A method, device, and electronic device for determining the priority of outbound call data

Abstract

The present disclosure relates to a method, apparatus and electronic device for determining the priority of outbound call data, which is applied to the communication field to solve the problem of low timeliness of dissuasion caused by making outbound calls in a certain chronological order. The method includes: obtaining a target call recording corresponding to a number to be called; performing semantic analysis on the target call recording to obtain abnormal call information in the target call recording; and determining the call priority of the number to be called according to the abnormal call information.

Description

Technical Field

[0001] The present disclosure relates to the field of communication technologies, and in particular, to a method, an apparatus, and an electronic device for determining the priority of outbound call data. Background Art

[0002] In recent years, with the rapid development of China's financial and communication industries, new types of illegal criminal acts in the telecommunications network have emerged. Among them, telephone fraud is the main means of fraud. In order to reduce the occurrence of telecommunications network fraud cases, when currently making telephone dissuasions to potential telephone fraud victims, the outbound call system sorts the detected potential victims according to the time sequence to form a list of potential victims' numbers, and then calls the potential victims in the order of this number list for dissuasion, so as to protect the property safety of potential victims. However, since some potential victims are in the process of being deceived, making calls in order will result in property losses for this part of potential victims, and there is a problem of weak timeliness in preventing fraud. Summary of the Invention

[0003] In order to solve the above technical problems or at least partially solve the above technical problems, the present disclosure provides a method, an apparatus, and an electronic device for determining the priority of outbound call data.

[0004] In a first aspect, the present disclosure provides a method for determining the priority of outbound call data, the method comprising:

[0005] Obtaining a target call recording corresponding to the number to be called;

[0006] Performing semantic analysis on the target call recording to obtain abnormal call information in the target call recording;

[0007] Determining the call priority of the number to be called according to the abnormal call information.

[0008] Optionally, performing semantic analysis on the target call recording to obtain abnormal call information in the target call recording, including:

[0009] Performing semantic analysis on the target call recording through a natural language processing model, and obtaining a vector matrix corresponding to a plurality of preset abnormal tags output by the model;

[0010] Determining the call priority of the number to be called according to the abnormal call information includes:

[0011] Calculating priority parameters of a plurality of preset abnormal tags according to the vector matrix corresponding to the plurality of preset abnormal tags and preset weight coefficients corresponding to the plurality of preset abnormal tags;

[0012] Determining the call priority of the number to be called according to the parameter ranges in which the priority parameters of the plurality of preset abnormal tags are located.

[0013] Optionally, obtaining a target call recording corresponding to the number to be dialed includes:

[0014] Obtaining multiple call recordings corresponding to the number to be dialed;

[0015] Determining, from the multiple call recordings, a target call recording whose calling number is an abnormal calling number; wherein, the abnormal calling number is a telephone number marked as an abnormal call.

[0016] Optionally, the method further includes:

[0017] Determining the type of person to which the user of the number to be dialed belongs according to the abnormal call information;

[0018] Obtaining a preset voice corresponding to the type of person;

[0019] When making an outbound call to the number to be dialed according to the call priority, playing the preset voice.

[0020] Optionally, performing semantic analysis on the target call recording through a natural language processing model includes:

[0021] Converting the audio information in the target call recording into text information;

[0022] Performing semantic analysis according to the text information to extract abnormal call information in the text information.

[0023] Optionally, converting the audio information in the target call recording into text information includes:

[0024] Performing voice detection on the target call recording to obtain voice audio;

[0025] The first audio of the first target person and the second audio of the second target person included in the voice audio, where the first target person and the second target person are the calling user and the called user of the target call recording respectively;

[0026] Converting both the first audio and the second audio into text information.

[0027] In a second aspect, the present disclosure provides an apparatus for determining the priority of outbound call data, and the apparatus includes:

[0028] An obtaining module, configured to obtain a target call recording corresponding to the number to be dialed;

[0029] A processing module, configured to perform semantic analysis on the target call recording, obtain abnormal call information in the call recording; and determine the call priority of the number to be dialed according to the abnormal call information.

[0030] Optionally, the processing module is specifically configured to perform semantic analysis on the target call recording through a natural language processing model, and obtain a vector matrix corresponding to a plurality of preset exception tags output by the natural language processing model;

[0031] Determining the call priority of the number to be dialed according to the abnormal call information includes:

[0032] Calculating the priority parameters of a plurality of preset exception tags according to the vector matrix corresponding to the plurality of preset exception tags and the preset weight coefficients corresponding to the plurality of preset exception tags;

[0033] Determine the call priority of the number to be dialed according to the parameter range in which the priority parameters of the plurality of preset exception tags are located.

[0034] Optionally, the obtaining module is specifically configured to obtain a plurality of call recordings corresponding to the number to be dialed;

[0035] Determine a target call recording whose calling number is an abnormal calling number from the plurality of call recordings; wherein, the abnormal calling number is a telephone number marked as an abnormal call.

[0036] Optionally, the processing module is further configured to determine the type of person to which the user of the number to be dialed belongs according to the abnormal call information;

[0037] Obtain a preset voice corresponding to the type of person;

[0038] When making an outbound call to the number to be dialed according to the call priority, play the preset voice.

[0039] Optionally, the processing module is specifically configured to convert the audio information in the target call recording into text information;

[0040] Perform semantic analysis on the text information to extract abnormal call information in the text information.

[0041] Optionally, the processing module is specifically configured to perform voice detection on the target call recording to obtain voice audio;

[0042] Identify the first audio of the first target person and the second audio of the second target person from the voice audio, where the first target person and the second target person are the calling user and the called user of the target call recording respectively;

[0043] Convert both the first audio and the second audio into text information.

[0044] In a third aspect, the present disclosure provides an electronic device, which includes: a processor, a memory, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, the method for determining the priority of outbound call data as described in the first aspect is implemented.

[0045] Fourthly, the present disclosure provides a computer-readable storage medium, which includes: a computer program stored on the computer-readable storage medium, and when the computer program is executed by a processor, it implements the method for determining the priority of outbound call data as described in the first aspect.

[0046] Fifthly, the present disclosure provides a computer program product, which includes: when the computer program product runs on a computer, it enables the computer to implement the method for determining the priority of outbound call data as described in the first aspect.

[0047] The technical solutions provided by the embodiments of the present disclosure have the following advantages compared with the prior art:

[0048] The present disclosure takes the phone number of the potential victim as the number to be dialed, obtains the target call recording corresponding to the number to be dialed, and the target call recording is the call recording between the scammer and the potential victim; performs semantic analysis on the target call recording to obtain abnormal call information therein, so as to determine whether the potential victim has been deceived, or determine the scamming means of the scammer, or determine information such as the amount of money deceived by the potential victim, etc.; then, determines the priority of the number to be dialed according to the abnormal call information, and further makes anti-fraud outbound calls to the number to be dialed according to the call priority, thereby improving the timeliness of anti-fraud dissuasion, realizing timely dissuasion of potential victims with a high possibility of being deceived, and protecting the property safety of potential victims. Description of the Drawings

[0049] The drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure.

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the following will briefly introduce the drawings required to be used in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0051] Figure 1 It is a schematic diagram of the application scenario of the method for determining the priority of outbound call data described in the embodiments of the present disclosure;

[0052] Figure 2 It is a schematic flowchart of the method for determining the priority of outbound call data described in the embodiments of the present disclosure;

[0053] Figure 3 It is a schematic diagram of the natural language processing model in the method described in the embodiments of the present disclosure;

[0054] Figure 4Structural diagram of a device for determining the priority of outbound call data according to an embodiment of the present disclosure;

[0055] Figure 5 Structural diagram of an electronic device according to an embodiment of the present disclosure. Detailed implementation manners

[0056] In order to more clearly understand the above objects, features and advantages of the present disclosure, the solutions of the present disclosure will be further described below. It should be noted that, without conflict, the embodiments of the present disclosure and the features in the embodiments may be combined with each other.

[0057] Many specific details are set forth in the following description in order to fully understand the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only a part of the embodiments of the present disclosure, rather than all of the embodiments.

[0058] The prior art generates a number list according to the detected time sequence, where the time is the time when a fraud number makes a call to the number of a potential victim, the fraud number is the calling number, and the number of the potential victim is the called number; the number of the potential victim is stored as the number to be dialed by the outbound call system in the number list, and the numbers of the potential victims are sorted in the number list according to the detection time, and the outbound call system makes calls to the potential victims in the order of the number list of the potential victims for dissuasion. During the process of making calls in order, potential victims with a higher likelihood of being deceived may be deceived because they do not answer the anti-fraud outbound call in time, resulting in the problem of low timeliness of anti-fraud dissuasion.

[0059] To solve the above problems, the present disclosure takes the telephone number of the potential victim as the number to be dialed, obtains the target call recording corresponding to the number to be dialed, where the target call recording is the call recording between the fraudster and the potential victim; performs semantic analysis on the target call recording to obtain the abnormal call information therein, and determines whether the potential victim has been deceived, or determines the fraud means of the fraudster, or determines information such as the amount of money deceived by the potential victim when the abnormal call information is fraud information; then, determines the priority of the number to be dialed according to the abnormal call information, and further performs anti-fraud outbound calls to the number to be dialed according to the call priority, thereby improving the timeliness of anti-fraud dissuasion, realizing timely dissuasion of potential victims with a high risk of being deceived, and protecting the property safety of potential victims.

[0060] The method for determining the priority of outbound call data described in the embodiments of the present disclosure can be applied to a device for determining the priority of outbound call data or an electronic device. Among them, the device for determining the priority of outbound call data can be a functional module and / or functional entity in the electronic device that can implement the method for determining the priority of outbound call data.

[0061] The above electronic devices may include: smart phones (such as Android phones, IOS phones, Windows Phone phones, etc.), tablet computers, handheld computers, laptop computers, video matrices, monitoring platforms, mobile Internet devices (MIDs), wearable devices, servers, for example, cloud servers and other devices. The above are only examples, not an exhaustive list, including but not limited to the above devices.

[0062] To more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the following will briefly introduce the technical terms required for the description of the embodiments or the prior art:

[0063] Outbound refers to: the phone automatically dials out the user's phone number through the computer and plays the pre-recorded voice to the user through the computer. It is an essential part of the modern customer service center system that integrates the computer and the phone.

[0064] Automated Speech Recognition (ASR) is a technology that converts human speech into text. The speech recognition technology includes training, recognition, and distortion measurement. Among them, training is to pre-analyze the speech feature parameters, create a speech template, and store it in the speech parameter library; recognition is to perform the same analysis on the speech to be recognized as during training to obtain the speech parameters, and then compare them one by one with the reference templates in the library, and use the decision method to find the template closest to the speech features to obtain the recognition result; distortion measurement is to calculate the loss during the comparison process.

[0065] Natural Language Processing (NLP) uses a computer to process, understand, and apply human languages (such as Chinese, English, etc.).

[0066] Figure 1 It is a schematic diagram of the application scenario of a method for determining the priority of outbound data in the embodiments of the present disclosure, as Figure 1As shown in the figure, the device in the figure includes an anti-fraud outbound call platform 110, a first terminal 120, and a second terminal 130. Among them, the first terminal 120 is the terminal device of the first potential victim, and the second terminal 130 is the terminal device of the second potential victim. The anti-fraud outbound call platform 110 uses the phone number corresponding to the first terminal 120 and the phone number corresponding to the second terminal 130 as the numbers to be called, obtains the target call recording corresponding to the numbers to be called, further performs semantic analysis on the target call recording to obtain the abnormal call information in the target call record, and then determines the call priority of the numbers to be called according to the abnormal call information. If it is determined that the call priority of the number to be called corresponding to the first terminal 120 is relatively high, then the number to be called corresponding to the first terminal 120 is made an anti-fraud outbound call according to the call priority to perform anti-fraud dissuasion, so as to protect the property safety of potential victims and improve the timeliness of anti-fraud dissuasion. The number of potential victim terminals is not specifically limited in the embodiments of the present disclosure. In the embodiments of the present invention, only two potential victim terminals are used as examples for illustration.

[0067] Figure 2 It is a flowchart of a method for determining the priority of outbound call data in an embodiment of the present disclosure. The method includes:

[0068] S201. Obtain the target call recording corresponding to the number to be called.

[0069] Among them, the number to be called is the phone number of the first target person, and the target call recording is the call recording of the call between the first target person and the second target person. In the present disclosure, the first target person is a fraudster, and the second target person is a potential victim. Usually, the number to be called and the target call recording are batch-imported into the device in the form of table data by the user. This device can apply the method for determining the priority of outbound call data provided in the embodiments of the present disclosure, and multiple numbers to be called are stored in this table. The arrangement of these multiple numbers to be called is determined according to the pre-detected time sequence. The time is the time when it is detected that a fraud number makes a call to a potential victim number, where the fraud number is the calling number and the potential victim number is the called number.

[0070] Exemplarily, the table includes three numbers to be called, T1, T2, and T3. The times when T1, T2, and T3 are detected are t1, t2, and t3 respectively, and t1 < t2 < t3, indicating that it is first detected that the number to be called T1 is the phone number of a potential victim, then T2 is detected, and finally T3. Then the order of the numbers to be called in the table is T1, T2, T3. In the prior art, outbound calls are made in sequence according to the order of T1, T2, and T3.

[0071] To ensure the timeliness of anti-fraud dissuasion, this disclosure needs to determine the priorities of multiple numbers to be called in a table based on the content of the target call recording. In some embodiments, a call recording corresponding to the number to be called is obtained. The call recording includes a call recording of a potential victim communicating with a normal user and a target call recording of the potential victim communicating with a scammer. An implementation method is provided in the embodiments of this disclosure. The call recording is determined as the target call recording based on the calling number being a fraud number. Here, the calling number is the number corresponding to the terminal that performs the call operation, and correspondingly, the called number is the number corresponding to the terminal that performs the answering operation; the abnormal call number is a telephone number marked as abnormal, including but not limited to: a number with more than 90% of the calls being outgoing calls, a number with very few incoming calls, a number with a call duration exceeding 8 hours in a day, an unrecorded number, and an overseas number; the calling number being a fraud number means that the number performing the call operation is a number marked as abnormal call.

[0072] In some embodiments, there are multiple numbers to be called, which are imported into the device by the user. The device can make outgoing calls based on the numbers to be called, such as an anti-fraud outbound call platform, and then stored in the storage medium of the device in the form of a table. Correspondingly, the call recordings corresponding to the numbers to be called are stored in the storage medium of the device. This disclosure does not make specific restrictions on the storage form and storage location of the numbers to be called and the call recordings corresponding to the numbers to be called in the device.

[0073] In the above embodiments, by using the telephone number of the potential victim as the number to be called and obtaining the target call recording corresponding to the number to be called, the recording of the potential victim communicating with the scammer is determined, which is convenient for subsequent analysis based on the target call recording and improves the efficiency of anti-fraud dissuasion.

[0074] S202. Perform semantic analysis on the target call recording to obtain abnormal call information in the target call recording.

[0075] Among them, the semantic analysis process includes but is not limited to: speech recognition, natural language processing; the abnormal call information is represented as a vector matrix corresponding to multiple preset abnormal labels in the device, where the preset abnormal labels are predefined by the user and include but are not limited to amount, fraud type, whether being deceived, etc.

[0076] The following will introduce the speech recognition and natural language processing processes included in the speech analysis:

[0077] (1) Speech recognition

[0078] Speech recognition can be roughly divided into four stages: preprocessing, acoustic feature extraction, speech-to-text conversion through an acoustic model, and syllable-to-text conversion through a language model. Among them, the acoustic model and the language model are two core modules of the speech recognition system model, corresponding to the calculation of the probability from speech to syllables and the calculation of the probability from syllables to words respectively.

[0079] In some embodiments, in the preprocessing stage, since the target call recording includes audio segments and noise segments, it is necessary to perform preliminary denoising on the target call recording to remove the noise segments included in the target call recording. Among them, the noise segments include, but are not limited to, environmental noise and white noise. For example, the noise segments include blank audio and electronic noise generated during the device recording process.

[0080] In the acoustic feature extraction stage, the acoustic feature is the acoustic wave spectrum feature carrying speech information, and the acoustic feature has specificity and relative stability, that is, the acoustic features of different users are specific, and the acoustic features of the same user are relatively stable. Therefore, based on the audio segments obtained after preprocessing, different users' audio segments can be extracted according to the acoustic features of different users. In an implementation manner provided by the embodiments of the present disclosure, according to the different acoustic features of the first target person and the second target person in the target call recording, the first audio segment of the first target person and the second audio segment of the second target person can be determined, where the first target person is a fraudster and is the calling user in the target call recording; the second target person is a potential victim and is the called user in the target call recording.

[0081] Further, in some embodiments, voice activity detection is performed on the target call recording, that is, voice activity detection is performed on the first audio segment of the first target person and the second audio segment of the second target person included therein. Among them, voice activity detection (VAD) is used to identify and eliminate long silent segments from the first audio segment. Voice activity detection methods include, but are not limited to: performing voice activity detection using automatic gain control (AGC) with a speech sense algorithm, and performing voice activity detection using a voice endpoint detector.

[0082] Taking the example of using a voice activity detector (VAD) for human voice detection, for each frame of the voice signal input to the VAD, the VAD scores based on the probability that the voice signal in the first audio segment is a voice frame or a noise frame. When the score value of the voice frame is greater than a pre-set decision threshold, it is determined as a voice frame; otherwise, it is a noise frame. The VAD distinguishes between voice frames and noise frames according to the above decision results to remove the noise frames in the first audio segment. Among them, the decision threshold in this embodiment adopts the default decision threshold in the source code of Web Real-Time Communication (Webrtc). This decision threshold is obtained by analyzing a large amount of data during the development of Webrtc technology to improve the discrimination effect and accuracy, and at the same time reduce the model training workload of the VAD.

[0083] After performing human voice detection on the target call recording, the long silent segments in the target call recording are removed, and the first audio information of the first target person and the second audio information of the second target person included in the target call recording are obtained.

[0084] Furthermore, the first audio information of the first target person and the second audio information of the second target person are converted into corresponding first text information and second text information. In the process of converting the first audio information and the second audio information into the corresponding first text information and second text information, first, the voice is converted into syllables through an acoustic model. Among them, the acoustic model is trained by methods such as deep neural networks (DNN, Deep Neural Networks) after extracting acoustic features. The optimization of the acoustic model depends on a large number of domain-related, rich-content, and accurately annotated audios. The first audio information of the victim and the second audio information of the scammer included in the target call voice are converted into syllables for processing through the acoustic model.

[0085] It should be noted that the recognition result of the acoustic model directly affects the output of the language model, thus affecting the accuracy of the final result. Therefore, it is particularly important to correct the recognition result of the acoustic model. In an embodiment provided in the present disclosure, the syllable information recognized by the acoustic model is corrected. For example, the N-Gram model is used to screen the syllable formation probability of the recognition result of the acoustic model to improve the accuracy of the recognition result of the acoustic model. Another example is to correct the syllables after the acoustic model recognition based on the confusion set and the candidate word library.

[0086] Then, in the stage of converting syllables to text through a language model, the syllables of the first audio information of the scammer and the second audio information of the victim are converted into text to obtain the first text information of the scammer and the second text information of the victim. The language model can also be trained through methods such as deep neural networks, which will not be elaborated in this disclosure.

[0087] In the above embodiment, the noise segment in the target call recording is removed to obtain an audio segment, reducing the computational complexity of subsequent processing and improving the accuracy of speech recognition at the same time; then, the audio segment is subjected to voice detection to obtain the first audio segment of the victim and the second audio segment of the scammer, thereby removing the long silent segments in the audio segment and improving the efficiency of subsequent speech-to-text conversion; the first audio segment of the scammer and the second audio segment of the victim are converted into the corresponding first text information and second text information through an acoustic model and a language model, facilitating semantic analysis and improving the efficiency of obtaining abnormal call information contained in the target call recording.

[0088] (2) Natural Language Processing

[0089] In some embodiments, through a natural language processing model, semantic analysis is performed on the first text information of the scammer and the second text information of the victim obtained after speech recognition. Among them, the natural language processing model includes but is not limited to classical pre-trained language models such as auto-regressive model, auto-encoding model, seq2seq model, and UNIfied pre-trained Language Model (UNILM). The training process of the natural language processing model uses manually annotated historical fraud materials as the training set and the validation set to train the natural language processing model to obtain a converged natural language processing model and output a vector matrix corresponding to a predefined abnormal label, so that in the subsequent process of using the natural language processing model for semantic analysis, a vector matrix corresponding to multiple preset labels can be obtained according to the input target call recording.

[0090] The natural language processing processes of different natural language processing models are the same or similar. The natural language processing process includes but is not limited to text information segmentation, one-hot encoding, Word Embedding, and obtaining a vector matrix through a deep learning network. The following will introduce the natural language processing process:

[0091] In some embodiments, text information segmentation involves tokenizing the text in the form of long sentences, converting it into word vectors, then statistically counting the number of occurrences of the word vectors in the text information and encoding the positions of the word vectors through one-hot encoding to obtain the position encoding corresponding to the word vectors, and then performing word embedding to map the word vectors in high-dimensional form and their corresponding position encodings into low-dimensional continuous vectors. Further, a vector matrix is obtained through a deep learning network, and this vector matrix represents the machine language corresponding to the preset abnormal labels read from the text information that the machine can understand. It should be emphasized that this vector matrix is used to determine the priority later. The natural language processing process also includes part-of-speech tagging, named entity recognition, dependency parsing, etc.

[0092] Among them, tokenization refers to the process of splitting continuous natural language text into a sequence of words with semantic rationality and integrity; part-of-speech tagging can refer to the process of assigning a part of speech to each word in natural language text; named entity recognition, that is, proper name recognition, can refer to identifying entities with specific meanings in natural language text, mainly including personal names, place names, organization names, time and dates, etc.; dependency parsing can refer to inputting a Chinese sentence to obtain the dependency syntactic structure information of the sentence, and using the dependency relationship between words in the sentence to represent the syntactic structure information of words (such as subject-predicate, verb-object, attributive-middle, etc. structural relationships), and using a tree structure to represent the structure of the whole sentence (such as subject-verb-object, attributive-adverbial-complement, etc.); word vector refers to inputting a single Chinese word to obtain the vector representation of the word, and the calculation of word vectors can be achieved through training methods, mainly relying on a large amount of high-quality data and deep neural network technology to map the words in the language vocabulary into a vector with a fixed length; word sense similarity can be calculated by relying on a large amount of high-quality data and deep neural network technology to calculate the similarity between two words through word vectorization; short text similarity refers to inputting two short Chinese texts to output the semantic similarity between the texts, which can help quickly implement applications such as recommendation, retrieval, and ranking.

[0093] The following will take the unified pre-trained language model as an example to illustrate the natural language processing process:

[0094] First, the text information is tokenized to obtain the input text of the unified pre-trained language model. During the tokenization of the first text information of the scammer and the second text information of the victim, the vocabulary in the text information is segmented according to the preset corpus to obtain the input text, and the input text is a sequence of words composed of one or more characters. Exemplarily, the first text information of the scammer includes "remit 3000 yuan to the bank", and after tokenization, multiple input texts are obtained including: go, bank, remit, 3000, yuan; or, go, silver, bank, remit, money, 3000 yuan.

[0095] Such as Figure 3As shown in the figure, the unified pre-trained language model may include an embedding layer 301 and a Transformer layer 302. For example, taking X1, X2......X5 as the input text, the embedding layer 111 is used to extract features from the input text and perform vectorized representation of the features, and output the feature vectors of the input text. The Transformer layer 302 is used to perform semantic learning based on the feature vectors of the input text and output semantic features h1, h2......h5.

[0096] For example, the embedding layer 301 may include a segment embedding layer, a position embedding layer, and a token embedding layer, which are respectively used to perform vectorized representation of sentences for the input text, vectorized representation of the position information of each word in the input text, and vectorized representation of each word, and output sentence features, position features, and word features.

[0097] For example, the Transformer layer 302 may include multiple layers of Transformer modules. The Transformer module is a classic model architecture in the field of natural language processing and can learn the correlation between words (or words) in a sentence. The sentence features, position features, and word features of the input text pass through multiple layers of Transformer modules, and semantic features representing the meaning of the input text can be output.

[0098] It should be understood that the unified pre-trained language model is trained using historical fraud data to obtain a converged unified pre-trained language model, so that the output semantic features correspond to preset abnormal labels. For different language processing tasks, the calculation processes for the input text X1, X2......X5 are different, and the meanings of the output semantic features h1, h2......h5 are also different.

[0099] In the above embodiment, the abnormal call information is obtained from the text information obtained after speech recognition through the natural language processing model, and a vector matrix that can be understood by the machine is obtained, so as to extract the features in the text information, realizing speech analysis of the target call recording.

[0100] S203. Determine the call priority of the number to be called according to the abnormal call information.

[0101] Among them, the abnormal call information includes but is not limited to fraud information and blank call information. The fraud information is a vector matrix corresponding to a preset label, and the blank call information indicates that the calling number did not speak during the entire call and no valid information was obtained after semantic analysis. The following will be described in the case where the abnormal call information is fraud information:

[0102] In some embodiments, the abnormal call information is fraud information, which is machine language obtained by performing semantic analysis on the target call recording through a natural language processing model and can be a vector matrix.

[0103] Among them, the fraud information can be manually labeled by developers according to historical fraud data, and at the same time, the weight coefficient corresponding to the fraud information is set. For example, the fraud information includes the fraud amount, fraud type, and whether being deceived. The preset weight coefficient corresponding to the fraud amount is set to 0.3, the preset weight coefficient corresponding to the fraud type is set to 0.2, and whether being deceived is set to 0.5. The present disclosure does not make specific limitations on the setting of the preset weight coefficient.

[0104] Further, according to the vector matrices corresponding to multiple preset abnormal tags and the preset weight coefficients corresponding to multiple preset abnormal tags, the priority parameters of multiple preset abnormal tags are calculated, and then according to the range where the priority parameters of multiple preset abnormal tags are located, the call priority of the number to be called is determined.

[0105] Among them, the range where the priority parameter is located is a pre-set range. In the embodiments of the present disclosure, the pre-set range of the priority parameter is the highest priority, the secondary important priority, and the lowest priority. The highest priority can indicate that the potential victim has a high risk of being deceived and needs to be called immediately for dissuasion. The secondary important priority can indicate that the potential victim has a risk of being deceived but has not been deceived yet, so as to raise vigilance. The lowest priority can indicate that the potential victim has answered a fraud call and has a relatively small risk of being deceived, and can be called later.

[0106] Exemplarily, continuing with the above example, the preset weight coefficient corresponding to the fraud amount is 0.3, the preset weight coefficient corresponding to the fraud type is 0.2, and the preset weight coefficient corresponding to whether being deceived is 0.5. Then, weighted calculation is performed according to the vector matrices corresponding to multiple preset abnormal tags. If the priority parameter corresponding to the fraud amount obtained is 30, the priority parameter corresponding to the fraud type is 20, and the priority parameter corresponding to whether being deceived is 50, and the pre-set range of the priority parameter corresponding to whether being deceived is divided into the lowest priority 0 - 30, the secondary important priority 31 - 60, and the highest priority 61 - 100. It can be seen that the calculated priority parameter corresponding to whether being deceived is 50 belongs to the secondary important priority, indicating that the potential victim has a risk of being deceived but has not been deceived yet. Therefore, the outbound call order is adjusted according to the priority parameter, thereby raising the vigilance of the potential victim and preventing the potential victim from being deceived.

[0107] It should be noted that after determining the range of the priority parameters where multiple numbers to be called are located, each range of priority parameters includes m numbers to be called. For the outbound call order of the m numbers to be called, the magnitudes of the priority parameters corresponding to the m numbers to be called can be compared to further determine the priority of the m numbers to be called.

[0108] Exemplarily, the to-be-called numbers T1, T2, and T3 are all at the secondary important priority level, and the corresponding priority parameters of the to-be-called numbers T1, T2, and T3 are n1, n2, and n3 respectively, and n1 < n2 < n3. Then, during the outbound call to multiple to-be-called numbers at the secondary important priority level, the outbound call is first made to the to-be-called number T1, and then the outbound calls are made to T2 and T3 in sequence.

[0109] In the above embodiment, the abnormal call information obtained through semantic analysis and the set weight coefficient are used to determine the priority parameter. Then, according to the range where the priority parameter is located, the to-be-called number that needs to be immediately outbound-called is first determined, and then the other to-be-called numbers are sorted and outbound-called in sequence, realizing the judgment of the deceived risk degree of potential victims based on the abnormal call information, so as to preferentially outbound-call potential victims with a higher deceived risk degree to protect the property safety of potential victims.

[0110] In some embodiments, during the anti-fraud outbound call to the to-be-called number according to the call priority, first, according to the fraud information, determine the type of person to which the user of the to-be-called number belongs; the type of person includes, but is not limited to, not deceived and already deceived.

[0111] Exemplarily, according to the preset abnormal label of whether being deceived included in the fraud information, determine whether the type of person is not deceived or already deceived. If it is determined that the victim is not deceived, then determine the corresponding first preset language subsequently to prompt the victim not to answer the fraud number anymore; if it is determined that the victim is deceived, the determined second preset voice is used to prompt the victim to report the case as soon as possible.

[0112] Furthermore, obtain the preset voice corresponding to the victim type, where the preset voice is set differently according to different victim types and can be used to explain to the victim whether the victim is deceived and suggest reporting the case as soon as possible, etc. Then, when making an anti-fraud outbound call to the to-be-called number according to the call priority, play the preset voice.

[0113] In summary, the present disclosure uses the telephone number of the potential victim as the to-be-called number, obtains the target call recording corresponding to the to-be-called number, and the target call recording is the call recording between the fraudster and the potential victim; performs semantic analysis on the target call recording to obtain the fraud information therein to determine whether the potential victim is deceived, or determine the fraud means of the fraudster, or determine information such as the deceived amount of the potential victim; then, determine the priority of the to-be-called number according to the fraud information, and further make an anti-fraud outbound call to the to-be-called number according to the call priority, thereby improving the timeliness of anti-fraud dissuasion, realizing the timely dissuasion of potential victims with a high possibility of being deceived, and protecting the property safety of potential victims.

[0114] Figure 4 A device for determining the priority of outbound call data provided by an embodiment of the present disclosure, the device includes:

[0115] An acquisition module 401, configured to acquire a target call recording corresponding to a number to be called;

[0116] A processing module 402, configured to perform semantic analysis on the target call recording to obtain abnormal call information in the call recording; and determine the call priority of the number to be called according to the abnormal call information.

[0117] Optionally, the processing module 402 is specifically configured to perform semantic analysis on the target call recording through a natural language processing model, and obtain a vector matrix corresponding to a plurality of preset abnormal tags output by the natural language processing model;

[0118] Determining the call priority of the number to be called according to the abnormal call information includes:

[0119] Calculating priority parameters of a plurality of preset abnormal tags according to the vector matrix corresponding to the plurality of preset abnormal tags and the preset weight coefficients corresponding to the plurality of preset abnormal tags;

[0120] Determine the call priority of the number to be called according to the parameter ranges where the priority parameters of the plurality of preset abnormal tags are located.

[0121] Optionally, the acquisition module 401 is specifically configured to acquire a plurality of call recordings corresponding to the number to be called;

[0122] Determine a target call recording whose calling number is an abnormal calling number from the plurality of call recordings; wherein, the abnormal calling number is a telephone number marked as an abnormal call.

[0123] Optionally, the processing module 402 is further configured to determine a target personnel type to which the user of the number to be called belongs according to the abnormal call information;

[0124] Obtain a preset voice corresponding to the target personnel type;

[0125] When making an outbound call to the number to be called according to the call priority, play the preset voice.

[0126] Optionally, the processing module 402 is specifically configured to convert the audio information in the target call recording into text information;

[0127] Perform semantic analysis on the text information to extract abnormal call information in the text information.

[0128] Optionally, the processing module 402 is specifically configured to perform voice detection on the target call recording to obtain voice audio;

[0129] Identify the first audio of the first target person and the second audio of the second target person from the human voice audio, where the first target person and the second target person are the calling user and the called user of the target call recording respectively;

[0130] Convert both the first audio and the second audio into text information.

[0131] In summary, the priority determination device for anti-fraud outbound calls provided by the present disclosure uses the phone number of the potential victim as the number to be dialed, obtains the target call recording corresponding to the number to be dialed, and this target call recording is the call recording between the scammer and the potential victim; performs semantic analysis on the target call recording to obtain abnormal call information therein, so as to determine whether the potential victim has been deceived, or determine the fraud means of the scammer, or determine information such as the amount of money deceived by the potential victim, etc.; then, determines the priority of the number to be dialed according to the abnormal call information, and further makes anti-fraud outbound calls to the number to be dialed according to the call priority, thereby improving the timeliness of anti-fraud dissuasion, realizing timely dissuasion of potential victims with a high possibility of being deceived, and protecting the property safety of potential victims.

[0132] It should be noted that in the embodiments of the above outbound call data priority determination device, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present disclosure.

[0133] As Figure 5 shown, the embodiments of the present disclosure provide an electronic device, which includes: a processor 501, a memory 502, and a computer program stored on the memory 502 and executable on the processor 501. This computer program can be executed by the processor to implement each process executed by the above-mentioned method for determining the priority of outbound call data by a terminal, and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0134] The embodiments of the present disclosure provide a computer-readable storage medium, which is characterized in that a computer program is stored on this computer-readable storage medium. When this computer program is executed by a processor, it implements each process executed by the electronic device of the above-mentioned method for determining the priority of outbound call data, and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0135] Among them, this computer-readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc, etc.

[0136] An embodiment of the present disclosure provides a computer program product, which includes: when the computer program product runs on a computer, it enables the computer to implement the above-mentioned method for determining the priority of outbound call data, and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0137] From the above description of the embodiments, those skilled in the art can clearly understand that the present disclosure can be implemented by means of software and necessary general-purpose hardware. Of course, it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present disclosure, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk, or optical disc of a computer, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present disclosure.

[0138] It should be noted that in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of another identical element in the process, method, article or device including the said element.

[0139] The above are only specific embodiments of the present disclosure, enabling those skilled in the art to understand or implement the present disclosure. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to these embodiments described herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for determining the priority of outbound call data, characterized in that, comprising: Obtaining a target call recording corresponding to the number to be dialed; Performing semantic analysis on the target call recording to obtain abnormal call information in the target call recording; Determining the call priority of the number to be dialed according to the abnormal call information; The performing semantic analysis on the target call recording to obtain abnormal call information in the target call recording includes: Performing semantic analysis on the target call recording through a natural language processing model, and obtaining a vector matrix corresponding to multiple preset abnormal tags output by the natural language processing model; The determining the call priority of the number to be dialed according to the abnormal call information includes: Calculating the priority parameters of the multiple preset abnormal tags according to the vector matrix corresponding to the multiple preset abnormal tags and the preset weight coefficients corresponding to the multiple preset abnormal tags; Determining the call priority of the number to be dialed according to the parameter range where the priority parameters of the multiple preset abnormal tags are located.

2. The method according to claim 1, characterized in that, the obtaining a target call recording corresponding to the number to be dialed includes: Obtaining multiple call recordings corresponding to the number to be dialed; Determining the target call recording whose calling number is an abnormal calling number from the multiple call recordings; wherein, the abnormal calling number is a telephone number marked as an abnormal call.

3. The method according to claim 1, characterized in that, the method further includes: Determining the type of person to which the user of the number to be dialed belongs according to the abnormal call information; Obtaining a preset voice corresponding to the type of person; Playing the preset voice when making an outbound call to the number to be dialed according to the call priority.

4. The method according to claim 1, characterized in that, the performing semantic analysis on the target call recording through a natural language processing model includes: Converting the audio information in the target call recording into text information; Performing semantic analysis according to the text information to extract the abnormal call information in the text information.

5. The method according to claim 4, characterized in that, the converting the audio information in the target call recording into text information includes: Performing voice detection on the target call recording to obtain voice audio; Identifying the first audio of the first target person and the second audio of the second target person from the voice audio, where the first target person and the second target person are the calling user and the called user of the target call recording respectively; Converting both the first audio and the second audio into text information.

6. An apparatus for determining the priority of outbound call data, characterized in that, comprising: An obtaining module, configured to obtain a target call recording corresponding to the number to be dialed; A processing module, configured to perform semantic analysis on the target call recording to obtain abnormal call information in the call recording; and determine the call priority of the number to be dialed according to the abnormal call information; The processing module is specifically configured to perform semantic analysis on the target call recording through a natural language processing model, and obtain a vector matrix corresponding to a plurality of preset exception tags output by the natural language processing model; Determining the call priority of the to-be-called number according to the abnormal call information includes: calculating a priority parameter of the plurality of preset exception tags according to the vector matrix corresponding to the plurality of preset exception tags and the preset weight coefficients corresponding to the plurality of preset exception tags; Determine the call priority of the to-be-called number according to the parameter range in which the priority parameters of the plurality of preset exception tags are located.

7. An electronic device, Characterized in that, Comprising: A processor, a memory, and a computer program stored on the memory and executable on the processor, where when the computer program is executed by the processor, it implements the method for determining the priority of outbound call data according to any one of claims 1 to 5.

8. A computer-readable storage medium, Characterized in that, Comprising: A computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, it implements the method for determining the priority of outbound call data according to any one of claims 1 to 5.

9. A computer program product, Characterized in that, Comprising: When the computer program product runs on a computer, the computer is caused to implement the method for determining the priority of outbound call data according to any one of claims 1 to 5.

Images