data of each discriminant vector extraction layer. In addition, the fraud assessment category of the first example discrimination vector output by the adjacent discrimination vector extraction layer is also related, that is, the first example discrimination vector related to the fraud assessment category of the current first example discrimination vector can be obtained by inputting the current first example discrimination vector output by the current discrimination vector extraction layer into the next discrimination vector extraction layer.

For example, the fraud identification model comprises six discrimination vector extraction layers, after the big data of the example fraud event is input into the first discrimination vector extraction layer, the first example discrimination vector output by the current feature layer is used as the fraud data of the next discrimination vector extraction layer, and the fraud assessment categories of the first example discrimination vectors output by the adjacent discrimination vector extraction layers are also related. And the fraud behavior recognition model extracts the discrimination vectors from shallow to deep from the big data of the example fraud event to obtain a first example discrimination vector representing characteristic information of different fraud assessment categories. It can be understood that one discrimination vector unit on the discrimination vector with the smaller magnitude of the fraud assessment category corresponds to a larger vector segment on the sample fraud event big data, so that the discrimination vector with the smaller magnitude of the fraud assessment category is suitable for detecting the target fraud trigger data with the larger magnitude of the fraud assessment category, and one discrimination vector unit on the discrimination vector with the larger magnitude of the fraud assessment category corresponds to a smaller vector segment on the sample fraud event big data, so that the discrimination vector with the larger magnitude of the fraud assessment category is suitable for detecting the target fraud trigger data with the smaller magnitude of the fraud assessment category.

Step S130, carrying out fraud intention identification on each first example discrimination vector to obtain first identification fraud intentions corresponding to each preset fraud intention, and determining second identification fraud intentions from each first identification fraud intention according to intention loss data between each preset fraud intention and the example fraud intention.

The fraud intention identification means calculating the big data characteristics of the fraud event to obtain fraud intention identification data. The first identified fraud intention refers to the identified fraud intention adjusted by the preset fraud intention.

In an independently implemented embodiment, after the first example discrimination vectors are obtained, the big data system 100 may perform fraud intention recognition on each first example discrimination vector, obtain first recognition difference data corresponding to a preset fraud intention according to a fraud event big data feature of the intention source location data where the preset fraud intention is located, and update the corresponding preset fraud intention according to the first recognition difference data to obtain the corresponding first recognition fraud intention. After the fraud intention identification, each preset fraud intention on each first example discrimination vector can obtain a corresponding first identified fraud intention. The big data system 100 may calculate intention loss data between the respective preset fraud intents and the example fraud intents, determine at least one preset fraud intention closest to the example fraud intention from the respective preset fraud intents as the example fraud intention according to the intention loss data, and regard a first identified fraud intention corresponding to the reference preset fraud intention as a second identified fraud intention.

In a separately implementable embodiment, there may be multiple example fraud intents in the example fraud event big data, i.e., multiple target fraud trigger data is included in the example fraud event big data. Then, the big data system 100 may determine, from the preset fraud intentions, at least one preset fraud intention respectively closest to the intention source data area of the example fraud intentions as the corresponding example fraud intention, so as to obtain at least one example fraud intention respectively corresponding to the example fraud intentions.

Step S140, a second example discrimination vector corresponding to the first example discrimination vector is obtained according to the first example discrimination vector, and the first example discrimination vector and the corresponding second example discrimination vector are subjected to vector weight fusion to obtain a corresponding third example discrimination vector.

In an independently implementable embodiment, the big data system 100 may perform feature derivation and association processing on the first example discrimination vector to obtain a second example discrimination vector corresponding to the first example discrimination vector. The first example discrimination vector and the corresponding second example discrimination vector are discrimination vectors of the same fraud assessment category. The big data system 100 performs vector weight fusion on the first example discrimination vector and the corresponding second example discrimination vector to obtain a third example discrimination vector corresponding to the first example discrimination vector.

In an embodiment that can be implemented independently, the big data system 100 may obtain, through feature derivation and association processing, second example discrimination vectors corresponding to the respective first example discrimination vectors, and then perform vector weight fusion on the respective first example discrimination vectors and the corresponding second example discrimination vectors to obtain third example discrimination vectors corresponding to the respective first example discrimination vectors. In order to reduce the calculation amount, the big data system 100 may also select a part of the first example discrimination vectors from each of the first example discrimination vectors to calculate corresponding second example discrimination vectors, and perform vector weight fusion on the first example discrimination vectors with the second example discrimination vectors and the corresponding second example discrimination vectors to obtain corresponding third example discrimination vectors.

In a separately implementable embodiment, the first example discrimination vector of the plurality of fraud assessment categories is a first example discrimination vector arranged by an assessment level of the fraud assessment category. Obtaining a second example discrimination vector corresponding to the first example discrimination vector according to the first example discrimination vector, including: and deriving the fraud assessment category of the current first example discrimination vector into the fraud assessment category of the derived assessment level corresponding to the current first example discrimination vector, and taking the derived current first example discrimination vector as a second example discrimination vector corresponding to the same fraud assessment category of the first example discrimination vector.

In an independently implementable embodiment, the big data system 100 may derive the fraud assessment category of the current first example discrimination vector as a fraud assessment category of a derived assessment level corresponding to the current first example discrimination vector, and then use the derived current first example discrimination vector as a second example discrimination vector that matches the first example discrimination vector associated with the fraud assessment category of the current first example discrimination vector. Subsequently, when the first example discrimination vector and the corresponding second example discrimination vector are subjected to vector weight fusion, feature transfer can be performed between different first example discrimination vectors, shallow features are transferred to deep layers, and deep features are transferred to shallow layers, so that accuracy of fraud intention identification is improved.

For example, the big data of the example fraud event passes through six discrimination vector extraction layers to obtain six first example discrimination vectors, and fraud assessment categories of the first example discrimination vectors are sequentially decreased progressively. The big data system 100 may extend the fraud assessment category of the first example discrimination vector corresponding to the latter fraud assessment category to the current fraud assessment category as the second example discrimination vector that matches the first example discrimination vector corresponding to the current fraud assessment category. For example, a first example discrimination vector corresponding to the second fraud assessment category is expanded into the first fraud assessment category, and the expanded first example discrimination vector is taken as the second example discrimination vector matched with the first example discrimination vector corresponding to the first fraud assessment category. And expanding the first example discrimination vector corresponding to the third fraud assessment category into a second fraud assessment category, and taking the expanded first example discrimination vector as a second example discrimination vector matched with the first example discrimination vector corresponding to the second fraud assessment category. Wherein the first example discrimination vector corresponding to the sixth fraud assessment category may be directly used as the second example discrimination vector to which the first example discrimination vector corresponding to the sixth fraud assessment category matches. The first example discrimination vector corresponding to the fifth fraud assessment category may also be mapped to a sixth fraud assessment category, and the mapped first example discrimination vector may be used as the second example discrimination vector to which the first example discrimination vector corresponding to the sixth fraud assessment category matches. It will be appreciated that the big data system 100 may also map a fraud assessment category of the first example discrimination vector corresponding to a previous fraud assessment category to a current fraud assessment category as a second example discrimination vector that matches the first example discrimination vector corresponding to the current fraud assessment category.

For example, after determining the second example discrimination vectors corresponding to the first example discrimination vectors, the big data system 100 may perform feature fusion on the first example discrimination vectors and the corresponding second example discrimination vectors to obtain third example discrimination vectors corresponding to the first example discrimination vectors.

Step S150, performing fraud intention identification on the third example discrimination vector according to the second identification fraud intention, and obtaining a third identification fraud intention corresponding to the second identification fraud intention.

Wherein the third identified fraud intention refers to the identified fraud intention adjusted by the second identified fraud intention.

In an independently implemented embodiment, the big data system 100 may use the second recognized fraud intention as a preset fraud intention in a third example discrimination vector, perform fraud intention recognition on the third example discrimination vector, obtain second recognition difference data corresponding to the second recognized fraud intention according to a fraud event big data feature of the source location data of the intention where the second recognized fraud intention is located, and update the second recognized fraud intention according to the second recognition difference data to obtain a corresponding third recognized fraud intention. On the whole, the big data system 100 performs fraud intention recognition on the first example discrimination vector to obtain first recognition difference data corresponding to a preset fraud intention, performs fraud intention recognition on the third example discrimination vector to obtain second recognition difference data corresponding to a second recognition fraud intention, and finally updates the corresponding preset fraud intention according to the second recognition difference data and the corresponding first recognition difference data to obtain a third recognition fraud intention. Namely, first mining identification is carried out to obtain first identification difference data so as to obtain a first identification cheating intention, then secondary mining identification is carried out to obtain second identification difference data, and the corresponding first identification cheating intention is corrected according to the second identification difference data so as to obtain an accurate third identification cheating intention.

And step S160, generating model loss data according to the first recognition fraud intention and the intention loss data of the example fraud intention, the third recognition fraud intention and the intention loss data of the example fraud intention, and configuring data according to model parameters of a second recognition fraud behavior recognition model of the model loss data until a model convergence condition is met to obtain a target fraud behavior recognition model.

In an independently implementable embodiment, after determining the first identified fraud intent and the third identified fraud intent, the big data system 100 may calculate intent loss data for the first identified fraud intent and the example fraud intent, and intent loss data for the third identified fraud intent and the example fraud intent, generate model loss data from the calculated intent loss data, perform back propagation updates from the model loss data, and model parameter configuration data for the second identified fraud identification model until a model convergence condition is satisfied, resulting in the target fraud identification model. The model convergence condition may be that the model loss data is smaller than a preset index, the number of model iterations reaches an iteration threshold, and the like.

In an embodiment that can be implemented independently, in order to further improve the accuracy of the fraud identification model, the big data system 100 may configure data of model parameters of the fraud identification model according to the model loss data and the accuracy loss data until a model convergence condition is satisfied, so as to obtain a target fraud identification model. In addition, the big data system 100 may also configure data according to the model parameter of the second recognition fraud identification model based on the model loss data and the global loss data until a model convergence condition is satisfied, so as to obtain a target fraud identification model. Of course, the big data system 100 may also configure data according to the model parameter of the second recognition fraudulent behavior recognition model, which is composed of the model loss data, the accuracy loss data, and the global loss data, until the model convergence condition is satisfied, so as to obtain the target fraudulent behavior recognition model. The accuracy loss data and the global difference refer to an information fraud identification method based on big data and artificial intelligence, which can be referred to in the following embodiments, in the global computing process.

In the above-mentioned big data and artificial intelligence based information fraud identification method, fraud intention identification is performed on each first example discrimination vector, therefore, the first identification fraud intention is obtained after the preset fraud intention is preliminarily updated, the third example discrimination vector with better fraud identification reference value can be obtained by fusing different discrimination vectors, and performing fraud intention recognition on the third example discrimination vector according to the second recognition fraud intention, and updating the second recognition fraud intention for the second time to obtain a third recognition fraud intention so that the third recognition fraud intention is closer to the actual fraud intention, further, model parameter configuration data is optimized according to the two intention updating results and the actual fraud intention calculation model loss data, a fraud identification model with higher fraud identification performance can be obtained, therefore, the fraud intention recognition is carried out according to the target fraud behavior recognition model, so that the reference value of fraud reference content push can be improved.

In an embodiment, the independently implemented method includes performing fraud intention recognition on each first example discrimination vector to obtain first recognition fraud intentions corresponding to each preset fraud intention, and determining a second recognition fraud intention from each first recognition fraud intention according to intention loss data between each preset fraud intention and an example fraud intention, including:

respectively carrying out fraud intention recognition on each first example discrimination vector to obtain a first recognition difference data sequence corresponding to each first example discrimination vector; the first identification difference data sequence comprises first identification difference data corresponding to each preset cheating intention on the first example discrimination vector; obtaining a corresponding first recognition fraud intention according to a preset fraud intention and corresponding first recognition difference data; in the current first example discrimination vector, according to the intention matching degree between each preset cheating intention and the example cheating intention, determining a reference preset cheating intention from each preset cheating intention, and taking a first recognition cheating intention corresponding to the reference preset cheating intention as a candidate recognition cheating intention corresponding to the current first example discrimination vector; and obtaining second identification fraud intentions according to the candidate identification fraud intentions respectively corresponding to the first example discrimination vectors.

In an embodiment that can be implemented independently, the big data system 100 can perform fraud intention identification on each first example discrimination vector respectively, so as to obtain a first identification difference data sequence corresponding to each first example discrimination vector respectively, where the first identification difference data sequence includes first identification difference data corresponding to each preset fraud intention on the first example discrimination vector respectively. The first identification difference data corresponding to the preset fraud intention is obtained by carrying out data analysis on the big data characteristics of the fraud event at the position of the preset fraud intention. Then, the big data system 100 can update the corresponding preset fraud intention according to the first identification difference data to obtain the first identification fraud intention.

After obtaining the first recognized fraud intentions corresponding to the preset fraud intentions in the first example discrimination vectors, the big data system 100 may search the preset fraud intention closest to the example fraud intention from the first example discrimination vectors as an example fraud intention, and obtain example fraud intentions corresponding to the first example discrimination vectors. The example fraud intention may specifically be determined by calculating intention matching degrees between each preset fraud intention and the example fraud intention in the current first example discrimination vector, and determining a reference preset fraud intention from each preset fraud intention according to the intention matching degrees. Specifically, the preset fraud intention with the highest intention matching degree may be used as the example fraud intention, or the preset fraud intentions with the highest intention matching degree may be processed in a descending order, and a plurality of preset fraud intentions with the intention matching degrees ranked at the top are used as the example fraud intentions. Then, the big data system 100 may take the first identified fraud intention corresponding to the reference preset fraud intention in the first example discrimination vector as the candidate identified fraud intention corresponding to the first example discrimination vector. Finally, the big data system 100 obtains a second recognition fraud intention according to the candidate recognition fraud intentions respectively corresponding to the first example discrimination vectors. That is, the big data system 100 may determine the second recognized fraud intention from the first recognized fraud intentions corresponding to the respective reference preset fraud intentions. Specifically, the candidate recognition fraud intention which is most matched with the example fraud intention may be used as the second recognition fraud intention, a plurality of candidate recognition fraud intentions which are more matched with the example fraud intention may be used as the second recognition fraud intention, and each of the candidate recognition fraud intentions may be used as the second recognition fraud intention.

In an embodiment, the determining, according to the intention matching degree between each preset fraud intention and the example fraud intention, a reference preset fraud intention from each preset fraud intention includes:

and taking the preset fraud intention corresponding to the maximum intention matching degree as an example fraud intention.

In an independently implementable embodiment, in the current first example discrimination vector, the big data system 100 may take a preset fraud intention corresponding to the maximum intention matching degree as an example fraud intention. That is, the big data system 100 may cull the preset fraud intention that is obviously not mapped on the target fraud trigger data, retain the preset fraud intention that is most accurately currently mapped, and take the preset fraud intention that is most accurately currently mapped as an example fraud intention. It will be appreciated that in calculating the intent match between the preset fraud intent and the example fraud intent, the preset fraud intent and the example fraud intent need to be directionally coupled to the same fraud assessment category for comparison.

In an embodiment, the obtaining the second recognition fraud intention according to the candidate recognition fraud intention corresponding to each first example discrimination vector includes:

among the respective candidate recognition fraudulent intentions, the candidate recognition fraudulent intention which matches the example fraudulent intention most closely is taken as the second recognition fraudulent intention.

In an embodiment that can be implemented independently, after obtaining the candidate recognition fraud intentions corresponding to the respective first example discrimination vectors, the big data system 100 may calculate the intention matching degrees between the respective candidate recognition fraud intentions and the example fraud intentions, and select the candidate recognition fraud intention with the greatest matching degree with the example fraud intention as the second recognition fraud intention. That is, the big data system 100 further performs a preferred process to further select an optimal candidate recognition fraud intention as the second recognition fraud intention from candidate recognition fraud intentions corresponding to a plurality of preset fraud intentions with more accurate current mappings. It will be appreciated that in calculating the intent match of the candidate recognition fraud intent and the example fraud intent, there is a need to compare the candidate recognition fraud intent and the example fraud intent directionally coupled to the same fraud assessment category.

In this embodiment, the fraud intention recognition is performed on each first example discrimination vector to obtain a first recognition fraud intention corresponding to each preset fraud intention on each first example discrimination vector, the first recognition fraud intention is subjected to first screening, candidate recognition fraud intents corresponding to each first example discrimination vector are screened out from the first recognition fraud intents, the second screening is performed on each candidate recognition fraud intention, and a second recognition fraud intention is screened out from the second recognition fraud intents. In this way, the fraud intention closest to the example fraud intention can be accurately screened out from the fraud intention recognition results of the first example discrimination vectors of the respective fraud assessment categories through two screening.

In an embodiment, the method for recognizing a fraud intention of a third example discrimination vector according to a second recognized fraud intention to obtain a third recognized fraud intention corresponding to the second recognized fraud intention includes:

determining fraud directed graph information among the first example discrimination vectors according to the evaluation level of the fraud evaluation category of the first example discrimination vectors; connecting the second identification fraud intention direction to a third example discrimination vector corresponding to the first example discrimination vector according to the fraud directed graph information; and carrying out fraud intention identification on the directionally connected third example discrimination vector to obtain a third identification fraud intention corresponding to the second identification fraud intention.

In an independently implementable embodiment, the big data system 100 requires a synchronous directional connection of the second recognized fraud intention to each third example discrimination vector when performing fraud intention recognition on the third example discrimination vector, because the second recognized fraud intention is the first recognized fraud intention on a particular first example discrimination vector, and the fraud assessment categories of the first example discrimination vector and the third example discrimination vector do not necessarily coincide. The big data system 100 may first determine fraud directed graph information between the respective first example discrimination vectors based on the evaluation level of the fraud evaluation category of the respective first example discrimination vectors. The fraud directed graph information is used for representing the corresponding relation of each characteristic discrimination vector unit of the same original discrimination vector unit on the big data of the example fraud event among the first example discrimination vectors. For example, the fraud assessment category of the first example discrimination vector 1 is smaller than that of the first example discrimination vector 2, one feature discrimination vector unit on the first example discrimination vector 1 corresponds to one vector segment on the first example discrimination vector 2, the feature discrimination vector unit and the vector segment both represent the fraud event big data feature corresponding to the same area on the example fraud event big data, and the feature discrimination vector unit and each feature discrimination vector unit on the vector segment have a corresponding relationship. Then, the big data system 100 directionally connects the second identification fraud intention to the third example discrimination vector corresponding to the first example discrimination vector according to the fraud directed graph information, and then performs fraud intention identification on the directionally connected third example discrimination vector to obtain a third identification fraud intention corresponding to the second identification fraud intention. In mapping the second identified fraud intent, the big data system 100 needs to first determine a mapping segment of the second identified fraud intent on the third example discrimination vector, and then generate the second identified fraud intent of the fixed fraud assessment category on the mapping segment. That is, the fraud assessment categories on the respective third example discrimination vectors identifying the fraud intention are the same, and the data locations are different. It will be appreciated that the respective third example discrimination vectors differ in fraud assessment category, and that similarly too small a range of second identified fraudulent intentions, even if the mapped segments correspond to the same original discrimination vector unit, are vector segments of different sizes on the corresponding example fraud event big data. Therefore, the fraud intention recognition is carried out on the directionally connected third example discrimination vector, the loss data of the second recognition fraud intention relative to the target fraud trigger data is further analyzed, and the target fraud trigger data can be determined more accurately.

In an embodiment that can be implemented independently, performing fraud intention recognition on the directionally-connected third example discrimination vector to obtain a third recognized fraud intention corresponding to the second recognized fraud intention, the method includes:

carrying out fraud intention identification on the directionally connected third example discrimination vector to obtain second identification difference data corresponding to a second identification fraud intention; and obtaining a corresponding third recognition fraud intention according to the second recognition fraud intention and the corresponding second recognition difference data.

In an embodiment that can be implemented independently, the big data system 100 can perform fraud intention identification on the directionally-connected third example discrimination vector, obtain second identification difference data corresponding to the second identification fraud intention according to a fraud event big data feature of a position of the second identification fraud intention on the third example discrimination vector, and update the second identification fraud intention according to the second identification difference data to obtain a corresponding third identification fraud intention.

In this embodiment, the second identification fraud intention is directionally connected to each third example discrimination vector, fraud intention identification is performed on the directionally connected third example discrimination vector, secondary mining identification can be performed by integrating each fraud evaluation category on the basis of primary mining identification, and the fraud intention identified by primary mining identification is corrected according to the secondary mining identification result, so that a more accurate identification fraud intention is obtained.

In an embodiment, the independently implemented method for obtaining a target fraud recognition model according to model parameter configuration data of a second recognition fraud recognition model of model loss data until a model convergence condition is satisfied includes:

in step S210, the reference recognition accuracy corresponding to the preset fraud intention and the second recognized fraud intention, the intention matching degree of which is greater than the preset matching degree, of the example fraud intention is determined as the first recognition accuracy, and the reference recognition accuracy corresponding to the preset fraud intention and the second recognized fraud intention, the intention matching degree of which is less than or equal to the preset matching degree, of the example fraud intention is determined as the second recognition accuracy.

Wherein the recognition accuracy is a category of the intent tag attribute with which the fraud intent is determined. When the identification accuracy corresponding to the fraud intention is the first identification accuracy, the target fraud trigger data is determined to be bound by the fraud intention association. When the recognition accuracy corresponding to the fraud intention is the second recognition accuracy, it is determined that the fraud intention association is bound to not be the target fraud trigger data. The first recognition accuracy and the second recognition accuracy may be set as needed, for example, the first recognition accuracy is determined to be 1 and the second recognition accuracy is determined to be 0. The preset matching degree may also be set as needed, for example, the preset matching degree is set to 0.5.

In an embodiment, in order to further improve the identification effectiveness of the target fraud trigger data, in addition to training the network to correct the preset fraud intention to obtain the identified fraud intention, the network may be further trained to output the category of the identified fraud intention, so that the location data and the category of the intention source for comprehensively identifying the fraud intention can more accurately determine the target fraud trigger data. The big data system 100 may classify the preset fraud intention on the first example discrimination vector, use the preset fraud intention with the matching degree with the example fraud intention larger than the preset matching degree as a forward sample, use the preset fraud intention with the matching degree with the example fraud intention smaller than or equal to the preset matching degree as a backward sample, and similarly, the big data system 100 may also classify the second identified fraud intention on the third example discrimination vector, use the second identified fraud intention with the matching degree with the example fraud intention larger than the preset matching degree as a forward sample, and use the second identified fraud intention with the matching degree with the example fraud intention smaller than or equal to the preset matching degree as a backward sample. The big data system 100 may determine the identification accuracy corresponding to the forward samples as a first identification accuracy and the identification accuracy corresponding to the backward samples as a second identification accuracy. In this way, the big data system 100 can perform supervised training on the model according to the forward samples and the backward samples, so that the network can accurately identify the identification accuracy corresponding to the fraud intention.

It can be understood that, when the intention matching degree between the preset fraud intention and the example fraud intention and the intention matching degree between the second identified fraud intention and the example fraud intention are calculated, the preset fraud intention and the example fraud intention are connected to the same fraud assessment category and then calculated, and the second identified fraud intention and the example fraud intention are connected to the same fraud assessment category and then calculated.

Step S220, generating accuracy loss data according to the fraud accuracy and the reference identification accuracy corresponding to the preset fraud intention and the fraud accuracy and the reference identification accuracy corresponding to the second identification fraud intention; the preset fraud accuracy corresponding to the fraud intention is obtained by carrying out fraud intention recognition on the first example discrimination vector, and the fraud accuracy corresponding to the second fraud intention recognition is obtained by carrying out fraud intention recognition on the third example discrimination vector.

And step S230, second identifying the model parameter configuration data of the fraud behavior identification model according to the model loss data and the accuracy loss data until the model convergence condition is met, and obtaining the target fraud behavior identification model.

In an embodiment that can be implemented independently, when the fraud intention recognition is performed on the first example discrimination vector, the big data system 100 may obtain not only the first recognition difference data corresponding to the preset fraud intention but also the fraud accuracy corresponding to the preset fraud intention, and similarly, when the fraud intention recognition is performed on the third example discrimination vector, the big data system 100 may obtain not only the first recognition difference data corresponding to the second recognition fraud intention but also the fraud accuracy corresponding to the second recognition fraud intention. Therefore, the big data system 100 may calculate accuracy loss data according to the recognition accuracy difference between the fraud accuracy and the reference recognition accuracy corresponding to the preset fraud intention and the recognition accuracy difference between the fraud accuracy and the reference recognition accuracy corresponding to the second recognition fraud intention, perform back propagation updating by combining the model loss data and the accuracy loss data, and obtain the target fraud recognition model by configuring the model parameter of the second recognition fraud recognition model until the model convergence condition is satisfied. In this way, when the target fraud behavior recognition model is applied, the target fraud trigger data can be accurately detected by combining the intention source location data and the category of the fraud intention.

In the embodiment, the model is trained according to the model loss data and the accuracy loss data, and the target fraud identification model can simultaneously identify the intention source location data and the category of the fraud intention, so that the target fraud trigger data is accurately located according to the intention source location data and the category of the fraud intention.

In an embodiment, the second identifying fraud identification model based on the model loss data and the accuracy loss data until a model convergence condition is satisfied to obtain a target fraud identification model includes:

step S310, inputting big data of example fraud events into a global fraud intention analysis model after training is finished, and obtaining first global discrimination vectors corresponding to all first example discrimination vectors and second global discrimination vectors corresponding to all third example discrimination vectors; the model parameter configuration data quantity of the global fraud intention analysis model is larger than that of the fraud behavior recognition model, and a corresponding relation exists between the global fraud intention analysis model and a discrimination vector extraction layer of the fraud behavior recognition model;

step S320, generating global loss data according to a first cross degree between the first example discrimination vector and the corresponding first global discrimination vector and a second cross degree between the third example discrimination vector and the corresponding second global discrimination vector;

and step S330, configuring data according to the model parameters of the second identification fraudulent behavior identification model of the model loss data, the accuracy loss data and the global loss data until the model convergence condition is met, and obtaining the target fraudulent behavior identification model.

The global fraud intention analysis model is a parent fraud intention analysis model, and the fraud behavior identification model is a child fraud intention analysis model. The model parameter configuration data amount of the parent fraud intention analysis model is larger than that of the child fraud intention analysis model, and the model structure composition of the parent fraud intention analysis model and the model structure composition of the child fraud intention analysis model may be the same or different. The parameter quantity of the discrimination vector obtained by the parent fraud intention analysis model through the discrimination vector extraction of the input fraud event big data is larger than that of the discrimination vector obtained by the child fraud intention analysis model through the discrimination vector extraction of the input fraud event big data, and the method mainly comprises the step that the kernel function quantity of the discrimination vector obtained by the parent fraud intention analysis model is larger. The parent fraud intention analysis model and the child fraud intention analysis model both comprise a discrimination vector extraction layer, the discrimination vector extraction layer of the parent fraud intention analysis model and the discrimination vector extraction layer of the child fraud intention analysis model can have the same number of layers, and certainly, the discrimination vector extraction layer of the parent fraud intention analysis model can also have more layers than the discrimination vector extraction layer of the child fraud intention analysis model.

In an independently implementable embodiment, the fraud identification model has high requirements on cloud computing performance during application, and the traditional fraud identification model is often large in calculation amount, so that the learning and prediction efficiency is low, and therefore, in order to improve the fraud identification modelRecognition of speedAnd the fraud behavior recognition model can be further processed to obtain a lightweight fraud behavior recognition model. The big data system 100 can obtain the global fraud intention analysis model after training, and input the same example fraud event big data into the global fraud intention analysis model and the fraud behavior recognition model respectively. The big data system 100 performs data processing on the big data of the example fraud event through a fraud behavior recognition model to obtain a first example discrimination vector and a third example discrimination vector, and performs data processing on the big data of the example fraud event through a global fraud intention analysis model to obtain a first global discrimination vector and a second global discrimination vector. The sub-fraud intention analysis model extracts the discrimination vector of the example fraud event big data through the discrimination vector extraction layer to obtain an initial discrimination vector, and the first example discrimination vector and the first global discrimination vector output by the discrimination vector extraction layer with the corresponding relation also have the corresponding relation. The model is used for deriving and fusing the initial discrimination vector to obtain a corresponding target discrimination vector, and a third example discrimination vector and a second global discrimination vector obtained by fusing the first example discrimination vector and the first global discrimination vector which have corresponding relations also have corresponding relationsIs described.

Because the global fraud intention analysis model has strong feature expression capability, the first example discrimination vector can be learned to the first global discrimination vector, the first example discrimination vector can be approached to the first global discrimination vector, the third example discrimination vector can be learned to the second global discrimination vector, and the third example discrimination vector can be approached to the second global discrimination vector. The big data system 100 may calculate a first cross degree between a first example discrimination vector and a corresponding first global discrimination vector, calculate a second cross degree between a third example discrimination vector and a corresponding second global discrimination vector, generate global loss data according to the first cross degree and the second cross degree, perform back propagation updating jointly with the model loss data, the accuracy loss data, and the global loss data, and configure data for model parameters of a second recognition fraudulent behavior recognition model until a model convergence condition is satisfied, so as to obtain a target fraudulent behavior recognition model.

In an independently implementable embodiment, the global fraud intention analysis model is pre-trained, and the training process is the same as the fraud recognition model and is based on model loss data or model loss data and accuracy loss data.

In an embodiment, the global fraud intention analysis model and the fraud behavior recognition model have the same number of layers of the discrimination vector extraction layers, and there is a correspondence before the discrimination vector extraction layers with the same sequence. For example, the global fraud intention analysis model and the fraud recognition model each include three discrimination vector extraction layers, a first discrimination vector extraction layer of the global fraud intention analysis model corresponds to a first discrimination vector extraction layer of the fraud recognition model, a second discrimination vector extraction layer of the global fraud intention analysis model corresponds to a second discrimination vector extraction layer of the fraud recognition model, and a third discrimination vector extraction layer of the global fraud intention analysis model corresponds to a third discrimination vector extraction layer of the fraud recognition model.

In an independently implementable embodiment, the global fraud intention analysis model has more discriminant vector extraction layers than the fraud identification model. The first discriminant vector extraction layer of the fraud identification model corresponds to the first discriminant vector extraction layer of the global fraud intention analysis model, and the last discriminant vector extraction layer of the fraud identification model corresponds to the last discriminant vector extraction layer of the global fraud intention analysis model, so that the capability that the shallow and deep features extracted from the big data of the fraud by the fraud identification model are close to the global fraud intention analysis model is guaranteed. And the other discriminant vector extraction layers of the fraud behavior recognition model respectively correspond to one discriminant vector extraction layer of the global fraud intention analysis model, but it should be noted that the discriminant vector extraction layers cannot be in cross correspondence. For example, the global fraud intention analysis model comprises six discrimination vector extraction layers, the fraud recognition model comprises four discrimination vector extraction layers, the first discrimination vector extraction layer of the fraud recognition model corresponds to the first discrimination vector extraction layer in the global fraud intention analysis model, the fourth discrimination vector extraction layer of the fraud recognition model corresponds to the sixth discrimination vector extraction layer in the global fraud intention analysis model, when the second discrimination vector extraction layer of the fraud recognition model corresponds to the third discrimination vector extraction layer of the global fraud intention analysis model, the third discrimination vector extraction layer of the fraud recognition model cannot correspond to the second discrimination vector extraction layer in the global fraud intention analysis model and cannot form a cross-correspondence, and at this time, and the third discriminant vector extraction layer of the fraud behavior recognition model can only correspond to the fourth discriminant vector extraction layer or the fifth discriminant vector extraction layer in the global fraud intention analysis model.

In the embodiment, knowledge distillation is further performed on the fraud recognition model during AI training, so that a lightweight fraud recognition model can be obtained, and therefore, when the target fraud recognition model is applied, the recognition accuracy can be guaranteed, and the recognition speed can be guaranteed to be high.

In an independently implementable embodiment, generating global impairment data from a first degree of intersection between a first example discrimination vector and a corresponding first global discrimination vector, and a second degree of intersection between a third example discrimination vector and a corresponding second global discrimination vector, comprises:

carrying out fraud assessment category mapping on each first example discrimination vector so that the fraud assessment categories of the first example discrimination vector mapped by each fraud assessment category and the corresponding first global discrimination vector are the same; calculating first discrimination vector intersection degrees between the first example discrimination vectors and the corresponding first global discrimination vectors after fraud assessment category mapping, and obtaining first intersection degrees according to the first discrimination vector intersection degrees; carrying out fraud assessment type mapping on each third example discrimination vector so as to enable the fraud assessment type of each fraud assessment type mapped third example discrimination vector to be the same as that of the corresponding second global discrimination vector; calculating second discrimination vector intersection degrees between the third example discrimination vector after fraud assessment category mapping and the corresponding second global discrimination vector, and obtaining second intersection degrees according to the second discrimination vector intersection degrees; and generating global loss data according to the first and second intersection degrees.

In an embodiment, since the amount of model parameter configuration data of the global fraud intention analysis model is greater than the amount of model parameter configuration data of the target fraud trigger data model, the fraud assessment category of the first global discrimination vector is greater than the corresponding first example discrimination vector, and the fraud assessment category of the second global discrimination vector is greater than the corresponding third example discrimination vector. Therefore, when calculating the cross degree, the discrimination vectors with the corresponding relationship need to be converted into the same fraud assessment category, and the component similarity between the discrimination vectors is measured by the vector similarity between the discrimination vectors with the same fraud assessment category. The big data system 100 may perform fraud assessment category mapping on each first example discrimination vector, so that fraud assessment categories of the first example discrimination vector mapped by each fraud assessment category are the same as fraud assessment categories of the corresponding first global discrimination vector, then calculate a first discrimination vector cross degree between the first example discrimination vector mapped by the fraud assessment category and the corresponding first global discrimination vector, and obtain the first cross degree between the first example discrimination vector and the corresponding first global discrimination vector according to the first discrimination vector cross degree. For example, the first discrimination vector cross degree is directly used as the first cross degree. Similarly, the big data system 100 may perform fraud assessment category mapping on each third example discrimination vector, so that the fraud assessment categories of the third example discrimination vector after the fraud assessment category mapping are the same as the fraud assessment categories of the corresponding second global discrimination vector, then calculate a second discrimination vector cross degree between the third example discrimination vector after the fraud assessment category mapping and the corresponding second global discrimination vector, and obtain a second cross degree between the third example discrimination vector and the corresponding second global discrimination vector according to the second discrimination vector cross degree. Finally, the big data system 100 generates global loss data from the first and second degrees of intersection, e.g., taking the sum of the first and second degrees of intersection as the global loss data.

The intersection calculation process is described by taking the first intersection as an example. Assume that the fraud identification model and the global fraud intention analysis model each include six discrimination vector extraction layers, and there is a correspondence before the discrimination vector extraction layers that are ordered the same. After the big data of the same example fraud event are respectively input into a fraud behavior recognition model and a global fraud intention analysis model, six first example discrimination vectors and six first global discrimination vectors can be obtained. The first example discrimination vector and the first global discrimination vector output by the discrimination vector extraction layer having correspondence relationship correspond to each other. The big data system 100 may perform fraud assessment category mapping on the first example discrimination vector, so that the fraud assessment categories of the first example discrimination vector and the corresponding first global discrimination vector are the same, further calculate vector similarities between the first example discrimination vector and the corresponding first global discrimination vector, which are the same in fraud assessment categories, to obtain six vector similarities, and then obtain a first cross degree according to the six vector similarities.

In the embodiment, after the two discrimination vectors are converted into the same fraud assessment category, the cross degree between the discrimination vectors is obtained according to the vector similarity between the discrimination vectors, and the component similarity between the discrimination vectors can be accurately measured.

In an embodiment that can be implemented independently, referring to fig. 2, there is provided a big data and artificial intelligence based information fraud identification method, where the big data and artificial intelligence based information fraud identification method includes the following steps:

step S510, obtaining the fraud event big data to be identified of the digital service device.

Step S520, extracting a discrimination vector of big data of a fraud event to be recognized according to the fraud behavior recognition model meeting the model convergence requirement to obtain a first fraud discrimination vector of a plurality of fraud assessment categories; each first fraud discrimination vector comprises a corresponding preset fraud intention.

Step S530, fraud intention recognition is carried out on each first fraud judgment vector, and candidate fraud intentions corresponding to the target fraud trigger data on the big data of the fraud event to be recognized are obtained.

And step S540, obtaining a second fraud discrimination vector corresponding to the first fraud discrimination vector according to the first fraud discrimination vector, and performing vector weight fusion on the first fraud discrimination vector and the corresponding second fraud discrimination vector to obtain a corresponding third fraud discrimination vector.

And step S550, carrying out fraud intention identification on the third fraud judgment vector according to the candidate fraud intention, obtaining a target fraud intention corresponding to the target fraud trigger data on the big data of the fraud event to be identified, and carrying out fraud reference content push on the digital service equipment according to the target fraud intention.

In an independently implementable embodiment, the big data system 100 may obtain the big data of the fraud event to be identified locally, or from other terminals or cloud service platforms, for example, in a credit social entity network, the credit fraud data of the credit service flow may be obtained from the credit service platform, the data sequence in the credit fraud data is used as the big data of the fraud event to be identified, and the big data of the fraud event to be identified is classified with credit intention. The big data system 100 may extract a discrimination vector from big data of a fraud event to be identified, obtain first fraud discrimination vectors of a plurality of fraud assessment categories, and mark and add at least one preset fraud intention on each first fraud discrimination vector. The big data system 100 identifies fraud intention to each first fraud discrimination vector to obtain candidate analytic fraud intents corresponding to each preset fraud intention, and determines candidate fraud intents corresponding to target fraud trigger data on the big data of the fraud event to be identified from each candidate analytic fraud intention. The big data system 100 may specifically select, as the candidate fraud intentions, candidate analytic fraud intentions with identification accuracy greater than a preset matching degree from the candidate analytic fraud intentions, where the identification accuracy is also obtained by the big data system 100 performing fraud intention identification on each first fraud discrimination vector, and the candidate analytic fraud intentions corresponding to the preset fraud intentions are obtained by performing fraud intention identification on the first fraud discrimination vector to obtain initial identification loss data corresponding to each preset fraud intention, and are obtained according to the preset fraud intention and the corresponding initial identification loss data. Next, the big data system 100 performs feature derivation and association processing on the first example discrimination vector to obtain a second example discrimination vector corresponding to the first example discrimination vector, and performs vector weight fusion on the first example discrimination vector and the corresponding second example discrimination vector to obtain a corresponding third example discrimination vector, where the first example discrimination vector and the corresponding second example discrimination vector are discrimination vectors with the same fraud assessment category. The big data system 100 may use the candidate fraud intention as a preset fraud intention on a third fraud discrimination vector, perform fraud intention identification on the third fraud discrimination vector to obtain intermediate analysis fraud intentions corresponding to the candidate fraud intention, and determine a target fraud intention corresponding to target fraud trigger data on the fraud event big data to be identified from each intermediate analysis fraud intention. The big data system 100 may specifically select, from the intermediate analyzed fraudulent intentions, an intermediate analyzed fraudulent intention whose identification accuracy is greater than a preset matching degree as the target fraudulent intention, where the identification accuracy is also obtained by the big data system 100 performing fraudulent intention identification on the third fraudulent judgment vector, and the intermediate analyzed fraudulent intention corresponding to the candidate fraudulent intention is obtained by performing fraudulent intention identification on the third fraudulent judgment vector to obtain target identification loss data corresponding to the candidate fraudulent intention, and is obtained according to the candidate fraudulent intention and the corresponding target identification loss data.

It can be understood that, the specific processes of extracting the discrimination vector for the big data of the fraud event, identifying the fraud intention for the discrimination vector, and generating the third fraud discrimination vector may refer to the big data and artificial intelligence based information fraud identification method of each related embodiment of the big data and artificial intelligence based information fraud identification method, and are not described herein again. The big data and artificial intelligence based information fraud recognition method of each relevant embodiment of the big data and artificial intelligence based information fraud recognition method can be realized not only by a model, but also by designing a corresponding algorithm or formula.

In an independently implementable embodiment, the big data system 100 can perform big data and artificial intelligence based information fraud identification on big data of fraud events to be identified by means of an AI model. The big data system 100 may input the big data of the fraud event to be recognized into the target fraud behavior recognition model, and the model outputs a target fraud intention corresponding to the target fraud trigger data on the big data of the fraud event to be recognized. The training process of the fraud behavior recognition model may refer to the big data and artificial intelligence based information fraud recognition methods of the related embodiments of the big data and artificial intelligence based information fraud recognition methods, and details are not repeated here.

The information fraud identification method based on big data and artificial intelligence carries out fraud intention identification on each first fraud discrimination vector, therefore, after the preset fraud intention is preliminarily updated, candidate fraud intentions are obtained, a third example discrimination vector with better fraud identification reference value can be obtained by fusing different discrimination vectors, and carrying out fraud intention identification on the third example discrimination vector according to the candidate fraud intention, thus obtaining the target fraud intention after secondary updating of the candidate fraud intention, wherein the target fraud intention can more accurately reflect the actual fraud intention of the target fraud trigger data, therefore, the identification effectiveness of the target fraud trigger data is improved, the reference value of carrying out fraud reference content pushing on the digital service equipment is improved, and the actual fraud intention of a fraud source can be more effectively reflected by the fraud reference content.

In an independently implementable embodiment, with respect to the foregoing step S550, the embodiment of the present disclosure provides an artificial intelligence based fraud reference content pushing method, which can be implemented independently of other embodiments, and specifically can be implemented by the following steps.

Step A601, a fraud track data sequence of the target fraud intention is obtained.

For example, the fraud trajectory data sequence comprises a series of fraud trajectory data for the target fraud intent and the target fraud reference content data.

Step A602, obtaining a key fraud migration behavior matched with the preset fraud risk service characteristics from a fraud behavior track data sequence to obtain fraud migration data of the key fraud migration behavior; wherein the key fraud migration behavior comprises direct fraud migration behavior and indirect fraud migration behavior.

For example, the key fraud migration behavior is used for recording fraud migration information of different fraud trace data, such as a time sequence characteristic, a data source characteristic and the like of fraud migration.

Step A603, extracting target fraud unit data of the associated fraud reference content for determining the target fraud intention from the fraud migration data.

Step A604, obtaining the associated fraud reference content of the target fraud intention based on the target fraud unit data to obtain the content tracing data of the associated fraud reference content of the target fraud intention; and determining target fraud reference content data from the fraud track data sequence through the content tracing data, and pushing fraud reference content to the target fraud reference content data.

For example, the content traceability data is used to screen the fraud trace data sequence for different content data.

Therefore, the key fraud migration behavior matched with the preset fraud risk service characteristics can be obtained from the obtained fraud behavior track data sequence to obtain fraud migration data, and the target fraud unit data is extracted from the fraud migration data, so that the target fraud intention can be associated with fraud reference content based on the target fraud unit data to obtain content tracing data. Therefore, target fraud reference content data can be screened from the fraud track data sequence through the content tracing data, and fraud reference content pushing is carried out. Because the associated fraud reference content of the target fraud intention is considered when the target fraud reference content data is determined, the associated target fraud reference content data can be determined from the fraud behavior track data sequence of the target fraud intention on the premise of not influencing the normal fraud push information of the target fraud intention, and the fraud reference content data is pushed.

In an embodiment that can be implemented independently, the target fraud intention includes a direct fraud intention tag and an indirect fraud intention tag, based on which, the obtaining of the key fraud migration behavior matching the preset fraud risk service feature from the fraud trajectory data sequence described in step a602, and obtaining fraud migration data of the key fraud migration behavior may include step a6021 and step a 6022.

Step A6021, a first fraud track data sequence corresponding to the direct fraud intention label and a second fraud track data sequence corresponding to the indirect fraud intention label are obtained from the fraud track data sequence.

Step A6022, respectively obtaining key fraud migration behaviors matched with the preset fraud risk service characteristics from each fraud behavior track data of the first fraud behavior track data sequence and each fraud behavior track data of the second fraud behavior track data sequence, determining that the key fraud migration behaviors matched with the preset fraud risk service characteristics obtained from each fraud behavior track data of the first fraud behavior track data sequence are indirect fraud migration behaviors, and determining that the key fraud migration behaviors matched with the preset fraud risk service characteristics obtained from each fraud behavior track data of the second fraud behavior track data sequence are direct fraud migration behaviors.

In the foregoing, the indirect fraudulent intent tag does not include a directly accessed fraudulent intent tag.

In another embodiment, which can be implemented independently, the fraud migration data includes fraud migration call data, and the fraud migration call data is used for summarizing the fraud migration call situation of the fraud migration data, so as to facilitate subsequent analysis. On this basis, the extracting of the target fraud unit data of the associated fraud reference content for determining the target fraud intention from the fraud migration data described in step a603 may include the following steps a6031 and a 6032.

Step A6031, extracting fraud migration data of direct fraud migration behavior from the fraud migration data.

For example, the fraud migration data includes fraud migration data for direct fraud migration and fraud migration data for indirect fraud migration.

And step A6032, generating a cheating migration calling knowledge network of the direct cheating migration behavior according to the cheating migration calling data of the direct cheating migration behavior. Wherein the target fraud unit data comprises a fraud migration invocation knowledge network for direct fraud migration behavior.

For example, the fraud migration calling knowledge network is used for carrying out feature expression of the knowledge network on fraud migration calling data of direct fraud migration behaviors, so that a large amount of fraud migration calling data can be materialized, and the accuracy of target fraud unit data is improved as much as possible on the premise of not changing the data features of the fraud migration calling data.

In another embodiment that can be implemented independently, the step a602 of obtaining the key fraud migration behavior matching the preset fraud risk service feature from the fraud trajectory data sequence to obtain the fraud migration data of the key fraud migration behavior, and the step a603 of extracting the target fraud unit data of the associated fraud reference content for determining the target fraud intention from the fraud migration data may be implemented in the following two implementations, and of course, in actual implementation, the implementation is not limited to the following two implementations.

A first embodiment.

(11) And performing fraud migration behavior recognition on each fraud behavior track data of the first fraud behavior track data sequence to generate fraud migration data of key fraud migration behaviors, wherein the fraud migration data comprises: and the method is used for distinguishing the target fraud migration flow direction and the fraud migration category of the key fraud migration behavior.

(12) Determining a direct fraud migration behavior in the key fraud migration behaviors according to the fraud migration category, and inputting fraud migration flow direction data corresponding to a target fraud migration flow direction of the direct fraud migration behavior in each fraud behavior track data of the first fraud behavior track data sequence into a fraud migration recognition network to obtain classification measurement value data of whether the direct fraud migration behavior corresponds to a preset migration label, wherein the preset migration label comprises: a live migration tag and/or a static migration tag.

For example, the fraud migration recognition network may be a trained AI neural model whose function may be adapted based on the above, and therefore will not be described further herein.

(13) When the direct cheating migration behavior is identified to correspond to the preset migration tag from the multiple continuous cheating behavior track data in the first cheating behavior track data sequence, recording the dynamic cheating migration behavior of the direct cheating migration behavior, wherein the target cheating unit data comprises the dynamic cheating migration behavior of the direct cheating migration behavior.

A second embodiment.

(21) And performing fraud migration behavior recognition on each fraud behavior track data of the second fraud behavior track data sequence to generate fraud migration data of key fraud migration behaviors, wherein the fraud migration data comprises: and the method is used for distinguishing the target fraud migration flow direction and the fraud migration category of the key fraud migration behavior.

(22) And determining indirect fraudulent migration behaviors in the key fraudulent migration behaviors according to the fraudulent migration categories, and respectively inputting fraudulent migration flow direction data corresponding to the target fraudulent migration flow direction of each indirect fraudulent migration behavior in each fraudulent migration behavior track data of the second fraudulent migration behavior track data sequence into the migration flow direction identification network to obtain the migration flow direction identification information of each indirect fraudulent migration behavior in each fraudulent migration behavior track data.

Similarly, the migration flow direction recognition network may be a pre-trained AI neural model, the function of which can be adaptively adjusted based on the above, and therefore will not be further described herein.

(23) And recording the fraud migration trend item behavior of the indirect fraud migration behavior when the trend floating value of the migration flow direction trend of the indirect fraud migration behavior identified from the associated fraud behavior track data in the second fraud behavior track data sequence is greater than the set trend floating value, wherein the target fraud unit data comprises the fraud migration trend item behavior of the indirect fraud migration behavior.

In the above embodiment, the target fraud migration flow of the indirect fraud migration behavior includes a first target fraud migration flow of an active fraud migration feature and a passive fraud migration feature for distinguishing the indirect fraud migration behavior and a second target fraud migration flow of a passive fraud migration specific manner corresponding to the passive fraud migration feature for distinguishing the indirect fraud migration behavior; wherein the migration flow direction identification network calculates the migration flow direction identification information of the indirect fraudulent migration behavior based on the migration flow direction request data of the first target fraudulent migration flow direction.

In another embodiment, the method may further include the following steps a 21-a 23.

Step a21, a first fraud intention path data blob and a second fraud intention path data blob of the target fraud intention are obtained, wherein the first fraud intention path data in the first fraud intention path data blob and the corresponding second fraud intention path data in the second fraud intention path data blob have different path directions.

Step a22, determining migration invoking data of the corresponding indirect fraudulent migration behavior according to the fraudulent migration routing log of the same indirect fraudulent migration behavior in the first fraudulent intention path data and the second fraudulent intention path data.

Step a23, respectively inputting fraud migration flow direction data corresponding to the target fraud migration flow direction of each indirect fraud migration behavior in each fraud behavior trajectory data of the fraud behavior trajectory data sequence into a migration flow direction identification network, and obtaining migration flow direction identification information of each indirect fraud migration behavior in each fraud behavior trajectory data sequence includes: and inputting fraud migration flow direction data corresponding to the target fraud migration flow direction of each indirect fraud migration behavior in each fraud behavior track data of the fraud behavior track data sequence and migration calling data of the corresponding marked indirect fraud migration behavior into a migration flow direction identification network to obtain actual migration flow direction identification information of each indirect fraud migration behavior in each fraud behavior track data.

In this way, the fraud intention path data of different data dimensions can be analyzed, so that the migration calling data of the indirect fraud migration behavior can be determined, and thus, the actual migration flow direction identification information of each indirect fraud migration behavior in each fraud behavior track data can be determined based on the migration calling data, so that the accuracy of the migration flow direction identification information is ensured, and the target fraud unit data is ensured to meet the actual requirements.

In an embodiment, the obtaining of the relevant fraudulent reference content of the target fraudulent intention based on the target fraudulent unit data in step a604 to obtain the content tracing data of the relevant fraudulent reference content of the target fraudulent intention may include the following steps a60411 to a 60415.

Step A60411, a plurality of traceability intelligence sequences for the target fraud intention are determined based on the target fraud unit data.

For example, the traceability intelligence sequence is used to indicate correlated fraud reference content acquisitions for the targeted fraud intent from different content acquisition dimensions.

Step A60412, for each traceability information sequence which does not meet the preset information condition, processing the traceability information atlas of the traceability information sequence to obtain a first traceability information group which meets an information contact template, and adding the first traceability information group into an information contact library corresponding to an information contact unit corresponding to the traceability information sequence, wherein the information contact template is as follows: and associating the intelligence contact template of the intelligence contact unit corresponding to the tracing intelligence sequence in the obtaining process of the fraud reference content.

For example, the intelligence communication unit may be a pre-configured algorithm model, which may be selected according to actual situations, and is not limited herein.

Step A60413, optimizing a first tracing situation party configured in an information contact library corresponding to the preset social entity network data for mining the social entity network sequence by adopting a pre-generated second tracing situation party, wherein the second tracing situation party is as follows: and processing dynamic traceability information corresponding to the target traceability information sequence to obtain traceability information parties meeting an information contact template for mining the social entity network sequence, wherein the target traceability information sequence is a traceability information sequence meeting a preset information condition, and the information contact library corresponding to the mined social entity network sequence and the information contact library corresponding to any information contact unit have a shared information contact library.

Step A60414, determining a reference tracing situation newspaper group with the information routing cost in the information contact library.

For example, intelligence routing costs are used to indicate the order of the number of content operations referencing the traceback actor, the higher the intelligence routing costs, the earlier the order of using the traceback actor.

Step A60415, the reference tracing situation newspaper group is adopted to obtain the associated fraud reference content of the target fraud intention, and the content tracing data of the associated fraud reference content of the target fraud intention is obtained.

Based on the steps a60411 to a60415, a plurality of tracing intelligence sequences can be determined first, and then the determination of the reference tracing intelligence group with the intelligence routing cost is realized, so that the reference tracing intelligence group can be used to perform associated fraud reference content acquisition on the target fraud intention, which can ensure that the content tracing data is matched with the current social network entity, thereby realizing the subsequent screening of the target fraud reference content data more accurately.

In an independently implementable embodiment, the step a60412 of processing the traceable intelligence map of the traceable intelligence sequence for each traceable intelligence sequence that does not satisfy the preset intelligence condition to obtain a first traceable intelligence group that satisfies the intelligence contact template, and adding the first traceable intelligence group to the intelligence contact library corresponding to the intelligence contact unit corresponding to the traceable intelligence sequence may include the following steps a604121 and a 604122.

Step A604121, for each traceability information sequence which does not meet the preset information condition, clustering traceability information maps of the traceability information sequences according to the information contact record of the information contact unit corresponding to the traceability information sequence in the acquisition flow of the associated fraud reference content to obtain a first traceability information group.

Step A604122, for each tracing intelligence sequence not meeting the preset intelligence condition, adding the first tracing intelligence group corresponding to the tracing intelligence sequence to the intelligence contact node corresponding to the intelligence contact unit corresponding to the tracing intelligence sequence in the intelligence contact library.

In an embodiment, the generation manner of the second tracing situation group described in step a60413 may include the following steps a604131 and a 604132.

Step A604131, obtaining the pre-stored dynamic traceability information corresponding to the preset information condition satisfied by the target traceability information sequence; when there are multiple dynamic provenance intelligence, for each dynamic provenance intelligence.

Step A604132, clustering the dynamic traceability information according to the information contact record of the information contact unit corresponding to the dynamic traceability information in the social entity network sequence to obtain a second traceability information cluster.

It can be understood that, by implementing the step a604131 and the step a604132, the matching degree between the second tracing situation newspaper and the actual social network entity can be ensured, so as to improve the resolution accuracy of the second tracing situation newspaper.

In another embodiment, the step of processing the traceable intelligence map of the traceable intelligence sequence for each traceable intelligence sequence that does not satisfy the predetermined intelligence condition, which is described in step a60412, may include: and when the migration flow direction mode represented by the target fraud unit data is a combined migration flow direction mode, processing the traceability intelligence map of the traceability intelligence sequence aiming at each traceability intelligence sequence which does not meet the preset intelligence condition.

In an embodiment, when the migration flow direction represented by the target fraud unit data is an independent migration flow direction, the method further includes the following steps a 11-a 13.

Step A11, for each tracing intelligence sequence which does not satisfy the preset intelligence condition, clustering the tracing intelligence atlas of the tracing intelligence sequence according to the intelligence contact record of the intelligence contact unit corresponding to the tracing intelligence sequence in the acquisition flow of the associated fraud reference content to obtain a first tracing intelligence group.

Step A12, for each tracing intelligence sequence not meeting the preset intelligence condition, adding the first tracing intelligence group corresponding to the tracing intelligence sequence to the intelligence contact node corresponding to the intelligence contact unit corresponding to the tracing intelligence sequence in the intelligence contact library.

Step A13, determining a reference tracing source situation party with intelligence routing cost in the intelligence contact base, wherein the reference tracing source situation party with intelligence routing cost in the intelligence contact base further comprises: the candidate information contact data which is added to the information contact library corresponding to the information contact unit corresponding to the target tracing information sequence in advance is as follows: and processing dynamic traceability information corresponding to a preset information condition met by the target traceability information sequence to obtain information contact data of an information contact template meeting an information contact unit corresponding to the target traceability information sequence.

It can be understood that by implementing the above steps a 11-a 13, the reference traceability system can determine the reference traceability system of the traceability system in different ways, thereby improving the flexibility of determining the reference traceability system and ensuring that different ways can be selected to determine the reference traceability system under different social network entities.

In an embodiment, the adding of the candidate intelligence contact data in step a13 may include the following steps a 131-a 133.

Step A131, obtaining the pre-stored dynamic traceability information corresponding to the preset information condition satisfied by the target traceability information sequence.

Step A132, clustering the dynamic traceability information according to the information contact records of the information contact units corresponding to the target traceability information sequence to obtain candidate information contact data.

Step A133, add the candidate intelligence contact data to the intelligence contact nodes corresponding to the intelligence contact units corresponding to the target tracing intelligence sequence in the intelligence contact library.

For example, in an independently implementable embodiment, in order to ensure accurate splitting of target fraudulent reference content data, different intelligence routing costs need to be considered, thereby avoiding the generation of influence of target fraudulent reference content data and fraudulent activity trace data. The step a604, determining target fraudulent reference content data from the fraudulent conduct track data sequence through the content tracing data, and performing fraudulent reference content push on the target fraudulent reference content data, may include the content described in step a 60421-step a 60416.

Step A60421, obtaining each first fraud candidate information contact data of a fraud behavior track data sequence and each second fraud candidate information contact data of each first fraud behavior associated data sequence based on the interval data of the information routing cost in the content tracing data; the fraud behavior track data sequence and each first fraud behavior related data sequence are data sequences with different data dimensions; each first fraud candidate intelligence contact data at least comprises corresponding behavior track searching information and behavior track offset information when a fraud behavior track data sequence is generated, and each second fraud candidate intelligence contact data of any first fraud behavior associated data sequence at least comprises corresponding behavior track searching information and behavior track offset information when the first fraud behavior associated data sequence is generated.

Step A60422, for each first fraud candidate information contact data and each first fraud associated data sequence, determining at least one third fraud candidate information contact data matched with the data dimension of the first fraud candidate information contact data in the second fraud candidate information contact data of the first fraud associated data sequence, and judging whether the first fraud candidate information contact data and each third fraud candidate information contact data meet the preset fraud reference content recommendation condition according to the behavior track search information and the behavior track offset information included in the first fraud candidate fraud information contact data and the behavior track search information and the behavior track offset information included in each third fraud candidate information contact data.

Step a60423, if yes, calculating a matching degree between each third fraud candidate information contact data and the first fraud candidate information contact data according to the behavior track search information and the behavior track offset information included in the first fraud candidate information contact data and the behavior track search information and the behavior track offset information included in each third fraud candidate information contact data.

Step a60424, when each first fraud candidate intelligence contact data and each third fraud candidate intelligence contact data of a first fraud associated data sequence all satisfy the predetermined fraud reference content recommendation condition, determining the first fraud associated data sequence as a candidate fraud associated data sequence.

Step A60425, determining the matching degree of each candidate fraudulent conduct related data sequence and the fraudulent conduct track data sequence according to the matching degree between each third fraudulent candidate intelligence related data of each candidate fraudulent conduct related data sequence and each first fraudulent candidate intelligence related data.

Step A60426, obtaining target fraud behavior associated data sequences with matching degrees larger than a preset matching degree from the candidate fraud behavior associated data sequences, establishing fraud propagation information of each target fraud behavior associated data sequence and the fraud propagation track data sequence, determining target fraud reference content data from the fraud propagation track data sequences according to fraud propagation strength corresponding to the fraud propagation information, and performing fraud reference content pushing on the target fraud reference content data.

It can be understood that, by executing the step a 60421-step a60426, each first fraud candidate intelligence contact data of the fraud track data sequence and each second fraud candidate intelligence contact data of each first fraud associated data sequence can be obtained based on the interval data of the intelligence routing cost in the content tracing data, so that the first fraud candidate intelligence contact data and the second fraud candidate fraud intelligence contact data are analyzed, and the fraud propagation strength is determined. Therefore, when the target fraud reference content data is determined, different intelligence routing costs and fraud propagation strength can be considered, so that accurate analysis of the target fraud reference content data can be ensured, and influence between the target fraud reference content data and fraud track data is avoided.

Fig. 3 is a schematic functional module diagram of an information fraud recognition apparatus 300 based on big data and artificial intelligence provided by an embodiment of the present disclosure, and the functions of the functional modules of the information fraud recognition apparatus 300 based on big data and artificial intelligence are described in detail below.

The obtaining module 310 is configured to obtain big data of the fraud event to be identified of the digital service device.

The extracting module 320 is configured to perform decision vector extraction on big data of a fraud event to be identified according to a fraud behavior identification model meeting a model convergence requirement, to obtain first fraud decision vectors of multiple fraud assessment categories, where each first fraud decision vector includes a corresponding preset fraud intention.

The identifying module 330 is configured to perform fraud intention identification on each first fraud discrimination vector to obtain a candidate fraud intention corresponding to the target fraud trigger data on the big data of the fraud event to be identified.

The fusion module 340 is configured to obtain a second fraud discrimination vector corresponding to the first fraud discrimination vector according to the first fraud discrimination vector, and perform vector weight fusion on the first fraud discrimination vector and the corresponding second fraud discrimination vector to obtain a corresponding third fraud discrimination vector.

And the pushing module 350 is configured to perform fraud intention identification on the third fraud discrimination vector according to the candidate fraud intention, obtain a target fraud intention corresponding to the target fraud trigger data on the big data of the fraud event to be identified, and perform fraud reference content pushing on the digital service device according to the target fraud intention.

Fig. 4 illustrates a hardware structure of the big data system 100 for implementing the big data and artificial intelligence based information fraud recognition method, as provided by the embodiment of the present disclosure, and as shown in fig. 4, the big data system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

In a specific implementation process, at least one processor 110 executes computer-executable instructions stored in the machine-readable storage medium 120, so that the processor 110 may execute the big data and artificial intelligence-based information fraud identification method according to the above method embodiment, the processor 110, the machine-readable storage medium 120, and the communication unit 140 are connected through the bus 130, and the processor 110 may be configured to control the transceiving action of the communication unit 140, so as to perform data transceiving with the aforementioned digital service device 200.

For a specific implementation process of the processor 110, reference may be made to the above-mentioned various method embodiments executed by the big data system 100, which implement principles and technical effects are similar, and details of this embodiment are not described herein again.

In addition, the embodiment of the disclosure also provides a readable storage medium, in which computer-executable instructions are preset, and when a processor executes the computer-executable instructions, the information fraud identification method based on big data and artificial intelligence is realized.

Finally, it should be understood that the examples in this specification are only intended to illustrate the principles of the examples in this specification. Other variations are also possible within the scope of this description. Accordingly, by way of example, and not limitation, alternative configurations of the embodiments of the specification can be seen as matching the teachings of the specification. Accordingly, the embodiments of the present description are not limited to only those embodiments explicitly described and depicted herein.

Claims

1. An information fraud identification method based on big data and artificial intelligence is applied to a big data system, wherein the big data system is in communication connection with a plurality of digital service devices, and the method comprises the following steps:

2. The big data and artificial intelligence based information fraud identification method according to claim 1, characterized in that said method further comprises:

obtaining example fraud event big data, and inputting the example fraud event big data into an initial fraud behavior recognition model, wherein the example fraud event big data comprises example fraud intention corresponding to target fraud trigger data;

extracting a discrimination vector from the sample fraud event big data to obtain first sample discrimination vectors of a plurality of fraud assessment categories, wherein each first sample discrimination vector comprises a corresponding preset fraud intention;

identifying fraud intentions of each first example discrimination vector to obtain first identification fraud intentions corresponding to each preset fraud intention, and determining second identification fraud intentions from each first identification fraud intention according to intention loss data between each preset fraud intention and example fraud intentions;

obtaining a second example discrimination vector corresponding to the first example discrimination vector according to the first example discrimination vector, and performing vector weight fusion on the first example discrimination vector and the corresponding second example discrimination vector to obtain a corresponding third example discrimination vector;

carrying out fraud intention identification on the third example discrimination vector according to the second identification fraud intention to obtain a third identification fraud intention corresponding to the second identification fraud intention;

and generating model loss data according to the first recognition fraud intention and the intention loss data of the example fraud intention, the third recognition fraud intention and the intention loss data of the example fraud intention, and optimizing the model parameter configuration data of the fraud behavior recognition model according to the model loss data until a model convergence condition is met to obtain a target fraud behavior recognition model.

3. The method for identifying information fraud based on big data and artificial intelligence according to claim 2, wherein the identifying fraud intention for each first example discrimination vector to obtain a first identified fraud intention corresponding to each preset fraud intention, and determining a second identified fraud intention from each first identified fraud intention according to intention loss data between each preset fraud intention and an example fraud intention comprises:

respectively carrying out fraud intention recognition on each first example discrimination vector to obtain a first recognition difference data sequence corresponding to each first example discrimination vector; the first identification difference data sequence comprises first identification difference data corresponding to each preset cheating intention on the first example discrimination vector;

obtaining a corresponding first recognition fraud intention according to a preset fraud intention and corresponding first recognition difference data;

in a current first example discrimination vector, according to the intention matching degree between each preset cheating intention and an example cheating intention, determining a reference preset cheating intention from each preset cheating intention, and taking a first recognition cheating intention corresponding to the reference preset cheating intention as a candidate recognition cheating intention corresponding to the current first example discrimination vector;

and obtaining the second identification fraud intention according to the candidate identification fraud intentions respectively corresponding to the first example discrimination vectors.

4. The big-data and artificial-intelligence based information fraud identification method of claim 2, wherein said first example discrimination vector of said plurality of fraud assessment categories is a first example discrimination vector arranged by assessment level of fraud assessment category;

the obtaining a second example discrimination vector corresponding to the first example discrimination vector according to the first example discrimination vector includes:

and deriving the fraud assessment category of the current first example discrimination vector into the fraud assessment category of the derived assessment level corresponding to the current first example discrimination vector, and taking the derived current first example discrimination vector as a second example discrimination vector corresponding to a first example discrimination vector which is the same as the fraud assessment category of the derived current first example discrimination vector.

5. The big data and artificial intelligence based information fraud identification method according to claim 2, wherein said performing fraud intention identification on said third example discrimination vector according to said second identified fraud intention to obtain a third identified fraud intention corresponding to the second identified fraud intention comprises:

determining fraud directed graph information among the first example discrimination vectors according to the evaluation level of the fraud evaluation category of the first example discrimination vectors;

connecting the second identified fraud intention direction to a third example discrimination vector corresponding to the first example discrimination vector according to the fraud directed graph information;

and carrying out fraud intention identification on the directionally connected third example discrimination vector to obtain a third identification fraud intention corresponding to the second identification fraud intention.

6. The method for identifying information fraud according to claim 5, wherein the identifying fraud intention on the directionally-connected third example discrimination vector to obtain a third identified fraud intention corresponding to the second identified fraud intention comprises:

carrying out fraud intention identification on the directionally connected third example discrimination vector to obtain second identification difference data corresponding to a second identification fraud intention;

and obtaining a corresponding third recognition fraud intention according to the second recognition fraud intention and the corresponding second recognition difference data.

7. The big data and artificial intelligence based information fraud recognition method of claim 2, wherein said optimizing model parameter configuration data of said fraud recognition model according to said model loss data until a model convergence condition is satisfied to obtain a target fraud recognition model comprises:

determining reference recognition accuracy corresponding to a preset fraud intention and a second recognized fraud intention of which the intention matching degree of the example fraud intention is greater than the preset matching degree as first recognition accuracy, and determining reference recognition accuracy corresponding to the preset fraud intention and the second recognized fraud intention of which the intention matching degree of the example fraud intention is less than or equal to the preset matching degree as second recognition accuracy;

generating accuracy loss data according to fraud accuracy and reference identification accuracy corresponding to a preset fraud intention, and fraud accuracy and reference identification accuracy corresponding to a second identification fraud intention, wherein the fraud accuracy corresponding to the preset fraud intention is obtained by carrying out fraud intention identification on a first example discrimination vector, and the fraud accuracy corresponding to the second identification fraud intention is obtained by carrying out fraud intention identification on a third example discrimination vector;

optimizing model parameter configuration data of the fraud behavior recognition model according to the model loss data and the accuracy loss data until a model convergence condition is met to obtain a target fraud behavior recognition model;

wherein, the optimizing the model parameter configuration data of the fraud identification model according to the model loss data and the accuracy loss data until meeting the model convergence condition to obtain the target fraud identification model comprises:

inputting the big data of the example fraud event into a global fraud intention analysis model after training is finished to obtain a first global discrimination vector corresponding to each first example discrimination vector and a second global discrimination vector corresponding to each third example discrimination vector; the model parameter configuration data volume of the global fraud intention analysis model is larger than that of the fraud behavior recognition model, and a corresponding relation exists between the global fraud intention analysis model and a discrimination vector extraction layer of the fraud behavior recognition model;

generating global loss data according to a first degree of intersection between the first example discrimination vector and the corresponding first global discrimination vector and a second degree of intersection between the third example discrimination vector and the corresponding second global discrimination vector;

updating model parameter configuration data of the fraud behavior recognition model according to the model loss data, the accuracy loss data and the global loss data until a model convergence condition is met to obtain a target fraud behavior recognition model;

wherein generating global loss data based on a first degree of intersection between a first example discrimination vector and a corresponding first global discrimination vector, and a second degree of intersection between a third example discrimination vector and a corresponding second global discrimination vector, comprises:

carrying out fraud assessment category mapping on each first example discrimination vector so that the fraud assessment categories of the first example discrimination vector mapped by each fraud assessment category and the corresponding first global discrimination vector are the same;

calculating first discrimination vector intersection degrees between the first example discrimination vectors and corresponding first global discrimination vectors after fraud assessment category mapping, and obtaining the first intersection degrees according to the first discrimination vector intersection degrees;

carrying out fraud assessment type mapping on each third example discrimination vector so as to enable the fraud assessment type of each fraud assessment type mapped third example discrimination vector to be the same as that of the corresponding second global discrimination vector;

calculating second discrimination vector intersection degrees between the third example discrimination vector after fraud assessment category mapping and a corresponding second global discrimination vector, and obtaining the second intersection degrees according to the second discrimination vector intersection degrees;

and generating global loss data according to the first and second intersection degrees.

8. The big data and artificial intelligence based information fraud identification method according to any of claims 1-7, wherein said step of pushing fraud reference content to said digital service device according to said target fraud intention comprises:

acquiring a fraud behavior track data sequence of the target fraud intention;

obtaining key fraud migration behaviors matched with preset fraud risk service characteristics from the fraud behavior track data sequence to obtain fraud migration data of the key fraud migration behaviors; wherein the critical fraudulent migration behavior comprises direct fraudulent migration behavior and indirect fraudulent migration behavior;

extracting target fraud unit data of associated fraud reference content for determining a target fraud intention from the fraud migration data;

obtaining associated fraud reference content of the target fraud intention based on the target fraud unit data to obtain content tracing data of the associated fraud reference content of the target fraud intention;

and determining target fraud reference content data from the fraud track data sequence through the content tracing data, and performing fraud reference content pushing on the target fraud reference content data.

9. The method for identifying information fraud based on big data and artificial intelligence according to claim 1, wherein the obtaining of the associated fraud reference content of the target fraud intention based on the target fraud unit data to obtain the content tracing data of the associated fraud reference content of the target fraud intention comprises:

determining a plurality of traceability intelligence sequences for the target fraud intent based on the target fraud unit data;

aiming at each traceability information sequence which does not meet the preset information condition, processing the traceability information atlas of the traceability information sequence to obtain a first traceability information group which meets an information contact template, and adding the first traceability information group into an information contact library corresponding to an information contact unit corresponding to the traceability information sequence, wherein the information contact template is as follows: associating an information contact template of an information contact unit corresponding to the tracing information sequence in the obtaining process of the fraud reference content;

optimizing a first tracing situation group configured in the information contact library corresponding to the preset social entity network data for mining the social entity network sequence by adopting a pre-generated second tracing situation group, wherein the second tracing situation group is as follows: processing dynamic traceability information corresponding to a target traceability information sequence to obtain traceability information parties meeting an information contact template of the mined social entity network sequence, wherein the target traceability information sequence is the traceability information sequence meeting the preset information condition, and an information contact library corresponding to the mined social entity network sequence and an information contact library corresponding to any information contact unit are stored in a shared information contact library;

determining a reference tracing situation newspaper group with information routing cost in the information contact library;

and obtaining the associated fraud reference content of the target fraud intention by adopting the reference tracing source report group to obtain the content tracing data of the associated fraud reference content of the target fraud intention.

10. A big data system, comprising a processor and a machine-readable storage medium having stored therein at least one instruction, at least one program, a set of codes, or a set of instructions, which is loaded and executed by the processor to implement the big data and artificial intelligence based information fraud identification method of any one of claims 1-9.