Fraud call identification method and device, computer equipment and storage medium

Abstract

The invention belongs to the technical field of artificial intelligence, and relates to a fraud phone recognition method and device, computer equipment and a storage medium. Wherein the call data comprises audio data and metadata; respectively carrying out feature extraction on the audio data and the metadata to obtain audio features and metadata features; performing fraud analysis on the audio features and the metadata features based on a telephone analysis model to obtain a fraud analysis result; if the fraud analysis result is that the call corresponding to the call data is a suspected fraud call, converting the audio data into text data; performing semantic feature extraction on the text data to obtain semantic features; performing fusion processing on the audio features, the metadata features and the semantic features to obtain comprehensive features; and performing fraud recognition on the comprehensive features based on a telephone recognition model to obtain a fraud telephone recognition result corresponding to the call. According to the invention, based on a model combination processing mode, the accuracy and efficiency of telephone fraud identification are effectively improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence development technology and financial technology, and in particular to methods, devices, computer equipment and storage media for identifying fraudulent calls. Background Art

[0002] In the financial sector, with the rapid development of telecommunications technology, telephone fraud has become an increasingly rampant and ever-changing form of illegal activity, seriously threatening the property safety of the general public and the stability of the financial order. In order to meet this challenge, a series of telephone fraud identification products and technologies have emerged on the market, aiming to provide early warning and intercept potential fraud calls, protecting users from fraud.

[0003] However, the current mainstream telephone fraud identification solutions mainly rely on traditional pattern recognition technologies, such as keyword filtering mechanisms and blacklist databases. Although these technologies can identify and prevent some known fraudulent behaviors to a certain extent, they have obvious limitations. For keyword filtering, keyword filtering sets a list of sensitive words and triggers an alarm when these words appear in the call content. However, fraudsters can often cleverly avoid directly mentioning sensitive words and use synonyms, metaphors or pinyin abbreviations, which greatly reduces the effectiveness of the keyword filtering mechanism. In addition, for the blacklist database method, the blacklist mechanism relies on known fraudulent phone numbers or number segments for interception. However, the source of fraudulent calls is often highly mobile, and fraudsters will frequently change numbers or use fake numbers to commit crimes, resulting in the blacklist update speed being difficult to keep up with the speed of change of fraudulent numbers, and there is no warning at all for fraudulent numbers that appear for the first time.

[0004] Therefore, existing methods of identifying telephone fraud have obvious deficiencies in both accuracy and timeliness when faced with constantly updated and highly concealed telephone fraud methods. Summary of the invention

[0005] The purpose of the embodiments of the present application is to propose a method, device, computer equipment and storage medium for identifying fraudulent calls, so as to solve the technical problem that the existing telephone fraud identification methods have obviously insufficient recognition accuracy and timeliness.

[0006] In order to solve the above technical problems, the present application embodiment provides a method for identifying fraudulent calls, which adopts the following technical solution:

[0007] Acquire the user's call data; wherein the call data includes audio data and metadata;

[0008] Performing feature extraction processing on the audio data and the metadata respectively to obtain corresponding audio features and metadata features;

[0009] Perform fraud analysis on the audio features and the metadata features based on a preset telephone analysis model to obtain corresponding fraud analysis results;

[0010] If the fraud analysis result is that the call corresponding to the call data is a suspected fraud call, converting the audio data into corresponding text data;

[0011] Extracting semantic features from the text data to obtain corresponding semantic features;

[0012] The audio feature, the metadata feature and the semantic feature are fused based on a preset fusion strategy to obtain corresponding comprehensive features;

[0013] The comprehensive features are used to identify fraud based on a preset phone identification model to obtain a fraud phone identification result corresponding to the call.

[0014] Furthermore, the step of performing feature extraction processing on the audio data and the metadata respectively to obtain corresponding audio features and metadata features specifically includes:

[0015] Preprocessing the audio data to obtain corresponding target audio data;

[0016] Extracting features from the target audio data based on the pre-trained model to obtain corresponding audio features;

[0017] Cleaning the metadata to obtain corresponding target metadata;

[0018] Perform feature engineering processing on the target metadata to obtain corresponding metadata features.

[0019] Furthermore, the step of extracting semantic features from the text data to obtain corresponding semantic features specifically includes:

[0020] Preprocessing the text data to obtain corresponding target text data;

[0021] Performing sentiment analysis on the target text data based on a preset sentiment analysis algorithm to obtain corresponding sentiment features;

[0022] Performing intent recognition on the target text data based on a preset intent recognition model to obtain corresponding intent features;

[0023] The semantic feature is generated based on the emotion feature and the intention feature.

[0024] Furthermore, the step of fusing the audio features, the metadata features, and the semantic features based on a preset fusion strategy to obtain corresponding comprehensive features specifically includes:

[0025] Normalizing the audio features, the metadata features, and the semantic features respectively to obtain processed designated audio features, designated metadata features, and designated semantic features;

[0026] The specified audio feature, the specified metadata feature and the specified semantic feature are concatenated to obtain corresponding multimodal features;

[0027] Performing dimensionality reduction processing on the multimodal features to obtain processed designated features;

[0028] The designated feature is used as the comprehensive feature.

[0029] Furthermore, the step of performing fraud identification on the comprehensive features based on a preset phone identification model to obtain a fraud phone identification result corresponding to the call specifically includes:

[0030] Performing inference processing on the comprehensive features based on the phone recognition model to calculate the probability that the call is a fraudulent call;

[0031] Determining whether the probability is greater than a preset probability threshold;

[0032] If so, generating a first fraud call identification result indicating that the call is a fraud call;

[0033] If not, a second fraud call identification result is generated indicating that the call is a normal call.

[0034] Furthermore, after the step of performing fraud identification on the comprehensive features based on the preset phone identification model to obtain a fraud phone identification result corresponding to the call, the method further includes:

[0035] Obtaining specified multimodal features of the identified fraudulent calls; wherein the number of the fraudulent calls includes a plurality;

[0036] Call the preset target clustering algorithm;

[0037] Performing cluster analysis on the designated multimodal features of the fraudulent calls based on the target clustering algorithm to obtain corresponding clustering results;

[0038] Analyze and process the clustering results based on a preset analysis strategy to obtain corresponding analysis results;

[0039] The clustering result and the analysis result are outputted.

[0040] Furthermore, after the step of performing fraud identification on the comprehensive features based on the preset phone identification model to obtain a fraud phone identification result corresponding to the call, the method further includes:

[0041] If the fraud call identification result is that the call is a fraud call, performing corresponding warning processing on the call;

[0042] generating a reminder message corresponding to the call;

[0043] Get the preset information push method;

[0044] Based on the information push method, the reminder information is sent to the user.

[0045] In order to solve the above technical problems, the embodiment of the present application also provides a fraudulent call identification device, which adopts the following technical solution:

[0046] A first acquisition module is used to acquire the user's call data; wherein the call data includes audio data and metadata;

[0047] A first extraction module, used to perform feature extraction processing on the audio data and the metadata respectively to obtain corresponding audio features and metadata features;

[0048] An analysis module, used to perform fraud analysis on the audio features and the metadata features based on a preset telephone analysis model to obtain corresponding fraud analysis results;

[0049] a conversion module, configured to convert the audio data into corresponding text data if the fraud analysis result indicates that the call corresponding to the call data is a suspected fraud call;

[0050] A second extraction module is used to extract semantic features from the text data to obtain corresponding semantic features;

[0051] A fusion module, used for fusing the audio feature, the metadata feature and the semantic feature based on a preset fusion strategy to obtain corresponding comprehensive features;

[0052] The identification module is used to identify fraud based on the comprehensive features based on a preset phone recognition model to obtain a fraud call identification result corresponding to the call.

[0053] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0054] Acquire the user's call data; wherein the call data includes audio data and metadata;

[0055] Performing feature extraction processing on the audio data and the metadata respectively to obtain corresponding audio features and metadata features;

[0056] Perform fraud analysis on the audio features and the metadata features based on a preset telephone analysis model to obtain corresponding fraud analysis results;

[0057] If the fraud analysis result is that the call corresponding to the call data is a suspected fraud call, converting the audio data into corresponding text data;

[0058] Extracting semantic features from the text data to obtain corresponding semantic features;

[0059] The audio feature, the metadata feature and the semantic feature are fused based on a preset fusion strategy to obtain corresponding comprehensive features;

[0060] The comprehensive features are used to identify fraud based on a preset phone identification model to obtain a fraud phone identification result corresponding to the call.

[0061] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0062] Acquire the user's call data; wherein the call data includes audio data and metadata;

[0063] Performing feature extraction processing on the audio data and the metadata respectively to obtain corresponding audio features and metadata features;

[0064] Perform fraud analysis on the audio features and the metadata features based on a preset telephone analysis model to obtain corresponding fraud analysis results;

[0065] If the fraud analysis result is that the call corresponding to the call data is a suspected fraud call, converting the audio data into corresponding text data;

[0066] Extracting semantic features from the text data to obtain corresponding semantic features;

[0067] The audio feature, the metadata feature and the semantic feature are fused based on a preset fusion strategy to obtain corresponding comprehensive features;

[0068] The comprehensive features are used to identify fraud based on a preset phone identification model to obtain a fraud phone identification result corresponding to the call.

[0069] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0070] The present application first obtains the user's call data; wherein the call data includes audio data and metadata; then, feature extraction is performed on the audio data and the metadata respectively to obtain corresponding audio features and metadata features; then, based on a preset phone analysis model, fraud analysis is performed on the audio features and the metadata features to obtain corresponding fraud analysis results; if the fraud analysis result is that the call corresponding to the call data is a suspected fraud call, the audio data is converted into corresponding text data; subsequently, semantic feature extraction is performed on the text data to obtain corresponding semantic features; further, based on a preset fusion strategy, the audio features, the metadata features and the semantic features are fused to obtain corresponding comprehensive features; finally, fraud identification is performed on the comprehensive features based on a preset phone recognition model to obtain a fraud call recognition result corresponding to the call. The present application obtains audio features and metadata features by performing feature extraction processing on the acquired user's call data, and then intelligently performs preliminary fraud analysis on the audio features and metadata features based on the use of a telephone analysis model. When it is detected that the fraud analysis result is that the call corresponding to the call data is a suspected fraud call, the audio data will be further converted into corresponding text data, and semantic features will be extracted from the text data to obtain semantic features. The audio features, metadata features and semantic features are then fused based on a fusion strategy to obtain comprehensive features. Subsequently, the comprehensive features will be used for accurate fraud identification of the call based on the use of a telephone recognition model, thereby achieving rapid and accurate generation of fraud call identification results corresponding to the call, effectively improving the accuracy and processing efficiency of telephone fraud identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0072] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0073] Figure 2 A flowchart of an embodiment of a method for identifying fraudulent calls according to the present application;

[0074] Figure 3 It is a structural schematic diagram of an embodiment of a fraud call identification device according to the present application;

[0075] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0076] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0077] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0078] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0079] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0080] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0081] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to the laptop computer 1011, tablet computer 1012 or mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0082] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0083] It should be noted that the method for identifying fraudulent calls provided in the embodiments of the present application is generally executed by a server / terminal device, and accordingly, the device for identifying fraudulent calls is generally set in the server / terminal device.

[0084] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0085] Continue to refer Figure 2 , a flowchart of an embodiment of a method for identifying fraudulent calls according to the present application is shown. According to different requirements, the order of the steps in the flowchart can be changed, and some steps can be omitted. The method for identifying fraudulent calls provided in the embodiment of the present application can be applied to any scenario where fraudulent call identification is required, and the method for identifying fraudulent calls can be applied to products in these scenarios, for example, fraudulent call identification in the financial insurance field. The method for identifying fraudulent calls includes the following steps:

[0086] Step S201, obtaining the user's call data; wherein the call data includes audio data and metadata.

[0087] In this embodiment, the fraud call identification method is executed on the electronic device (e.g. Figure 1The server / terminal device shown in the figure) can obtain call data through a wired connection or a wireless connection. It should be pointed out that the above-mentioned wireless connection methods may include but are not limited to 3G / 4G / 5G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or developed in the future. The executor of this application may be a fraudulent call identification system, or simply referred to as the system. This application can be applied to call identification scenarios in the financial field. The system can receive call data from service customers in financial enterprises through a pre-established data receiving interface. Call data includes audio data and metadata (such as call time, call duration, call frequency, calling and called numbers, etc.).

[0088] Step S202: performing feature extraction processing on the audio data and the metadata respectively to obtain corresponding audio features and metadata features.

[0089] In this embodiment, the above-mentioned specific implementation process of performing feature extraction processing on the audio data and the metadata respectively to obtain corresponding audio features and metadata features will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated in detail here.

[0090] Step S203: performing fraud analysis on the audio features and the metadata features based on a preset telephone analysis model to obtain corresponding fraud analysis results.

[0091] In this embodiment, the above-mentioned telephone analysis model is a model obtained by training the preliminary classification model using pre-built call sample data. The telephone analysis model will predict based on the input audio features and metadata features and output the prediction result, that is, whether the call is a fraud call. And according to the prediction results, the calls are divided into two categories: normal calls and suspected fraud calls. Among them, a threshold value (such as 0.5) can be set. When the predicted probability exceeds the threshold, the call is marked as a suspected fraud call, that is, a fraud analysis result is generated that the call is a suspected fraud call.

[0092] Specifically, by selecting a suitable preliminary classification model, such as support vector machine (SVM), random forest, etc., and using historical call data extracted within a historical time period, and using the historical audio features and historical metadata features of the extracted historical call data as call sample data, and using the call sample data to train the above preliminary classification model, by adjusting the parameters of the model, such as the C parameter and kernel function parameter in SVM, or the number and depth of trees in random forest, until a telephone analysis model that can specifically predict whether a call is a suspected fraud call (yes / no) and output the corresponding probability value is constructed. Among them, there is no specific limitation on the selection of the above historical time period, for example, it can be used within the past year.

[0093] In addition, if the fraud analysis result output by the telephone analysis model is that the call corresponding to the call data is a normal call, then the post-processing of fraud identification based on the comprehensive features of the preset telephone recognition model will not be performed subsequently. Instead, the above call will be directly treated as a normal call for processing, thereby effectively reducing the processing workload of the telephone recognition model for fraud identification based on the comprehensive features, and thus improving the processing efficiency of telephone fraud identification.

[0094] Step S204: If the fraud analysis result is that the call corresponding to the call data is a suspected fraud call, the audio data is converted into corresponding text data.

[0095] In this embodiment, the above audio data may be converted into corresponding text data by using speech recognition technology (such as ASR, automatic speech recognition).

[0096] Step S205: extracting semantic features from the text data to obtain corresponding semantic features.

[0097] In the present embodiment, the above-mentioned specific implementation process of extracting semantic features from the text data to obtain corresponding semantic features will be described in further detail in subsequent specific embodiments of the present application, and will not be elaborated on here.

[0098] Step S206: fusing the audio features, the metadata features, and the semantic features based on a preset fusion strategy to obtain corresponding comprehensive features.

[0099] In this embodiment, the above-mentioned specific implementation process of fusing the audio features, the metadata features and the semantic features based on the preset fusion strategy to obtain the corresponding comprehensive features will be further described in detail in the subsequent specific embodiments of this application and will not be elaborated in detail here.

[0100] Step S207, performing fraud identification on the comprehensive features based on a preset phone identification model to obtain a fraud call identification result corresponding to the call.

[0101] In this embodiment, the above-mentioned specific implementation process of performing fraud identification on the comprehensive features based on the preset phone identification model to obtain the fraud phone identification result corresponding to the call will be further described in detail in the subsequent specific embodiments of this application and will not be elaborated in detail here.

[0102] The present application obtains audio features and metadata features by performing feature extraction processing on the acquired user's call data, and then intelligently performs preliminary fraud analysis on the audio features and metadata features based on the use of a telephone analysis model. When it is detected that the fraud analysis result is that the call corresponding to the call data is a suspected fraud call, the audio data will be further converted into corresponding text data, and semantic features will be extracted from the text data to obtain semantic features. The audio features, metadata features and semantic features are then fused based on a fusion strategy to obtain comprehensive features. Subsequently, the comprehensive features will be used for accurate fraud identification of the call based on the use of a telephone recognition model, thereby achieving rapid and accurate generation of fraud call identification results corresponding to the call, effectively improving the accuracy and processing efficiency of telephone fraud identification.

[0103] In some optional implementations, step S202 includes the following steps:

[0104] The audio data is preprocessed to obtain corresponding target audio data.

[0105] In this embodiment, the above preprocessing includes denoising and segmenting the audio data. Specifically, a professional audio processing tool or library can be used to denoise the audio data to reduce the impact of background noise on feature extraction. In addition, the audio data is segmented to cut long audio segments into short segments or frames that are easier to process.

[0106] Feature extraction is performed on the target audio data based on the pre-trained model to obtain corresponding audio features.

[0107] In this embodiment, the above-mentioned pre-trained model can specifically adopt the WavLM model. By using the WavLM model to extract features of the pre-processed target audio data, the obtained audio features may include Mel-frequency cepstral coefficients (MFCC), spectrum features, energy features, etc. Among them, the feature extraction process should ensure that the extracted features can fully reflect the speech characteristics and content of the audio.

[0108] The metadata is cleaned to obtain corresponding target metadata.

[0109] In this embodiment, metadata in the call data is cleaned to remove invalid or redundant data.

[0110] Perform feature engineering processing on the target metadata to obtain corresponding metadata features.

[0111] In this embodiment, a feature engineering process can be designed according to the needs of telephone fraud identification to extract useful metadata features such as call time, call duration, call frequency, and called number. In addition, the extracted features can be further standardized or normalized to ensure that the dimensions of different features are consistent.

[0112] The present application obtains the corresponding target audio data by preprocessing the audio data; then extracts features from the target audio data based on a pretrained model to obtain corresponding audio features; then cleans the metadata to obtain corresponding target metadata; and subsequently performs feature engineering on the target metadata to obtain corresponding metadata features. The present application obtains the target audio data by preprocessing the audio data, and then extracts features from the target audio data based on the use of a pretrained model to obtain corresponding audio features. Also, by performing a cleaning process on the metadata to obtain target metadata, and then performing feature engineering on the target metadata to obtain corresponding metadata features, it is possible to quickly and accurately complete feature extraction processing of audio data and metadata, improve the processing efficiency of feature extraction processing, and ensure the data accuracy and standardization of the obtained audio features and metadata features.

[0113] In some optional implementations of this embodiment, step S205 includes the following steps:

[0114] The text data is preprocessed to obtain corresponding target text data.

[0115] In this embodiment, the preprocessed target text data is obtained by preprocessing the text data, including removing irrelevant information such as stop words, punctuation marks, and numbers, and performing stem extraction or word form restoration.

[0116] Based on a preset sentiment analysis algorithm, sentiment tendency analysis is performed on the target text data to obtain corresponding sentiment features.

[0117] In this embodiment, the sentiment analysis algorithm can be a sentiment classifier based on machine learning or deep learning. The sentiment analysis algorithm is used to judge the sentiment tendency of the target text data, and then the corresponding sentiment features, such as positive, negative, neutral, etc., are extracted.

[0118] The target text data is subjected to intent recognition based on a preset intent recognition model to obtain corresponding intent features.

[0119] In this embodiment, the above-mentioned intention recognition model can be a pre-trained machine learning model or deep learning model with intention recognition function. The intention recognition model is used to classify the target text data to extract the intention features, such as sales promotion, threat, inquiry, etc.

[0120] The semantic feature is generated based on the emotion feature and the intention feature.

[0121] In this embodiment, the emotional feature and the intention feature are integrated to obtain an integrated feature, and the integrated feature is used as the semantic feature.

[0122] The present application obtains the corresponding target text data by preprocessing the text data; then performs sentiment tendency analysis on the target text data based on a preset sentiment analysis algorithm to obtain corresponding sentiment features; then performs intent recognition on the target text data based on a preset intent recognition model to obtain corresponding intent features; and subsequently generates the semantic features based on the sentiment features and the intent features. The present application obtains the target text data by preprocessing the text data; then performs sentiment tendency analysis on the target text data based on a sentiment analysis algorithm to obtain sentiment features, and performs intent recognition on the target text data based on an intent recognition model to obtain intent features, thereby enabling rapid and accurate generation of corresponding semantic features based on the sentiment features and the intent features, thereby improving the processing efficiency of semantic feature extraction of text data and ensuring data accuracy and richness of the obtained semantic features.

[0123] In some optional implementations, step S206 includes the following steps:

[0124] The audio features, the metadata features, and the semantic features are respectively normalized to obtain processed designated audio features, designated metadata features, and designated semantic features.

[0125] In this embodiment, audio features (such as MFCC, spectrum features, etc.), metadata features (such as call time, duration, etc.) and semantic features (such as emotion features, intent features, etc.) are normalized respectively to ensure that they are comparable in value.

[0126] The specified audio feature, the specified metadata feature, and the specified semantic feature are concatenated to obtain corresponding multimodal features.

[0127] In this embodiment, a comprehensive multimodal feature is formed by splicing the processed designated audio features, designated metadata features, and designated semantic features, wherein the splicing may be weighted according to the dimension and importance of the features.

[0128] Perform dimensionality reduction processing on the multimodal features to obtain processed designated features.

[0129] In this embodiment, if the dimension of the feature vector is too high, it may lead to an increase in computational complexity and storage cost. In this case, feature dimensionality reduction technology (such as PCA, principal component analysis) can be used to reduce the dimensionality of the above multimodal features to obtain the specified features.

[0130] The designated feature is used as the comprehensive feature.

[0131] The present application performs normalization processing on the audio features, metadata features and semantic features respectively to obtain the processed designated audio features, designated metadata features and designated semantic features; then splices the designated audio features, designated metadata features and designated semantic features to obtain the corresponding multimodal features; then performs dimensionality reduction processing on the multimodal features to obtain the processed designated features; and subsequently uses the designated features as the comprehensive features. The present application performs normalization processing on the audio features, metadata features and semantic features respectively to obtain the processed designated audio features, designated metadata features and designated semantic features; then splices and performs dimensionality reduction processing on the obtained designated audio features, designated metadata features and designated semantic features, so as to achieve efficient and accurate feature fusion processing of multiple features, improve the processing efficiency of feature fusion processing, and ensure the accuracy and diversity of the generated comprehensive features.

[0132] In some optional implementations, step S207 includes the following steps:

[0133] The comprehensive features are inferred based on the telephone recognition model to calculate the probability that the call is a fraudulent call.

[0134] In this embodiment, the comprehensive features generated by the fusion process are input into a trained call recognition model. The call recognition model will perform reasoning based on the input comprehensive features to calculate the probability that the call is a fraudulent call.

[0135] It is determined whether the probability is greater than a preset probability threshold.

[0136] In this embodiment, there is no specific limitation on the value of the above probability threshold, which can be set according to actual business needs, for example, it can be set to 0.8.

[0137] If so, a first fraudulent call identification result is generated indicating that the call is a fraudulent call.

[0138] In this embodiment, if the probability that the call is a fraudulent call output by the telephone identification model is greater than the above-mentioned probability threshold, the call is determined to be a fraudulent call, and a first fraudulent call identification result that the call is a fraudulent call is generated.

[0139] If not, a second fraud call identification result is generated indicating that the call is a normal call.

[0140] In this embodiment, if the probability output by the telephone identification model that the call is a fraudulent call is less than the above probability threshold, the call is determined to be a normal call, and a second fraudulent call identification result that the call is a normal call is generated.

[0141] In addition, the fraudulent call identification results output by the call identification model can also be explained to explain the basis for each call being judged as a fraudulent call. Specifically, it can involve the analysis of comprehensive features and how the call identification model weighs the importance of different features during the reasoning process.

[0142] In addition, the construction process of the phone recognition model includes: according to the complexity of the problem and the characteristics of the data, selecting a suitable deep learning model, such as a convolutional neural network (CNN) for processing audio features, a recurrent neural network (RNN) or a long short-term memory network (LSTM) for processing time series data (such as call time series in metadata features), and an attention mechanism for enhancing the model's ability to capture key information. Then, the pre-constructed training set containing multimodal feature vectors (including historical audio features, historical metadata features, and historical semantic features of the above-mentioned historical call data) is used as input to train the deep learning model. During the training process, cross-validation and other techniques can be used to evaluate the performance of the model and adjust the model parameters to optimize the recognition accuracy. The trained deep learning model is further evaluated using a test data set (which can be composed of any proportion of data selected from the training set), including calculating indicators such as accuracy, recall, and F1 score to measure the performance of the model, until a phone recognition model with the function of accurately predicting whether a call is a fraudulent call (yes / no) and outputting the corresponding probability of the call being a fraudulent call is constructed.

[0143] Furthermore, the call recognition model can be optimized based on real-time feedback and expert annotations. Specifically, first collect real-time feedback data from users in actual use and expert annotation data on suspicious calls. These data should contain correct labels (such as whether it is a fraud call) and relevant feature information. Then preprocess the collected data, including removing duplicate data, processing missing values, and performing feature engineering. Then use online learning algorithms (such as SGD, Adam, etc.) to update the call recognition model. Online learning algorithms can fine-tune model parameters based on new data without retraining the entire model. This ensures that the call recognition model can quickly adapt to new fraud methods and features. The performance of the call recognition model can also be monitored regularly in the future, including indicators such as recognition accuracy and response time. If performance degradation is found, the model should be optimized or retrained in a timely manner. Model optimization of the call recognition model through real-time feedback and expert annotation can ensure that the call recognition model can continue to adapt to new fraud methods and features.

[0144] The present application performs reasoning processing on the comprehensive features based on the phone recognition model to calculate the probability that the call is a fraudulent call; then determines whether the probability is greater than a preset probability threshold; if so, generates a first fraudulent call recognition result that the call is a fraudulent call; if not, generates a second fraudulent call recognition result that the call is a normal call. The present application performs reasoning processing on the comprehensive features based on the use of the phone recognition model to calculate the probability that the call is a fraudulent call, and then compares the obtained probability with the preset probability threshold, and then based on the obtained comparison result, it can quickly and accurately generate a fraudulent call recognition result corresponding to the call, thereby improving the accuracy and intelligence of the generated fraudulent call recognition result.

[0145] In some optional implementations of this embodiment, after step S207, the electronic device may further perform the following steps:

[0146] Obtain specified multimodal features of identified fraudulent calls; wherein the number of the fraudulent calls includes multiple.

[0147] In this embodiment, the identified fraudulent calls can be selected by the phone recognition model after performing fraud recognition on multiple call data to identify the calls with fraud risks. The specified multimodal features may include audio features, metadata features, and semantic features corresponding to the call data.

[0148] Call the preset target clustering algorithm.

[0149] In this embodiment, a suitable clustering algorithm can be selected as the target clustering algorithm according to the characteristics of the data and the requirements of the problem. Preferably, the target clustering algorithm can be HDBSCAN algorithm, which is a density-based clustering algorithm suitable for processing clusters with different densities and shapes.

[0150] Based on the target clustering algorithm, a cluster analysis is performed on the designated multimodal features of the fraudulent calls to obtain corresponding clustering results.

[0151] In this embodiment, a feature vector is formed by merging the specified multimodal features of fraudulent calls. Then, the corresponding parameters are set according to the characteristics of the data and the requirements of the clustering algorithm. For example, in HDBSCAN, parameters such as the minimum cluster size (min_cluster_size) and the minimum number of samples (min_samples) need to be set. Subsequently, the target clustering algorithm is used to perform cluster analysis on the above feature vectors, analyze the clustering results, identify different clusters, and understand the type and pattern of fraudulent calls represented by each cluster.

[0152] The clustering results are analyzed and processed based on a preset analysis strategy to obtain corresponding analysis results.

[0153] In this embodiment, the trend analysis can be performed on the clustering results, that is, by observing the changes in the number, size and shape of clusters, the trend and changes of fraudulent calls can be analyzed. And according to the clustering results, different types of fraudulent calls can be further classified and analyzed, such as segmenting according to fraudulent methods, target groups, etc., and the results can be integrated to obtain corresponding analysis results.

[0154] The clustering result and the analysis result are outputted.

[0155] In this embodiment, the generated clustering results and analysis results can be output in the form of a report, including the number of clusters, the characteristic description of each cluster, the trend of fraudulent calls, etc. In addition, corresponding prevention strategies can be formulated based on the clustering results, such as strengthening the monitoring and early warning of specific types of fraudulent calls.

[0156] The present application obtains the designated multimodal features of the identified fraudulent calls; wherein the number of the fraudulent calls includes multiple; then calls a preset target clustering algorithm; then performs cluster analysis on the designated multimodal features of the fraudulent calls based on the target clustering algorithm to obtain corresponding clustering results; subsequently analyzes and processes the clustering results based on a preset analysis strategy to obtain corresponding analysis results; finally outputs the clustering results and the analysis results. The present application obtains the designated multimodal features of the identified fraudulent calls, and then intelligently performs cluster analysis on the designated multimodal features of the fraudulent calls based on the use of the target clustering algorithm to obtain clustering results, and further analyzes the clustering results based on the use of the analysis strategy to obtain analysis results, and subsequently outputs the obtained clustering results and analysis results, which is conducive to relevant personnel being able to efficiently and accurately formulate corresponding prevention strategies by consulting the clustering results and analysis results, thereby improving the prevention effect against fraudulent calls and improving the work efficiency and work experience of relevant personnel.

[0157] In some optional implementations of this embodiment, after step S207, the electronic device may further perform the following steps:

[0158] If the fraud call identification result is that the call is a fraud call, corresponding warning processing is performed on the call.

[0159] In this embodiment, when the fraud call identification result is detected that the call is a fraud call, the system will immediately take early warning measures, which may include blocking the call from continuing to prevent the user from being further deceived.

[0160] Generate reminder information corresponding to the call.

[0161] In this embodiment, the content of the above reminder information may include: reminding the user that the above call is a fraud call, reminding the user to remain vigilant and avoid leaking personal information or performing any form of transfer operation.

[0162] Get the preset information push method.

[0163] In this embodiment, the above-mentioned information push method may include text messages, phone calls or other methods.

[0164] Based on the information push method, the reminder information is sent to the user.

[0165] In this embodiment, the system can display the reminder information to the user in a clear and intuitive manner. This can be achieved through a user interface. The format of the output result should be customized according to the needs of the recipient. For example, for the user, a concise and clear text or graphical interface can be used.

[0166] In addition, the system can also record the relevant information of fraudulent calls in detail, including the caller number, call time, fraud type, etc., and generate corresponding reports for subsequent analysis and processing. Specifically, the system automatically generates visual reports and analysis results corresponding to fraudulent calls. These reports can include statistical data, trend analysis, geographical distribution and other information of fraudulent calls. Among them, in order to more intuitively display the analysis results, various chart types can be used, such as bar charts, line charts, pie charts, etc. These charts can clearly show the situation and trend of fraudulent calls and help users better understand the problem. In addition, the generated reports can be distributed to relevant users or departments. This can be distributed through email, internal systems or other means. At the same time, the system can also provide report download and printing functions to meet the needs of different users.

[0167] In addition, if the fraud call identification result is that the call is a normal call, the system will treat it as a normal call. According to business needs, the system can choose to store the call records in the database for subsequent query and analysis, or simply ignore the call and do not perform any further operations. In addition, the system will also record log information related to the above calls, including call time, call duration, identification results, etc., for subsequent auditing and tracking.

[0168] If the present application detects that the fraud call identification result is that the call is a fraud call, the corresponding early warning processing is performed on the call; then a reminder message corresponding to the call is generated; thereafter, a preset information push method is obtained; and subsequently, based on the information push method, the reminder message is sent to the user. When the present application detects that the fraud call identification result is that the call is a fraud call, the corresponding early warning processing is intelligently performed on the call to prevent the user from being further deceived. In addition, reminder information corresponding to the call is automatically generated, and the reminder information is sent to the user based on the preset information push method, which helps the user to better understand the situation of fraud calls, thereby taking more effective preventive measures and improving the user's experience.

[0169] In some optional implementations of this embodiment, the present application also provides a cross-platform collaborative defense strategy. The cross-platform collaborative defense strategy achieves information sharing and joint defense among different financial institutions by establishing a secure data sharing and analysis platform. The platform uses encryption technology and security protocols to ensure secure sharing of data.

[0170] The implementation steps include:

[0171] 1. Data sharing platform construction: Build a secure data sharing platform that supports data exchange between different financial institutions.

[0172] 2. Application of encryption technology: Apply encryption technology to ensure the protection of user privacy and data security during data sharing.

[0173] 3. Collaborative defense mechanism design: Design a collaborative defense mechanism that allows financial institutions to share information about fraudulent calls and take joint defense measures.

[0174] 4. Data analysis and sharing: Conduct data analysis on the data sharing platform and share fraud call information based on the analysis results to improve overall defense capabilities.

[0175] Through the above steps, the efficiency and accuracy of telephone fraud identification can be improved, and the system's adaptability and user interactivity can be enhanced, providing financial institutions with a new telephone fraud defense solution.

[0176] In addition, this application has great practical significance and far-reaching strategic value for financial enterprises. In the insurance business, this application can significantly improve the ability to identify and defend against telephone fraud, protect customers from economic losses, and enhance customers' trust in the services of financial enterprises. Through multimodal fusion technology, fraudulent calls can be accurately identified and classified, and potential fraud can be warned and prevented in a timely manner, thereby reducing the occurrence of insurance fraud cases and reducing the compensation costs of financial enterprises. The real-time feedback learning mechanism and cross-platform collaborative defense strategy further enhance the competitiveness of financial enterprises in the field of financial security. By sharing fraud information with partners, a more solid security line of defense is built, providing solid technical support for the sound operation and sustainable development of financial enterprises.

[0177] In some optional implementations, the user information obtained is subject to the user's consent and complies with relevant laws and policies.

[0178] In addition, the non-Company software tools or components that appear in the embodiments of the present application are merely examples and do not represent actual use.

[0179] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0180] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned fraudulent call identification results, the above-mentioned fraudulent call identification results can also be stored in a node of a blockchain.

[0181] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0182] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0183] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0184] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0185] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0186] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a fraud call identification device, which is similar to Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0187] like Figure 3 As shown, the fraud call identification device 300 described in this embodiment includes: a first acquisition module 301, a first extraction module 302, an analysis module 303, a conversion module 304, a second extraction module 305, a fusion module 306 and an identification module 307. Among them:

[0188] The first acquisition module 301 is used to acquire the user's call data; wherein the call data includes audio data and metadata;

[0189] A first extraction module 302, configured to perform feature extraction processing on the audio data and the metadata respectively to obtain corresponding audio features and metadata features;

[0190] An analysis module 303, configured to perform fraud analysis on the audio features and the metadata features based on a preset telephone analysis model to obtain corresponding fraud analysis results;

[0191] A conversion module 304, configured to convert the audio data into corresponding text data if the fraud analysis result indicates that the call corresponding to the call data is a suspected fraud call;

[0192] The second extraction module 305 is used to extract semantic features from the text data to obtain corresponding semantic features;

[0193] A fusion module 306, configured to fuse the audio feature, the metadata feature and the semantic feature based on a preset fusion strategy to obtain a corresponding comprehensive feature;

[0194] The identification module 307 is used to identify fraud based on the comprehensive features based on a preset phone recognition model to obtain a fraud call identification result corresponding to the call.

[0195] In some optional implementations of this embodiment, the first extraction module 302 includes:

[0196] A first preprocessing submodule, used to preprocess the audio data to obtain corresponding target audio data;

[0197] An extraction submodule, used to extract features from the target audio data based on a pre-trained model to obtain corresponding audio features;

[0198] A cleaning submodule, used for cleaning the metadata to obtain corresponding target metadata;

[0199] The first processing submodule is used to perform feature engineering processing on the target metadata to obtain corresponding metadata features.

[0200] In some optional implementations of this embodiment, the second extraction module 305 includes:

[0201] A second preprocessing submodule, used for preprocessing the text data to obtain corresponding target text data;

[0202] The first analysis submodule is used to perform sentiment tendency analysis on the target text data based on a preset sentiment analysis algorithm to obtain corresponding sentiment features;

[0203] The recognition submodule is used to perform intent recognition on the target text data based on a preset intent recognition model to obtain corresponding intent features;

[0204] The first generating submodule is used to generate the semantic feature based on the emotional feature and the intention feature.

[0205] In some optional implementations of this embodiment, the fusion module 306 includes:

[0206] A second processing submodule is used to perform normalization processing on the audio feature, the metadata feature and the semantic feature respectively to obtain a processed specified audio feature, a specified metadata feature and a specified semantic feature;

[0207] A splicing submodule, used for splicing the specified audio feature, the specified metadata feature and the specified semantic feature to obtain corresponding multimodal features;

[0208] A Jiangwei submodule, used for performing dimensionality reduction processing on the multimodal features to obtain processed designated features;

[0209] A determination submodule is used to use the specified feature as the comprehensive feature.

[0210] In some optional implementations of this embodiment, the identification module 307 includes:

[0211] An inference submodule, for performing inference processing on the comprehensive features based on the phone recognition model to calculate the probability that the call is a fraudulent call;

[0212] A judgment submodule, used to judge whether the probability is greater than a preset probability threshold;

[0213] A second generating submodule is used to generate a first fraudulent call identification result that the call is a fraudulent call if the call is a fraudulent call;

[0214] The third generating submodule is used to generate a second fraud call identification result indicating that the call is a normal call if no.

[0215] In some optional implementations of this embodiment, the device for identifying fraudulent calls further includes:

[0216] A second acquisition module is used to acquire a specified multimodal feature of the identified fraudulent phone calls; wherein the number of the fraudulent phone calls includes a plurality;

[0217] A calling module is used to call a preset target clustering algorithm;

[0218] A clustering module, used to perform cluster analysis on the specified multimodal features of the fraudulent calls based on the target clustering algorithm to obtain corresponding clustering results;

[0219] A processing module, used to analyze and process the clustering results based on a preset analysis strategy to obtain corresponding analysis results;

[0220] An output module is used to output the clustering result and the analysis result.

[0221] In some optional implementations of this embodiment, the device for identifying fraudulent calls further includes:

[0222] An early warning module, configured to execute corresponding early warning processing on the call if the fraud call identification result is that the call is a fraud call;

[0223] A generating module, used for generating reminder information corresponding to the call;

[0224] The third acquisition module is used to obtain a preset information push method;

[0225] A sending module is used to send the reminder information to the user based on the information push method.

[0226] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0227] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components 41-43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0228] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0229] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk equipped on the computer device 4, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of a method for identifying fraudulent calls, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0230] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run computer-readable instructions or process data stored in the memory 41, such as computer-readable instructions for running the fraud call identification method.

[0231] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0232] The present application also provides another embodiment, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the method for identifying fraudulent calls as described above.

[0233] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0234] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A method for identifying fraudulent calls, characterized in that: The steps include: Acquire the user's call data; wherein the call data includes audio data and metadata; Performing feature extraction processing on the audio data and the metadata respectively to obtain corresponding audio features and metadata features; Perform fraud analysis on the audio features and the metadata features based on a preset telephone analysis model to obtain corresponding fraud analysis results; If the fraud analysis result is that the call corresponding to the call data is a suspected fraud call, converting the audio data into corresponding text data; Extracting semantic features from the text data to obtain corresponding semantic features; The audio feature, the metadata feature and the semantic feature are fused based on a preset fusion strategy to obtain corresponding comprehensive features; The comprehensive features are used to identify fraud based on a preset phone identification model to obtain a fraud phone identification result corresponding to the call.

2. The method for identifying fraudulent calls according to claim 1, characterized in that: The step of performing feature extraction processing on the audio data and the metadata respectively to obtain corresponding audio features and metadata features specifically includes: Preprocessing the audio data to obtain corresponding target audio data; Extracting features from the target audio data based on the pre-trained model to obtain corresponding audio features; Cleaning the metadata to obtain corresponding target metadata; Perform feature engineering processing on the target metadata to obtain corresponding metadata features.

3. The method for identifying fraudulent calls according to claim 1, characterized in that: The step of extracting semantic features from the text data to obtain corresponding semantic features specifically includes: Preprocessing the text data to obtain corresponding target text data; Performing sentiment analysis on the target text data based on a preset sentiment analysis algorithm to obtain corresponding sentiment features; Performing intent recognition on the target text data based on a preset intent recognition model to obtain corresponding intent features; The semantic feature is generated based on the emotion feature and the intention feature.

4. The method for identifying fraudulent calls according to claim 1, characterized in that: The step of fusing the audio features, the metadata features, and the semantic features based on a preset fusion strategy to obtain corresponding comprehensive features specifically includes: Normalizing the audio features, the metadata features, and the semantic features respectively to obtain processed designated audio features, designated metadata features, and designated semantic features; The specified audio feature, the specified metadata feature and the specified semantic feature are concatenated to obtain corresponding multimodal features; Performing dimensionality reduction processing on the multimodal features to obtain processed designated features; The designated feature is used as the comprehensive feature.

5. The method for identifying fraudulent calls according to claim 1, characterized in that: The step of performing fraud identification on the comprehensive features based on a preset phone identification model to obtain a fraud phone identification result corresponding to the call specifically includes: Performing reasoning processing on the comprehensive features based on the phone recognition model to calculate the probability that the call is a fraudulent call; Determining whether the probability is greater than a preset probability threshold; If so, generating a first fraud call identification result indicating that the call is a fraud call; If not, a second fraud call identification result is generated indicating that the call is a normal call.

6. The method for identifying fraudulent calls according to claim 1, characterized in that: After the step of performing fraud identification on the comprehensive features based on the preset phone identification model to obtain a fraud phone identification result corresponding to the call, the method further includes: Obtaining specified multimodal features of the identified fraudulent calls; wherein the number of the fraudulent calls includes a plurality; Call the preset target clustering algorithm; Performing cluster analysis on the designated multimodal features of the fraudulent calls based on the target clustering algorithm to obtain corresponding clustering results; Analyze and process the clustering results based on a preset analysis strategy to obtain corresponding analysis results; The clustering result and the analysis result are outputted.

7. The method for identifying fraudulent calls according to claim 1, characterized in that: After the step of performing fraud identification on the comprehensive features based on the preset phone identification model to obtain a fraud phone identification result corresponding to the call, the method further includes: If the fraud call identification result is that the call is a fraud call, performing corresponding warning processing on the call; generating a reminder message corresponding to the call; Get the preset information push method; Based on the information push method, the reminder information is sent to the user.

8. A device for identifying fraudulent calls, characterized in that: include: A first acquisition module is used to acquire the user's call data; wherein the call data includes audio data and metadata; A first extraction module, used to perform feature extraction processing on the audio data and the metadata respectively to obtain corresponding audio features and metadata features; An analysis module, used to perform fraud analysis on the audio features and the metadata features based on a preset telephone analysis model to obtain corresponding fraud analysis results; a conversion module, configured to convert the audio data into corresponding text data if the fraud analysis result indicates that the call corresponding to the call data is a suspected fraud call; A second extraction module is used to extract semantic features from the text data to obtain corresponding semantic features; A fusion module, used for fusing the audio feature, the metadata feature and the semantic feature based on a preset fusion strategy to obtain corresponding comprehensive features; The identification module is used to identify fraud based on the comprehensive features based on a preset phone recognition model to obtain a fraud call identification result corresponding to the call.

9. A computer device, characterized in that: It includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the method for identifying fraudulent calls as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the method for identifying fraudulent calls as described in any one of claims 1 to 7.

Images