Telephone fraud risk identification system and method based on big data

Abstract

The invention discloses a big data-based telephone fraud risk identification system and method, relates to the technical field of telephone fraud risk identification, and solves the problems that the prior art mainly depends on historical data, and the identification capability of a novel or unseen fraud means such as a voice synthesis fraud means is limited, so that the fraud risk cannot be identified. And the technical problem of high false alarm rate or missing report rate caused by the fact that the characteristics of all fraudulent behaviors cannot be fully covered by judging only according to the voice record in the call, the incoming call number, the attribution of the incoming call and other information is obviously solved. According to the invention, the communication data of the telephone is obtained in real time; based on the big data, constructing a historical telephone fraud database; calculating through a similarity algorithm to obtain a target similarity between the communication data and a historical telephone fraud database; calculating a telephone fraud risk assessment coefficient based on the communication data and the target similarity; and judging whether to carry out early warning based on the telephone fraud risk assessment coefficient. The technical problem is solved.

Description

Technical Field

[0001] The present invention belongs to the field of telephone fraud risk identification, and specifically relates to a telephone fraud risk identification system and method based on big data. Background Art

[0002] With the rapid development of information technology, telephone communications have become an indispensable part of people's daily lives; however, telephone fraud cases occur frequently, which not only poses a threat to personal property safety, but also seriously affects the harmony and stability of society; traditional telephone fraud prevention measures mostly rely on user reports or simple blacklist mechanisms. These methods have lags and limitations, and it is difficult to effectively prevent new types of fraud.

[0003] The prior art (invention application with application number 2017114916534) discloses a fraudulent call identification method, device, equipment and medium, including the following steps: obtaining the basic call data of the user; performing statistics and calculation on the basic call data to generate each phone number and its data features of the call with the user; selecting a plurality of known blacklist numbers and a plurality of known non-blacklist numbers from the phone numbers of the call with the user; establishing a binary classification training model, inputting the data features of the plurality of known blacklist numbers and the data features of the plurality of known non-blacklist numbers into the binary classification training model, training the binary classification training model to obtain a trained binary classification training model; and inputting the data features of the new number into the trained binary classification training model to identify whether the new number is a fraudulent call. The invention application can directly identify fraudulent calls, without the need for the user to make identification judgments and without the need to install a client, and can perform analysis and judgment on the network side to discover fraudulent calls. However, existing technologies mainly rely on historical data. Their ability to identify new or unseen fraud methods, such as voice synthesis fraud, is limited. In addition, judgments based solely on voice records, caller numbers and their locations during calls are obviously unable to fully cover the characteristics of all fraudulent behaviors, resulting in high false alarm or missed alarm rates.

[0004] Therefore, the present invention proposes a telephone fraud risk identification system and method based on big data to solve the above problems. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a telephone fraud risk identification system and method based on big data, which is used to solve the technical problems that the prior art mainly relies on historical data, and its recognition ability is limited for new or unseen fraud methods, such as voice synthesis fraud methods, and it is obviously impossible to fully cover the characteristics of all fraudulent behaviors by judging only based on information such as voice records in calls, caller numbers and their locations, resulting in a high false alarm rate or missed alarm rate.

[0006] To achieve the above-mentioned purpose, the first aspect of the present invention provides a telephone fraud risk identification system based on big data, comprising: a data collection module, a data analysis module, a risk identification module and an early warning module;

[0007] Data collection module: used to obtain telephone communication data in real time; and to build a historical telephone fraud database based on big data;

[0008] Data analysis module: calculates the target similarity between the communication data and the historical telephone fraud database through a similarity algorithm;

[0009] Risk identification module: calculates the telephone fraud risk assessment coefficient based on communication data and target similarity;

[0010] Early warning module: Determine whether to issue an early warning based on the telephone fraud risk assessment coefficient.

[0011] Preferably, the real-time acquisition of telephone communication data includes:

[0012] Extract the communication data in the phone; wherein the communication data includes: the voice record of the call, the incoming call number and the location corresponding to the incoming call number.

[0013] Preferably, the construction of a historical telephone fraud database based on big data includes:

[0014] By obtaining existing known telephone fraud cases from big data, and extracting corresponding historical communication data from each telephone fraud case; marking the extracted historical communication data of each telephone fraud case as reference data;

[0015] Extracting emotional data contained in the voice recording of each reference data through an emotional analysis tool; wherein the emotional analysis tool includes: VADER or TextBlob; the emotional data includes: emotional polarity and emotional intensity;

[0016] The speech recording in each reference data is converted into speech text by speech recognition technology and marked as initial text data; wherein the speech recognition technology includes: Google Speech-to-Text API or IBM Watson Speech toText;

[0017] Integrate the sentiment data of each reference data, each word in the initial text data, the incoming call number, and the location corresponding to the incoming call number into initial data;

[0018] Each initial data is converted into a vector of fixed dimension through a word vector model and marked as a feature vector; wherein the word vector model includes: Word2Vec or GloVe;

[0019] By storing the feature vector in a preset database, a historical telephone fraud database is obtained; wherein the preset database includes: MySQL, PostgreSQL or MongoDB.

[0020] It should be noted that sentiment polarity describes the direction of sentiment, which is usually divided into positive, negative and neutral; sentiment intensity describes the strength or degree of sentiment, measuring the intensity of sentiment; combining sentiment polarity and sentiment intensity can more comprehensively describe the state of emotion and better understand the emotional information in text or speech.

[0021] Preferably, the method of calculating the target similarity between the communication data and the historical telephone fraud database by a similarity algorithm includes:

[0022] Extract emotional data contained in voice records in communication data through sentiment analysis tools;

[0023] The voice records in the communication data are converted into voice text through voice recognition technology and marked as text data;

[0024] Integrate the emotional data contained in the voice recording in the communication data, each word in the text data, the caller number and the location corresponding to the caller number into the data to be identified;

[0025] The data to be identified is converted into a vector of fixed dimension through the word vector model and marked as the vector to be identified;

[0026] The similarity between the vector to be identified and the feature vector of each initial data in the historical telephone fraud database is calculated by a similarity algorithm to obtain several similarities; wherein the similarity algorithm includes: cosine similarity or Euclidean distance;

[0027] The highest similarity is selected as the target similarity.

[0028] Preferably, the telephone fraud risk assessment coefficient is calculated based on the communication data and the target similarity, including:

[0029] Calculate the incoming call speech synthesis evaluation coefficient based on the communication data;

[0030] Filter the number of sensitive keywords contained in the voice records in the communication data based on the preset sensitive keywords;

[0031] The telephone fraud risk assessment coefficient is calculated based on the incoming call speech synthesis assessment coefficient, the number of sensitive keywords and the target similarity.

[0032] It should be noted that the preset sensitive keywords refer to common and potentially risky words or phrases in telephone fraud; for example, "bank", "account", "password", "transfer", "remittance", "credit card", "verification code", etc.

[0033] Preferably, the step of calculating the incoming call speech synthesis evaluation coefficient based on the communication data includes:

[0034] Extract voice recordings from communication data;

[0035] Extracting audio feature data of the speech recording by feature extraction technology, and calculating the standard deviation of each feature in the audio feature data; wherein the audio feature data includes: spectral energy, fundamental frequency, formant, zero crossing rate, background noise and speech rate;

[0036] By formula: The incoming call speech synthesis evaluation coefficient is calculated; wherein YP is the incoming call speech synthesis evaluation coefficient, fi is the standard deviation of the i-th feature in the audio feature data, wi is the weight coefficient of the standard deviation of the i-th feature, i is the sequence number of each feature in the audio feature data, and i = {1, 2, ..., n}, n is the number of features in the audio feature data.

[0037] It should be noted that the feature extraction techniques include: STFT, autocorrelation method, cepstrum method, LPC, short-time energy ratio, ASR and VAD, etc.

[0038] The audio feature data of natural human voices will change with factors such as pronunciation, intonation, and emotion, resulting in a larger standard deviation of the features in the audio feature data; while the audio feature data of synthetic speech is usually very regular and lacks these natural changes. Therefore, the smaller the standard deviation of the features in the audio feature data, the less the audio feature data changes, and the greater the probability of synthetic speech.

[0039] Preferably, the method of screening the number of sensitive keywords contained in the voice recording in the communication data based on preset sensitive keywords includes:

[0040] Extracting text data corresponding to the voice record in the communication data;

[0041] The text data corresponding to the voice record in the communication data is screened based on the preset sensitive keywords to obtain the number of sensitive keywords contained in the text data.

[0042] Preferably, the telephone fraud risk assessment coefficient is calculated based on the incoming call speech synthesis assessment coefficient, the number of sensitive keywords and the target similarity, including:

[0043] The number of sensitive keywords is marked as M, and the target similarity is marked as X;

[0044] By formula: The telephone fraud risk assessment coefficient is calculated; wherein P is the telephone fraud risk assessment coefficient; A, B, and C are proportional coefficients, and 0<A<1, 0<B<1, and 0<C<1.

[0045] It should be noted that the proportional coefficients A, B, and C are set by experts in this field based on experience according to historical fraud cases.

[0046] Preferably, the determining whether to issue an early warning based on the telephone fraud risk assessment coefficient includes:

[0047] Determine whether the telephone fraud risk assessment coefficient is greater than the preset risk assessment threshold; if yes, send a warning message to the client; if not, continue to monitor and judge.

[0048] It should be noted that the preset risk assessment threshold is set by experts in this field based on experience of historical fraud cases.

[0049] A second aspect of the present invention provides a method for identifying telephone fraud risks based on big data, comprising:

[0050] Step 1: Obtain telephone communication data in real time; and build a historical telephone fraud database based on big data;

[0051] Step 2: Calculate the target similarity between the communication data and the historical telephone fraud database through a similarity algorithm;

[0052] Step 3: Calculate the telephone fraud risk assessment coefficient based on the communication data and target similarity;

[0053] Step 4: Determine whether to issue an early warning based on the telephone fraud risk assessment coefficient.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] 1. The existing technology mainly relies on historical data. For new or unseen fraud methods, such as voice synthesis fraud methods, its recognition ability will be limited. In addition, judging only by voice records, caller numbers and their locations during calls, it is obviously impossible to fully cover the characteristics of all fraud behaviors, resulting in a high false alarm rate or missed alarm rate. The present invention not only obtains the communication data of the phone in real time, including voice records, caller numbers and locations, but also extracts the emotional polarity and intensity in the voice through sentiment analysis tools, and uses advanced speech recognition technology to convert the voice into text, and combines the word vector model to convert multi-dimensional data into feature vectors, and constructs a huge historical telephone fraud database. On this basis, the present invention calculates the similarity between real-time communication data and historical fraud cases through a similarity algorithm, comprehensively considers the caller voice synthesis evaluation coefficient, the number of sensitive keywords and the target similarity, and calculates the telephone fraud risk assessment coefficient, thereby realizing accurate identification and early warning of potential fraud behaviors. Through multi-dimensional data analysis, big data-driven learning mechanism and intelligent evaluation and early warning mechanism, the accuracy and efficiency of telephone fraud risk identification are significantly improved, and the problem of high false alarm rate or missed alarm rate in the existing technology is solved.

[0056] 2. Since the audio features of natural human voices are highly variable and unpredictable due to multiple factors such as pronunciation, intonation, and emotion, the standard deviation of their audio feature data is relatively large; in contrast, synthetic speech usually exhibits a high degree of consistency and regularity, lacking the fluctuations inherent in natural language, and therefore the standard deviation of its audio feature data is relatively small; the present invention extracts voice records from communication data, extracts the audio feature data of the voice records through feature extraction technology, and calculates the standard deviation of these features; combines preset weight coefficients, and uses a specific formula to calculate the electronic speech synthesis evaluation coefficient; it can not only distinguish between natural human voices and synthetic speech, thereby more accurately identifying possible telephone fraud, but also improve the ability to respond to new fraud methods such as speech synthesis fraud, reduce false alarm rates and missed alarm rates, and provide users with a safer and more reliable communication environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0058] Figure 1 A schematic diagram of a system module according to an embodiment of the present invention;

[0059] Figure 2 A schematic diagram of a specific process of an embodiment of the present invention;

[0060] Figure 3 Schematic diagram of the method steps of an embodiment of the present invention. DETAILED DESCRIPTION

[0061] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] See also Figure 1-Figure 2 , the first aspect of the present invention provides a telephone fraud risk identification system based on big data, including: a data collection module, a data analysis module, a risk identification module and an early warning module;

[0063] Data collection module: used to obtain telephone communication data in real time; and to build a historical telephone fraud database based on big data;

[0064] Data analysis module: calculates the target similarity between the communication data and the historical telephone fraud database through a similarity algorithm;

[0065] Risk identification module: calculates the telephone fraud risk assessment coefficient based on communication data and target similarity;

[0066] Early warning module: Determine whether to issue an early warning based on the telephone fraud risk assessment coefficient.

[0067] Get phone communication data in real time, including:

[0068] Extract the communication data in the phone; wherein the communication data includes: the voice record of the call, the incoming call number and the location corresponding to the incoming call number.

[0069] Build a historical telephone fraud database based on big data, including:

[0070] By obtaining existing known telephone fraud cases from big data, and extracting corresponding historical communication data from each telephone fraud case; marking the extracted historical communication data of each telephone fraud case as reference data;

[0071] Extracting emotional data contained in the voice recording of each reference data through an emotional analysis tool; wherein the emotional analysis tool includes: VADER or TextBlob; the emotional data includes: emotional polarity and emotional intensity;

[0072] The speech recording in each reference data is converted into speech text by speech recognition technology and marked as initial text data; wherein the speech recognition technology includes: Google Speech-to-Text API or IBM Watson Speech toText;

[0073] Integrate the sentiment data of each reference data, each word in the initial text data, the incoming call number, and the location corresponding to the incoming call number into initial data;

[0074] Each initial data is converted into a vector of fixed dimension through a word vector model and marked as a feature vector; wherein the word vector model includes: Word2Vec or GloVe;

[0075] By storing the feature vector in a preset database, a historical telephone fraud database is obtained; wherein the preset database includes: MySQL, PostgreSQL or MongoDB.

[0076] The target similarity between the communication data and the historical telephone fraud database is calculated by the similarity algorithm, including:

[0077] Extract emotional data contained in voice records in communication data through sentiment analysis tools;

[0078] The voice records in the communication data are converted into voice text through voice recognition technology and marked as text data;

[0079] Integrate the emotional data contained in the voice recording in the communication data, each word in the text data, the caller number and the location corresponding to the caller number into the data to be identified;

[0080] The data to be identified is converted into a vector of fixed dimension through the word vector model and marked as the vector to be identified;

[0081] The similarity between the vector to be identified and the feature vector of each initial data in the historical telephone fraud database is calculated by a similarity algorithm to obtain several similarities; wherein the similarity algorithm includes: cosine similarity or Euclidean distance;

[0082] The highest similarity is selected as the target similarity.

[0083] The telephone fraud risk assessment coefficient is calculated based on the communication data and target similarity, including:

[0084] Calculate the incoming call speech synthesis evaluation coefficient based on the communication data;

[0085] Filter the number of sensitive keywords contained in the voice records in the communication data based on the preset sensitive keywords;

[0086] The telephone fraud risk assessment coefficient is calculated based on the incoming call speech synthesis assessment coefficient, the number of sensitive keywords and the target similarity.

[0087] The incoming call speech synthesis evaluation coefficient is calculated based on the communication data, including:

[0088] Extract voice recordings from communication data;

[0089] Extracting audio feature data of the speech recording by feature extraction technology, and calculating the standard deviation of each feature in the audio feature data; wherein the audio feature data includes: spectral energy, fundamental frequency, formant, zero crossing rate, background noise and speech rate;

[0090] By formula: The incoming call speech synthesis evaluation coefficient is calculated; wherein YP is the incoming call speech synthesis evaluation coefficient, fi is the standard deviation of the i-th feature in the audio feature data, wi is the weight coefficient of the standard deviation of the i-th feature, i is the sequence number of each feature in the audio feature data, and i = {1, 2, ..., n}, n is the number of features in the audio feature data.

[0091] Filter the number of sensitive keywords contained in the voice records in the communication data based on the preset sensitive keywords, including:

[0092] Extracting text data corresponding to the voice record in the communication data;

[0093] The text data corresponding to the voice record in the communication data is screened based on the preset sensitive keywords to obtain the number of sensitive keywords contained in the text data.

[0094] The telephone fraud risk assessment coefficient is calculated based on the incoming call speech synthesis assessment coefficient, the number of sensitive keywords and the target similarity, including:

[0095] The number of sensitive keywords is marked as M, and the target similarity is marked as X;

[0096] By formula: The telephone fraud risk assessment coefficient is calculated; wherein P is the telephone fraud risk assessment coefficient; A, B, and C are proportional coefficients, and 0<A<1, 0<B<1, and 0<C<1.

[0097] Determine whether to issue an early warning based on the telephone fraud risk assessment coefficient, including:

[0098] Determine whether the telephone fraud risk assessment coefficient is greater than the preset risk assessment threshold; if yes, send a warning message to the client; if not, continue to monitor and judge.

[0099] For example, suppose a user receives a call from an unfamiliar number one day, and the content of the call involves bank accounts and transfers. Now it is necessary to identify the risk of telephone fraud for this call, as follows:

[0100] 1. Data collection module: real-time acquisition of communication data; and building a historical telephone fraud database based on big data;

[0101] The system first extracts the following communication data from the call:

[0102] Voice recording (a call recording of about 5 minutes), caller ID and location of the caller ID.

[0103] 2. Sentiment analysis and speech recognition;

[0104] 2.1 Sentiment analysis;

[0105] The system uses sentiment analysis tools (such as VADER or TextBlob) to perform sentiment analysis on the voice recordings of the call and extract sentiment polarity and sentiment intensity. Assume that the sentiment analysis results show:

[0106] Sentiment polarity: Negative (-0.8), indicating that there is a clear negative sentiment in the call.

[0107] Emotional intensity: High (0.9), indicating that the emotional expression is very strong.

[0108] 2.2 Speech recognition;

[0109] The system converts the voice recording into text through speech recognition technology (such as Google Speech-to-Text API or IBM Watson Speech to Text). Assume that the converted text is: "Hello, I am your bank customer service. There is an abnormality in your account. Please provide your bank card number and verification code to confirm your identity, otherwise your account may be frozen."

[0110] 3. Feature extraction and vectorization;

[0111] 3.1 Audio feature extraction;

[0112] The system uses a variety of feature extraction techniques (such as STFT, autocorrelation method, LPC, etc.) to analyze the voice recordings and extract audio features: spectral energy, fundamental frequency, formant, zero-crossing rate, background noise, and speech rate;

[0113] 3.2 Calculate the standard deviation;

[0114] Calculate the standard deviation of each audio feature. Assume the calculation results are as follows:

[0115] Standard deviation of spectrum energy: 5dB;

[0116] Standard deviation of fundamental frequency: 2Hz;

[0117] Standard deviation of the resonance peak: F1 = 10 Hz;

[0118] Standard deviation of zero-crossing rate: 0.01;

[0119] Standard deviation of background noise: 1dB;

[0120] Standard deviation of speaking speed: 5 words / minute;

[0121] 3.3 Incoming call speech synthesis evaluation coefficient;

[0122] According to the formula: Calculate the evaluation coefficient of incoming call speech synthesis;

[0123] Assume the weight coefficient is:

[0124] Spectral energy w1: 0.2;

[0125] Fundamental frequency w2: 0.2;

[0126] Resonance peak w3: 0.2;

[0127] Zero crossing rate w4: 0.1;

[0128] Background noise w5: 0.1;

[0129] Speech speed w6: 0.2;

[0130] By calculation, we get YP≈4.5;

[0131] 4. Screening of sensitive keywords;

[0132] The call text is screened based on the preset sensitive keyword library (such as "bank", "account", "password", "transfer", "remittance", "credit card", "verification code", etc.). Assume that the number of sensitive keywords found in the call text is: 15 times, so M = 15.

[0133] 5. Similarity algorithm calculates target similarity

[0134] 5.1 Build a historical telephone fraud database;

[0135] Obtain known cases of telephone fraud from big data and extract historical communication data for each case, including voice records, sentiment data, caller numbers and locations, etc. These data are marked as reference data and converted into fixed-dimensional feature vectors through word vector models (such as Word2Vec or GloVe) and stored in a preset database (such as MySQL, PostgreSQL or MongoDB).

[0136] 5.2 Calculate similarity;

[0137] Compare the feature vector of the current call with the feature vector of each initial data in the historical telephone fraud database, and use the cosine similarity algorithm to calculate the similarity. Assume that the similarities calculated by the system are: 0.85, 0.78, 0.92, 0.65, 0.88;

[0138] The highest similarity is selected as the target similarity, that is, X=0.92.

[0139] 6. Calculate the telephone fraud risk assessment coefficient;

[0140] According to the formula: Calculate telephone fraud risk assessment coefficient;

[0141] Assume the scale factor is:

[0142] A = 0.3;

[0143] B = 0.4;

[0144] C = 0.3;

[0145] The calculation yields P≈1.75;

[0146] 7. Early warning mechanism;

[0147] Determine whether to issue an early warning based on the preset risk assessment threshold.

[0148] Assume that the preset risk assessment threshold is 1.0.

[0149] Since the calculated telephone fraud risk assessment coefficient (P=1.75) is greater than the threshold value of 1.0, the system will automatically send an early warning message to the client to remind the user that this call may involve a fraud risk and advise the user to handle it with caution.

[0150] How does the present invention comprehensively calculate the telephone fraud risk assessment coefficient and issue a warning in a timely manner through multi-dimensional data analysis, sentiment analysis, speech synthesis evaluation, sensitive keyword screening and similarity algorithm? Compared with the existing technology, the present invention can not only identify traditional fraud methods, but also effectively deal with new speech synthesis fraud methods, significantly improve the accuracy and efficiency of telephone fraud risk identification, reduce the false alarm rate and missed alarm rate, and provide users with a safer and more reliable communication environment.

[0151] See also Figure 3 The second aspect of the present invention provides a method for identifying telephone fraud risks based on big data, including:

[0152] Step 1: Obtain telephone communication data in real time; and build a historical telephone fraud database based on big data;

[0153] Step 2: Calculate the target similarity between the communication data and the historical telephone fraud database through a similarity algorithm;

[0154] Step 3: Calculate the telephone fraud risk assessment coefficient based on the communication data and target similarity;

[0155] Step 4: Determine whether to issue an early warning based on the telephone fraud risk assessment coefficient.

[0156] Part of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0157] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A telephone fraud risk identification system based on big data, characterized in that: include: Data collection module, data analysis module, risk identification module and early warning module; Data collection module: used to obtain telephone communication data in real time; and, building a historical telephone fraud database based on big data; Data analysis module: calculates the target similarity between the communication data and the historical telephone fraud database through a similarity algorithm; Risk identification module: calculates the telephone fraud risk assessment coefficient based on communication data and target similarity; Early warning module: Determine whether to issue an early warning based on the telephone fraud risk assessment coefficient.

2. According to the big data-based telephone fraud risk identification system of claim 1, it is characterized in that: The real-time acquisition of telephone communication data includes: Extract the communication data in the phone; wherein the communication data includes: the voice record of the call, the incoming call number and the location corresponding to the incoming call number.

3. According to the big data-based telephone fraud risk identification system of claim 1, it is characterized in that: The historical telephone fraud database is constructed based on big data, including: By obtaining existing known telephone fraud cases from big data, and extracting corresponding historical communication data from each telephone fraud case; marking the extracted historical communication data of each telephone fraud case as reference data; Extracting emotional data contained in the voice recording of each reference data through an emotional analysis tool; wherein the emotional analysis tool includes: VADER or TextBlob; the emotional data includes: emotional polarity and emotional intensity; The speech recording in each reference data is converted into speech text by speech recognition technology and marked as initial text data; wherein the speech recognition technology includes: Google Speech-to-Text API or IBM Watson Speech toText; Integrate the sentiment data of each reference data, each word in the initial text data, the incoming call number, and the location corresponding to the incoming call number into initial data; Each initial data is converted into a vector of fixed dimension through a word vector model and marked as a feature vector; wherein the word vector model includes: Word2Vec or GloVe; By storing the feature vector in a preset database, a historical telephone fraud database is obtained; wherein the preset database includes: MySQL, PostgreSQL or MongoDB.

4. According to the big data-based telephone fraud risk identification system of claim 1, it is characterized in that: The target similarity between the communication data and the historical telephone fraud database is calculated by the similarity algorithm, including: Extract emotional data contained in voice records in communication data through sentiment analysis tools; The voice records in the communication data are converted into voice text through voice recognition technology and marked as text data; Integrate the emotional data contained in the voice recording in the communication data, each word in the text data, the caller number and the location corresponding to the caller number into the data to be identified; The data to be identified is converted into a vector of fixed dimension through the word vector model and marked as the vector to be identified; The similarity between the vector to be identified and the feature vector of each initial data in the historical telephone fraud database is calculated by a similarity algorithm to obtain several similarities; wherein the similarity algorithm includes: cosine similarity or Euclidean distance; The highest similarity is selected as the target similarity.

5. According to the big data-based telephone fraud risk identification system of claim 1, it is characterized in that: The telephone fraud risk assessment coefficient calculated based on the communication data and the target similarity includes: Calculate the incoming call speech synthesis evaluation coefficient based on the communication data; Filter the number of sensitive keywords contained in the voice records in the communication data based on the preset sensitive keywords; The telephone fraud risk assessment coefficient is calculated based on the incoming call speech synthesis assessment coefficient, the number of sensitive keywords and the target similarity.

6. According to the big data-based telephone fraud risk identification system of claim 5, it is characterized in that: The calculation based on the communication data to obtain the incoming call speech synthesis evaluation coefficient includes: Extract voice recordings from communication data; Extracting audio feature data of the speech recording by feature extraction technology, and calculating the standard deviation of each feature in the audio feature data; wherein the audio feature data includes: spectral energy, fundamental frequency, formant, zero crossing rate, background noise and speech rate; By formula: The incoming call speech synthesis evaluation coefficient is calculated; wherein YP is the incoming call speech synthesis evaluation coefficient, fi is the standard deviation of the i-th feature in the audio feature data, wi is the weight coefficient of the standard deviation of the i-th feature, i is the sequence number of each feature in the audio feature data, and i = {1, 2, ..., n}, n is the number of features in the audio feature data.

7. The telephone fraud risk identification system based on big data according to claim 5 is characterized in that: The screening of the number of sensitive keywords contained in the voice recording in the communication data based on the preset sensitive keywords includes: Extracting text data corresponding to the voice record in the communication data; The text data corresponding to the voice record in the communication data is screened based on the preset sensitive keywords to obtain the number of sensitive keywords contained in the text data.

8. The telephone fraud risk identification system based on big data according to claim 5 is characterized in that: The telephone fraud risk assessment coefficient is calculated based on the incoming call speech synthesis assessment coefficient, the number of sensitive keywords and the target similarity, including: The number of sensitive keywords is marked as M, and the target similarity is marked as X; By formula: The telephone fraud risk assessment coefficient is calculated; wherein P is the telephone fraud risk assessment coefficient; A, B, and C are proportional coefficients, and 0<A<1, 0<B<1, and 0<C<1.

9. The telephone fraud risk identification system based on big data according to claim 1 is characterized in that: The determination of whether to issue an early warning based on the telephone fraud risk assessment coefficient includes: Determine whether the telephone fraud risk assessment coefficient is greater than the preset risk assessment threshold; if yes, send a warning message to the client; if not, continue to monitor and judge.

10. A telephone fraud risk identification method based on big data, applied to a telephone fraud risk identification system based on big data as claimed in any one of claims 1 to 9, characterized in that: include: Step 1: Obtain telephone communication data in real time; and build a historical telephone fraud database based on big data; Step 2: Calculate the target similarity between the communication data and the historical telephone fraud database through a similarity algorithm; Step 3: Calculate the telephone fraud risk assessment coefficient based on the communication data and target similarity; Step 4: Determine whether to issue an early warning based on the telephone fraud risk assessment coefficient.

Images