Risk data identification and model training method and device and electronic equipment

Abstract

The invention discloses a risk data identification and model training method and device and electronic equipment. The method comprises the steps that to-be-recognized text data is acquired, and the to-be-recognized text data is obtained by converting voice data in a voice call; the to-be-recognized text data is preprocessed according to a priori knowledge lexicon, the priori knowledge lexicon comprises multiple keywords corresponding to the risk features, and the preprocessing operation is used for screening key phrases containing the keywords in the to-be-recognized text data; and adopting the target model to identify the to-be-identified text data after the preprocessing operation to obtain an identification result, the identification result including that the voice call corresponding to the to-be-identified text data has an abnormal risk and the voice call corresponding to the to-be-identified text data is normal. According to the method and the device, the technical problem that a traditional risk prevention monitoring means is poor in risk data identification effect in a voice special line type service scene in the related technology is solved.

Description

Technical Field

[0001] The present application relates to the fields of network technology and security technology, and more specifically, to a method, device and electronic equipment for risk data identification and model training. Background Art

[0002] With the advent of a new era of communications, the scale of telecommunications users continues to expand, and the number of new cybercrime cases, mainly represented by telecommunications network fraud, continues to rise. Criminals use mobile phones, text messages, landlines and other channels to commit telecommunications network fraud, especially using cloud relays, voice dedicated lines and other business scenarios to carry out batch harassment, as well as fraud-related outbound calls, which have seriously infringed upon the lives and property rights of the people.

[0003] However, the traditional risk prevention and monitoring methods in related technologies have single functions and are difficult to achieve accurate analysis and identification of complex call traffic scenarios. There is also a technical problem of poor risk data identification effect in voice dedicated line business scenarios.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present application provide a risk data identification and model training method, device and electronic device to at least solve the technical problem that traditional risk prevention and monitoring means in related technologies are not effective in identifying risk data in voice dedicated line business scenarios.

[0006] According to one aspect of an embodiment of the present application, a risk data identification method is provided, including: obtaining text data to be identified, wherein the text data to be identified is converted from voice data in a voice call; performing a preprocessing operation on the text data to be identified based on a priori knowledge vocabulary, wherein the priori knowledge vocabulary contains multiple keywords corresponding to risk characteristics, and the preprocessing operation is used to screen key phrases containing the keywords in the text data to be identified; using a target model to identify the text data to be identified after the preprocessing operation, and obtain a recognition result, wherein the recognition result includes: there is an abnormal risk in the voice call corresponding to the text data to be identified, and the voice call corresponding to the text data to be identified is normal.

[0007] Optionally, the method also includes: obtaining a basic corpus, wherein the basic corpus contains a plurality of text data obtained by speech translation, and label information corresponding to the text data, the label information including: a first label for characterizing that the speech corresponding to the text data has an abnormal risk, and a second label for characterizing that the speech corresponding to the text data is normal; determining the text length of the text data for each piece of label information as the first label, and based on the text length, determining the number of keywords planned to be extracted from the text data; according to the number of keywords, extracting word segments used to characterize risk characteristics from the text data, and determining the word segments as keywords to obtain a priori knowledge vocabulary.

[0008] Optionally, after obtaining the prior knowledge vocabulary, the method also includes: determining a first number of data items in the basic corpus, which are text data whose label information contains the keyword as a first label; determining the total number of text data whose label information is the first label in the basic corpus as a second number of data items; determining the ratio of the first number of data items to the second number of data items as the occurrence ratio corresponding to the keyword, and determining a keyword level corresponding to the keyword based on a ratio interval corresponding to the occurrence ratio, wherein the keyword level is used to characterize the degree of influence of the keyword on determining whether the voice call corresponding to the text data has an abnormal risk.

[0009] Optionally, preprocessing the text data to be recognized based on the prior knowledge vocabulary includes: when the text length of the text data to be recognized exceeds a preset length threshold, dividing the text data to be recognized into multiple sentence fragments according to preset characters, wherein the preset characters include: punctuation characters; based on the prior knowledge vocabulary, determining the keywords contained in each sentence fragment, and determining the target retention probability corresponding to the sentence fragment and other sentence fragments adjacent to the sentence fragment in the text data to be recognized based on the keyword level corresponding to the keyword, wherein the target retention probability is used to characterize the probability that the sentence fragment is retained in the text data to be recognized after the preprocessing operation; according to the target retention probability, retaining the sentence fragments in the text data to be recognized to obtain the text data to be recognized after the preprocessing operation, wherein the text length of the text data to be recognized after the preprocessing operation is less than the text length of the text data to be recognized that has not undergone the preprocessing operation, and the length difference ratio between the text length of the text data to be recognized after the preprocessing operation and the preset length threshold is less than the preset length difference ratio.

[0010] Optionally, determining the target retention probability corresponding to the sentence fragment, and other sentence fragments adjacent to the sentence fragment in the text data to be recognized includes: determining the text length of the text data to be recognized that has not undergone preprocessing operations, and determining the initial retention probability based on the text length and a preset length threshold; determining the highest keyword level among the keyword levels corresponding to the keywords contained in the sentence fragment, and determining the adjustment coefficient corresponding to the highest keyword level, wherein the higher the keyword level, the greater the degree of influence of the keyword on judging whether there is an abnormal risk in the voice call corresponding to the text data; determining the target retention probability corresponding to the sentence fragment based on the adjustment coefficient and the initial retention probability; determining the context influence increment based on the target retention probability corresponding to the sentence fragment, and adjusting the target retention probability corresponding to other sentence fragments adjacent to the sentence fragment in the text data to be recognized based on the context influence increment.

[0011] Optionally, the target model is obtained by training based on the training data set, including: determining text data in the basic corpus whose text length exceeds a preset length threshold; performing preprocessing operations on the text data whose text length exceeds the preset length threshold based on a priori knowledge vocabulary, and using the preprocessed text data to replace the corresponding text data in the basic corpus that has not been preprocessed, to obtain an updated basic corpus; based on the updated basic corpus, constructing a training data set, and using the training data set to train the initial natural language processing model to obtain the target model, wherein the training data set contains multiple training samples and label information corresponding to the training samples, and the natural language processing model includes: a text convolutional neural network model.

[0012] Optionally, based on the updated basic corpus, constructing a training data set includes: performing word segmentation processing on each text data in the updated basic corpus to obtain an initial training sample and label information corresponding to the training sample; randomly replacing a preset proportion of words in the initial training sample with homophones to obtain a first training sample; determining the pinyin data corresponding to the initial training sample and / or the first training sample to obtain a second training sample; and constructing a training data set based on the initial training sample, the first training sample and the second training sample.

[0013] According to another aspect of an embodiment of the present application, a model training method is provided, including: obtaining a basic corpus, determining text data in the basic corpus whose text length exceeds a preset length threshold, wherein the basic corpus contains a plurality of text data obtained by speech translation, and label information corresponding to the text data, the label information including: a first label for characterizing that the speech corresponding to the text data has an abnormal risk, and a second label for characterizing that the speech corresponding to the text data is normal; performing a preprocessing operation on the text data whose text length exceeds the preset length threshold according to a priori knowledge vocabulary, wherein the priori knowledge vocabulary contains a plurality of keywords corresponding to risk features, and the preprocessing operation is used to screen key phrases containing the keywords in the text data; using the text data after the preprocessing operation to replace the corresponding text data in the basic corpus that has not been preprocessed, to obtain an updated basic corpus; based on the updated basic corpus, training an initial natural language processing model to obtain a target model, wherein the target model is used to identify the text data to be identified, to obtain a recognition result, the text data to be identified is obtained by converting the speech data in a voice call, and the recognition result includes: the voice call corresponding to the text data to be identified has an abnormal risk, and the voice call corresponding to the text data to be identified is normal.

[0014] According to another aspect of the embodiment of the present application, a risk data identification device is also provided, including: a data acquisition module, used to acquire text data to be identified, wherein the text data to be identified is converted from voice data in a voice call; a long text processing module, used to perform preprocessing operations on the text data to be identified based on a priori knowledge vocabulary, wherein the priori knowledge vocabulary contains multiple keywords corresponding to risk characteristics, and the preprocessing operation is used to screen key phrases containing keywords in the text data to be identified; a risk identification module, used to use a target model to identify the text data to be identified after the preprocessing operation, and obtain an identification result, wherein the identification result includes: there is an abnormal risk in the voice call corresponding to the text data to be identified, and the voice call corresponding to the text data to be identified is normal.

[0015] According to another aspect of an embodiment of the present application, there is also provided an electronic device, including: a memory and a processor, the processor being used to run a program stored in the memory, wherein the risk data identification method or the model training method is executed when the program is run.

[0016] According to another aspect of an embodiment of the present application, a non-volatile storage medium is further provided, the non-volatile storage medium including a stored computer program, wherein the device where the non-volatile storage medium is located executes the risk data identification method or the model training method by running the computer program.

[0017] According to another aspect of the embodiments of the present application, a computer program product is also provided, including a computer program, which implements the steps of the risk data identification method or the model training method when the computer program is executed by a processor.

[0018] In an embodiment of the present application, text data to be identified is obtained, wherein the text data to be identified is converted from voice data in a voice call; a preprocessing operation is performed on the text data to be identified based on a priori knowledge vocabulary, wherein the priori knowledge vocabulary contains multiple keywords corresponding to risk characteristics, and the preprocessing operation is used to screen key phrases containing keywords in the text data to be identified; a target model is used to identify the text data to be identified after the preprocessing operation to obtain a recognition result, wherein the recognition result includes: there is an abnormal risk in the voice call corresponding to the text data to be identified, and the voice call corresponding to the text data to be identified is normal. By utilizing natural language processing technology and deep learning algorithms, a priori knowledge vocabulary in the field of telecommunication risk identification is constructed, and the core essence of long-length voice translation texts is grasped through core keywords, fully retaining the characteristics of data in the telecommunication risk field, achieving the purpose of improving the accuracy of risk data identification in voice dedicated line business scenarios, thereby solving the technical problem that traditional risk prevention and monitoring means in related technologies are not effective in risk data identification in voice dedicated line business scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0020] Figure 1 It is a hardware structure block diagram of a computer terminal (or electronic device) for implementing a method for risk data identification provided in an embodiment of the present application;

[0021] Figure 2 It is a schematic diagram of a method flow for risk data identification provided according to an embodiment of the present application;

[0022] Figure 3 It is a schematic diagram of a method flow for intelligent identification of risk data based on natural language processing provided in an embodiment of the present application;

[0023] Figure 4 It is a schematic diagram of a long text-to-text data processing flow provided according to an embodiment of the present application;

[0024] Figure 5 It is a schematic diagram of a method flow of model training provided according to an embodiment of the present application;

[0025] Figure 6It is a structural schematic diagram of a risk data identification device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] 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 drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] In order to facilitate those skilled in the art to better understand the embodiments of the present application, some technical terms or nouns involved in the embodiments of the present application are explained as follows:

[0029] NLP (Natural Language Processing): is a machine learning algorithm and technology that enables the model to effectively interpret, process and understand human language and semantics.

[0030] textCNN (Convolutional Neural Network for text): A deep learning model that uses multiple different kernel_sizes to extract key information from sentences for natural language analysis.

[0031] With the advent of a new era of communications, the scale of telecommunications users continues to expand. Criminals use mobile phones, text messages, landlines and other channels to carry out telecommunications network fraud by using methods such as fake order rebates, false investment and financial management, false shopping services, impersonating e-commerce logistics customer service, and false credit fraud. In particular, they use cloud relays, dedicated voice lines and other business scenarios to carry out batch harassment and fraud-related outbound calls, which have seriously infringed upon the lives and property rights of the people.

[0032] In the related technologies, the monitoring and early warning technical means of basic operators for such voice services mostly focus on single call behavior data, which makes it difficult to achieve accurate analysis and identification of complex voice scenarios. In view of the problem that traditional risk monitoring means have single functions and the length of the general NLP model speech translation text varies greatly in special voice service scenarios, the embodiments of this application provide relevant solutions, using improved natural language processing technology and deep learning algorithms to specifically build NLP models and recognition methods based on voice service scenarios, which are described in detail below.

[0033] According to an embodiment of the present application, a method embodiment of risk data identification is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0034] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal (or electronic device) for implementing the risk data identification method is shown. Figure 1 As shown, the computer terminal 10 (or electronic device) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0035] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or electronic device). As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0036] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the risk data identification method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the above-mentioned risk data identification method is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0037] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0038] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or electronic device).

[0039] In the above operating environment, the embodiment of the present application provides a risk data identification method. Figure 2 is a schematic diagram of a method flow for risk data identification provided according to an embodiment of the present application, such as Figure 2 As shown, the method comprises the following steps:

[0040] Step S202, obtaining text data to be recognized, wherein the text data to be recognized is obtained by converting voice data in a voice call;

[0041] Step S204, performing a preprocessing operation on the text data to be identified according to the prior knowledge word library, wherein the prior knowledge word library contains a plurality of keywords corresponding to risk characteristics, and the preprocessing operation is used to screen key phrases containing the keywords in the text data to be identified;

[0042] Step S206, using the target model to identify the text data to be identified after the preprocessing operation, to obtain a recognition result, wherein the recognition result includes: the voice call corresponding to the text data to be identified has an abnormal risk, and the voice call corresponding to the text data to be identified is normal.

[0043] Through the above steps, by utilizing natural language processing technology and deep learning algorithms, a priori knowledge vocabulary in the field of telecommunications risk identification is constructed, and the core essence of long-length speech translation texts is grasped through core keywords, which fully retains the characteristics of data in the telecommunications risk field, and achieves the purpose of improving the accuracy of risk data identification in voice dedicated line business scenarios, thereby solving the technical problem that traditional risk prevention and monitoring methods in related technologies are not effective in identifying risk data in voice dedicated line business scenarios.

[0044] The following further introduces the risk data identification method in steps S202 to S206 of the embodiment of the present application.

[0045] Figure 3 is a schematic diagram of a method flow for intelligent identification of risk data based on natural language processing provided in an embodiment of the present application, such as Figure 3 As shown, the following steps are included: constructing a basic corpus in the field of telecommunication network risk identification (divided into risk data and normal data), constructing a telecommunication network risk identification prior knowledge vocabulary based on risk data (i.e., text data with label information as the first label), data processing for long-length voice media translation texts, constructing and generating trainable telecommunication network risk identification text data, training NLP-specific models in the field of telecommunication network risk, testing and tuning NLP-specific models in the field of telecommunication network risk identification, and reasoning applications of NLP-specific models in the field of telecommunication network risk identification. The following is a detailed introduction.

[0046] First, the basic corpus in the field of telecommunication network risk identification is obtained, and a priori knowledge vocabulary is constructed based on the basic corpus. The specific steps are as follows.

[0047] In some embodiments of the present application, the method also includes the following steps: obtaining a basic corpus, wherein the basic corpus contains a plurality of text data obtained by speech translation, and label information corresponding to the text data, the label information including: a first label for characterizing that the speech corresponding to the text data has an abnormal risk, and a second label for characterizing that the speech corresponding to the text data is normal; determining the text length of the text data for each piece of label information as the first label, and based on the text length, determining the number of keywords planned to be extracted from the text data; according to the number of keywords, extracting word segments used to characterize risk characteristics from the text data, and determining the word segments as keywords to obtain a priori knowledge vocabulary.

[0048] Specifically, a basic corpus in the field of telecommunication network risk identification is constructed. For example, the relevant media speech translation text data can be annotated and labeled according to the two categories of normal media speech (i.e., the second label) and risky media speech (i.e., the first label). For individual media data with inaccurate translated text, auxiliary annotation can be performed with the help of manual secondary analysis to further enhance the accuracy of the basic corpus.

[0049] Afterwards, a prior knowledge vocabulary can be constructed based on the text data (i.e., risk data) with label information as the first label in the basic corpus. In this embodiment, the core keywords in all risk data can be extracted through keyword extraction technology, and finally a prior knowledge vocabulary in the field of telecommunications risk identification can be constructed for all risk data in the basic corpus, as follows.

[0050] First, the number of extracted keywords N can be set according to the length of the media speech translation text data (i.e., the text length of the text data whose tag information is the first tag). For example, if the longest input text word number of the set text classification model (i.e., the target model) is Tmax, then when the word number of the text length is less than Tmax / 2 (rounded), N=3 can be set, and when the word number is greater than Tmax, N=6 can be set;

[0051] In an embodiment of the present application, by traversing the text data whose label information is the first label in the basic corpus, the text data is preprocessed (including word segmentation, removal of invalid values / punctuation / stop words / interference words, etc.), and core keywords are extracted according to the set number of keywords N. After the keyword extraction of all data is completed, the prior knowledge vocabulary can be obtained.

[0052] Furthermore, it is also necessary to determine the keyword level of each keyword in the prior knowledge vocabulary. The specific steps are as follows.

[0053] In some embodiments of the present application, after obtaining the prior knowledge vocabulary, the method also includes the following steps: determining a first data number of text data in the basic corpus whose label information contains the keyword as a first label; determining the total number of text data in the basic corpus whose label information is the first label as a second data number; determining the ratio of the first data number to the second data number as the occurrence ratio corresponding to the keyword, and determining the keyword level corresponding to the keyword based on the ratio interval range corresponding to the occurrence ratio, wherein the keyword level is used to characterize the degree of influence of the keyword on judging whether the voice call corresponding to the text data has an abnormal risk.

[0054] Specifically, the appearance ratio X of the keyword in all risk data (the number of texts with the label information being the first label) is counted, where X=the number of risk data items that appear (i.e., the number of first data items mentioned above) / the number of all risk data items (i.e., the number of second data items mentioned above), and the keyword level is marked according to the appearance ratio X. For example, in the embodiment of the present application, levels 1-5 can be set, and the larger the number, the higher the level, and the higher the corresponding confidence level. For example: the core keyword appearance ratio X is less than 1.6%, and the level is marked as 1; the core keyword appearance ratio X is greater than or equal to 1.6%, and less than 5.25%, and the level is marked as 2; the core keyword appearance ratio X is greater than or equal to 5.25%, and less than 10%, and the level is marked as 3; the core keyword appearance ratio X is greater than or equal to 10%, and less than 35%, and the level is marked as 4; the core keyword appearance ratio X is greater than or equal to 35%, and the level is marked as 5.

[0055] After obtaining the prior knowledge vocabulary and determining the keyword level corresponding to each keyword, the prior indicator vocabulary can be used to preprocess the text data in the basic corpus that is too long to update the basic corpus, and further generate a training data set to train the NLP model, thereby obtaining the target model in the field of telecommunications network risk identification. The following is a detailed introduction.

[0056] In some embodiments of the present application, the target model is obtained by training based on a training data set, including: determining text data in a basic corpus whose text length exceeds a preset length threshold; performing a preprocessing operation on the text data whose text length exceeds the preset length threshold based on a priori knowledge vocabulary, and using the preprocessed text data to replace the corresponding text data in the basic corpus that has not been preprocessed, to obtain an updated basic corpus; based on the updated basic corpus, constructing a training data set, and using the training data set to train an initial natural language processing model to obtain a target model, wherein the training data set contains multiple training samples and label information corresponding to the training samples, and the natural language processing model includes: a text convolutional neural network model.

[0057] Specifically, based on the prior knowledge vocabulary, preprocessing operations are performed on the speech media translation text data that is too long in the basic corpus (i.e., text data whose text length exceeds a preset length threshold), focusing on retaining sentence fragments (sentences) near core keywords to form data that meets the training length, and updating to form a new basic electronic corpus.

[0058] like Figure 4 As shown, for the text data in the basic corpus whose text length exceeds Tmax (i.e., the preset length threshold) characters, processing is performed. First, the total number of characters (text length) of the current long text data is determined to be Tlong; according to the preset characters, the current long text data is divided into multiple sentence fragments (sentences). In this embodiment, the sentences can be segmented by ending with punctuation marks such as commas, periods, question marks, and exclamation marks; then, all the split sentence fragments (sentences) are traversed, and the keywords appearing in each sentence fragment (sentence) are matched according to the multi-mode matching algorithm based on the prior knowledge vocabulary; then, according to the keyword level corresponding to the keyword, the target retention probability corresponding to the sentence fragment and other sentence fragments adjacent to the sentence fragment in the text data is determined, as follows.

[0059] First, calculate the initial retention probability. In this embodiment, take the initial retention probability P = (Tmax / Tlong)*80%; and set the context influence probability and floating influence probability corresponding to the key sentence fragment, for example, the context influence probability q = 60%, and the floating influence probability x = 5%.

[0060] Afterwards, the target retention probability corresponding to the sentence fragment and the target retention probabilities corresponding to other sentence fragments adjacent to the sentence fragment can be calculated according to the adjustment coefficient corresponding to the highest keyword level among the keyword levels corresponding to the keywords contained in the sentence fragment, and the initial retention probability. For example, in the embodiment of the present application, the target retention probabilities corresponding to all sentence fragments (sentences) can be calculated according to the following strategy:

[0061] 1) If the core keyword with level 1 appears at least once, the target retention probability increases by p1 (p1 = (1-P) * 60%), and the target retention probability of the sentence fragments (sentences) before and after the sentence fragment (sentence) increases by p1 * q;

[0062] 2) If the core keyword with level 2 appears at least once, the target retention probability is increased by p2 (p2 = (1-P) * 70%), and the target retention probability of the sentence fragments (sentences) before and after the sentence fragment (sentence) is increased by p2 * q;

[0063] 3) If there is at least one core keyword of level 3, the target retention probability is increased by p3 (p3 = (1 - P) * 80%), and the target retention probabilities of the statement fragments (sentences) before and after this statement fragment (sentence) are increased by p3 * q;

[0064] 4) If there is at least one core keyword of level 4, the target retention probability is increased by p4 (p4 = (1 - P) * 90%), and the target retention probabilities of the statement fragments (sentences) before and after this statement fragment (sentence) are increased by p4 * q;

[0065] 5) If there is at least one core keyword of level 5, the target retention probability is increased by p5 (p5 = 1 - P), and the target retention probabilities of the statement fragments (sentences) before and after this statement fragment (sentence) are increased by p5 * q;

[0066] 6) If a single statement fragment (sentence) contains multiple core keywords, the target retention probability is calculated only according to the core keyword with the highest level;

[0067] 7) The cumulative probability of the target retention probability of the statement fragment (sentence) does not exceed 1.

[0068] After calculating the target retention probability corresponding to each statement fragment (sentence), according to the target retention probability, retain the statement fragments in the text data to obtain the text data after the preprocessing operation. Assume that the text length (total number of words) of the retained statement fragments (sentences) is Tshort.

[0069] 1) If Tshort > Tmax and (Tmax / Tshort) < 80%, then set q = q - x, and then repeat the above calculation of the target retention probability and perform the retention operation according to the target retention probability. Otherwise, use the currently retained sentences as the final translated text data, that is, use the text data after the preprocessing operation to replace the corresponding text data in the basic corpus that has not undergone the preprocessing operation to obtain the updated basic corpus;

[0070] 2) If Tshort < Tmax and (Tshort / Tmax) < 80%, then set q = q + x, and then repeat the above calculation of the target retention probability and perform the retention operation according to the target retention probability. Otherwise, use the currently retained sentences as the final translated text data, that is, use the text data after the preprocessing operation to replace the corresponding text data in the basic corpus that has not undergone the preprocessing operation to obtain the updated basic corpus;

[0071] 3) If the last Tshort value is greater than Tmax and the current value is less than Tmax, or the current Tshort value is greater than Tmax and the last value is less than Tmax, the corresponding retained sentence when the Tshort value is greater than Tmax is taken as the final translated text data.

[0072] It should be noted that if the long text data undergoing preprocessing operations is text data that has been marked as normal in the training phase, or is long speech translation text data (text data to be identified) in the application reasoning process, then a keyword extraction algorithm can be used to extract N core keywords from the text data, the level of the extracted core keywords is set to 5, and the initial sentence retention probability P = (Tmax / Tlong)*90% is reset.

[0073] After obtaining the updated basic corpus, training data and test data for the NLP model can be generated based on the updated basic corpus, that is, a training data set is constructed. The specific steps are as follows.

[0074] In some embodiments of the present application, constructing a training data set based on an updated basic corpus includes the following steps: performing word segmentation processing on each text data in the updated basic corpus to obtain an initial training sample and label information corresponding to the training sample; randomly replacing a preset proportion of words in the initial training sample with homophones to obtain a first training sample; determining the pinyin data corresponding to the initial training sample and / or the first training sample to obtain a second training sample; and constructing a training data set based on the initial training sample, the first training sample, and the second training sample.

[0075] Specifically, each text data in the updated basic corpus is processed (including but not limited to: word segmentation, removal of punctuation / stop words / interference words, etc.) to obtain an initial training sample. Afterwards, in order to enhance the adaptability of the text classification model and improve the problem caused by the inaccurate translation text results, more sample data is covered. The embodiment of the present application can randomly select 10% of the words for homophone replacement based on the initial training sample, and expand 1 sample to 1-3 samples (first training sample). The original label category of the sample remains unchanged, that is, it originally corresponds to the first label, and the transformed data still corresponds to the first label; in addition, it can also be based on the aforementioned processing. Pinyin translation is performed to obtain a second training sample, and the pinyin maintains the Chinese word segmentation structure, and the Chinese pinyin and the original sample are spliced. Finally, a training data set that can be used to train the target model is formed.

[0076] After that, the training data set can be used to train the initial natural language processing model to obtain the target model. In this embodiment, the initial natural language processing model can use the textCNN model, which is a 5-layer structure, including an input layer, a convolution layer, a maximum pooling layer, a fully connected layer, and a classification layer. The text classification algorithm is used to train the training data set generated by the above steps to generate an NLP-specific model in the field of telecommunications risk identification, namely the target model, to achieve accurate classification and identification of fraud-related texts.

[0077] In addition, the NLP-specific models in the field of telecommunications risk identification can be further tested and tuned. The recognition effect of the target model obtained through training can be detected based on the test data. In actual use, the above steps of training data set construction and model training can be repeatedly cycled, and data can be added and replaced for optimization until the model meets the application effect.

[0078] When the trained target model is used for actual reasoning applications, it is also necessary to perform long-length processing (i.e. the preprocessing operation mentioned above) on the obtained translated text results (i.e. the text data to be recognized), Chinese word segmentation processing, Chinese pinyin splicing and other operations, and then use them as model input for reasoning application and output classification recognition results. The specific steps for preprocessing the text data to be recognized are as follows.

[0079] In some embodiments of the present application, preprocessing the text data to be recognized based on the prior knowledge vocabulary includes the following steps: when the text length of the text data to be recognized exceeds a preset length threshold, dividing the text data to be recognized into multiple sentence fragments according to preset characters, wherein the preset characters include: punctuation characters; determining the keywords contained in each sentence fragment based on the prior knowledge vocabulary, and determining the target retention probability corresponding to the sentence fragment and other sentence fragments adjacent to the sentence fragment before and after in the text data to be recognized based on the keyword level corresponding to the keyword, wherein the target retention probability is used to characterize the probability that the sentence fragment is retained in the text data to be recognized after the preprocessing operation; retaining the sentence fragments in the text data to be recognized according to the target retention probability to obtain the text data to be recognized after the preprocessing operation, wherein the text length of the text data to be recognized after the preprocessing operation is less than the text length of the text data to be recognized that has not been preprocessed, and the length difference ratio between the text length of the text data to be recognized after the preprocessing operation and the preset length threshold is less than the preset length difference ratio.

[0080] In some embodiments of the present application, determining the target retention probability corresponding to a sentence fragment, and other sentence fragments adjacent to the sentence fragment in the text data to be recognized includes the following steps: determining the text length of the text data to be recognized that has not undergone preprocessing operations, and determining the initial retention probability based on the text length and a preset length threshold; determining the highest keyword level among the keyword levels corresponding to the keywords contained in the sentence fragment, and determining the adjustment coefficient corresponding to the highest keyword level, wherein the higher the keyword level, the greater the degree of influence of the keyword on judging whether there is an abnormal risk in the voice call corresponding to the text data; determining the target retention probability corresponding to the sentence fragment based on the adjustment coefficient and the initial retention probability; determining the context influence increment based on the target retention probability corresponding to the sentence fragment, and adjusting the target retention probability corresponding to other sentence fragments adjacent to the sentence fragment in the text data to be recognized based on the context influence increment.

[0081] The above method flow for preprocessing the text data to be recognized based on the prior knowledge vocabulary in the actual reasoning application stage of the target model is similar to the method flow for preprocessing the long text data in the basic corpus in the training data set construction stage. It will not be repeated here. For the specific process steps, please refer to the relevant explanation of the preprocessing operation of the long text data in the basic corpus.

[0082] This application solution constructs a priori knowledge vocabulary in the field of telecommunications risk identification, and uses core keywords to grasp the core essence of long-length speech-to-text data, forming text classification data that meets the training requirements, fully retaining the characteristics of data in the field of telecommunications risk identification, and ensuring the accuracy of training professional NLP models in the field of telecommunications risk identification. In addition, since the training data set adds Chinese pinyin information during the construction process, and innovatively utilizes technologies such as long-length media text core essence processing, Chinese word segmentation processing, and Chinese pinyin splicing, the NLP-specific model in the field of telecommunications network risk identification that is finally trained can also achieve relatively accurate judgment and analysis in special circumstances such as interference in the media environment and errors in speech-to-text results, thereby making the model highly available.

[0083] According to an embodiment of the present application, a model training method is also provided. Figure 5 is a schematic diagram of a method flow for risk data identification provided according to an embodiment of the present application, such as Figure 5 As shown, the method comprises the following steps:

[0084] Step S502, obtaining a basic corpus, and determining text data in the basic corpus whose text length exceeds a preset length threshold, wherein the basic corpus contains a plurality of text data obtained by speech translation, and label information corresponding to the text data, and the label information includes: a first label for indicating that the speech corresponding to the text data has an abnormal risk, and a second label for indicating that the speech corresponding to the text data is normal;

[0085] Step S504, performing a preprocessing operation on the text data whose text length exceeds a preset length threshold according to the prior knowledge word library, wherein the prior knowledge word library contains a plurality of keywords corresponding to risk characteristics, and the preprocessing operation is used to screen key phrases containing the keywords in the text data;

[0086] Step S506, using the text data after the preprocessing operation to replace the corresponding text data not subjected to the preprocessing operation in the basic corpus, to obtain an updated basic corpus;

[0087] Step S508, based on the updated basic corpus, the initial natural language processing model is trained to obtain a target model, wherein the target model is used to recognize the text data to be recognized and obtain a recognition result, the text data to be recognized is obtained by converting the voice data in the voice call, and the recognition result includes: the voice call corresponding to the text data to be recognized has an abnormal risk, and the voice call corresponding to the text data to be recognized is normal.

[0088] It should be noted that the wind model training method provided in this embodiment is similar to Figure 2 The risk data identification method shown corresponds to a method embodiment, therefore, the relevant explanations and descriptions of the above-mentioned risk data identification method are also applicable to the embodiments of the present application and will not be repeated here.

[0089] According to an embodiment of the present application, an embodiment of a risk data identification device is also provided. Figure 6 Schematic diagram of a risk data identification device provided according to an embodiment of the present application. Figure 6 As shown, the device comprises:

[0090] The data acquisition module 60 is used to acquire the text data to be recognized, wherein the text data to be recognized is obtained by converting the voice data in the voice call;

[0091] The long text processing module 62 is used to perform a preprocessing operation on the text data to be identified based on the prior knowledge word library, wherein the prior knowledge word library contains a plurality of keywords corresponding to the risk characteristics, and the preprocessing operation is used to screen the key phrases containing the keywords in the text data to be identified;

[0092] The risk identification module 64 is used to use the target model to identify the text data to be identified after the preprocessing operation to obtain the identification result, wherein the identification result includes: the voice call corresponding to the text data to be identified has an abnormal risk, and the voice call corresponding to the text data to be identified is normal.

[0093] Optionally, the data acquisition module 60 is also used to: obtain a basic corpus, wherein the basic corpus contains a plurality of text data obtained by speech translation, and label information corresponding to the text data, the label information including: a first label for characterizing that the speech corresponding to the text data has an abnormal risk, and a second label for characterizing that the speech corresponding to the text data is normal; determine the text length of the text data for each piece of label information as the first label, and determine the number of keywords planned to be extracted from the text data based on the text length; extract word segmentations used to characterize risk characteristics from the text data according to the number of keywords, and determine the word segmentations as keywords to obtain a priori knowledge vocabulary.

[0094] Optionally, after obtaining the prior knowledge vocabulary, the risk data identification device is also used to: determine the first number of data items in the basic corpus, which are text data whose label information contains the keyword as the first label; determine the total number of text data whose label information is the first label in the basic corpus as the second number of data items; determine the ratio of the first number of data items to the second number of data items as the occurrence ratio corresponding to the keyword, and determine the keyword level corresponding to the keyword based on the ratio interval range corresponding to the occurrence ratio, wherein the keyword level is used to characterize the degree of influence of the keyword on judging whether the voice call corresponding to the text data has abnormal risks.

[0095] Optionally, preprocessing the text data to be recognized based on the prior knowledge vocabulary includes: when the text length of the text data to be recognized exceeds a preset length threshold, dividing the text data to be recognized into multiple sentence fragments according to preset characters, wherein the preset characters include: punctuation characters; based on the prior knowledge vocabulary, determining the keywords contained in each sentence fragment, and determining the target retention probability corresponding to the sentence fragment and other sentence fragments adjacent to the sentence fragment in the text data to be recognized based on the keyword level corresponding to the keyword, wherein the target retention probability is used to characterize the probability that the sentence fragment is retained in the text data to be recognized after the preprocessing operation; according to the target retention probability, retaining the sentence fragments in the text data to be recognized to obtain the text data to be recognized after the preprocessing operation, wherein the text length of the text data to be recognized after the preprocessing operation is less than the text length of the text data to be recognized that has not undergone the preprocessing operation, and the length difference ratio between the text length of the text data to be recognized after the preprocessing operation and the preset length threshold is less than the preset length difference ratio.

[0096] Optionally, determining the target retention probability corresponding to the sentence fragment, and other sentence fragments adjacent to the sentence fragment in the text data to be recognized includes: determining the text length of the text data to be recognized that has not undergone preprocessing operations, and determining the initial retention probability based on the text length and a preset length threshold; determining the highest keyword level among the keyword levels corresponding to the keywords contained in the sentence fragment, and determining the adjustment coefficient corresponding to the highest keyword level, wherein the higher the keyword level, the greater the degree of influence of the keyword on judging whether there is an abnormal risk in the voice call corresponding to the text data; determining the target retention probability corresponding to the sentence fragment based on the adjustment coefficient and the initial retention probability; determining the context influence increment based on the target retention probability corresponding to the sentence fragment, and adjusting the target retention probability corresponding to other sentence fragments adjacent to the sentence fragment in the text data to be recognized based on the context influence increment.

[0097] Optionally, the target model is obtained by training based on the training data set, including: determining text data in the basic corpus whose text length exceeds a preset length threshold; performing preprocessing operations on the text data whose text length exceeds the preset length threshold based on a priori knowledge vocabulary, and using the preprocessed text data to replace the corresponding text data in the basic corpus that has not been preprocessed, to obtain an updated basic corpus; based on the updated basic corpus, constructing a training data set, and using the training data set to train the initial natural language processing model to obtain the target model, wherein the training data set contains multiple training samples and label information corresponding to the training samples, and the natural language processing model includes: a text convolutional neural network model.

[0098] Optionally, based on the updated basic corpus, constructing a training data set includes: performing word segmentation processing on each text data in the updated basic corpus to obtain an initial training sample and label information corresponding to the training sample; randomly replacing a preset proportion of words in the initial training sample with homophones to obtain a first training sample; determining the pinyin data corresponding to the initial training sample and / or the first training sample to obtain a second training sample; and constructing a training data set based on the initial training sample, the first training sample and the second training sample.

[0099] It should be noted that the various modules in the above-mentioned risk data identification device can be program modules (for example, a set of program instructions that implement a certain specific function) or hardware modules. For the latter, it can be expressed in the following forms, but is not limited to this: the expression form of each of the above-mentioned modules is a processor, or the functions of each of the above-mentioned modules are implemented by a processor.

[0100] It should be noted that the risk data identification device provided in this embodiment can be used to perform Figure 2The risk data identification method shown, therefore, the relevant explanations and descriptions of the above-mentioned risk data identification method are also applicable to the embodiments of the present application and will not be repeated here.

[0101] An embodiment of the present application also provides a non-volatile storage medium, the non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the following risk data identification method or model training method by running the computer program: obtaining text data to be identified, wherein the text data to be identified is converted from voice data in a voice call; performing a preprocessing operation on the text data to be identified based on a priori knowledge vocabulary, wherein the priori knowledge vocabulary contains multiple keywords corresponding to risk characteristics, and the preprocessing operation is used to screen key phrases containing keywords in the text data to be identified; using a target model to identify the text data to be identified after the preprocessing operation to obtain an identification result, wherein the identification result includes: there is an abnormal risk in the voice call corresponding to the text data to be identified, and the voice call corresponding to the text data to be identified is normal.

[0102] Alternatively, a basic corpus is obtained, and text data in the basic corpus whose text length exceeds a preset length threshold is determined, wherein the basic corpus contains a plurality of text data obtained by speech translation, and label information corresponding to the text data, and the label information includes: a first label for characterizing that the speech corresponding to the text data has an abnormal risk, and a second label for characterizing that the speech corresponding to the text data is normal; based on a priori knowledge vocabulary, a preprocessing operation is performed on the text data whose text length exceeds the preset length threshold, wherein the priori knowledge vocabulary contains a plurality of keywords corresponding to risk features, and the preprocessing operation is used to screen key phrases containing the keywords in the text data; the text data after the preprocessing operation is used to replace the corresponding text data in the basic corpus that has not been preprocessed, to obtain an updated basic corpus; based on the updated basic corpus, an initial natural language processing model is trained to obtain a target model, wherein the target model is used to recognize the text data to be recognized and obtain a recognition result, the text data to be recognized is obtained by converting the speech data in the voice call, and the recognition result includes: the voice call corresponding to the text data to be recognized has an abnormal risk, and the voice call corresponding to the text data to be recognized is normal.

[0103] The embodiments of the present application also provide a computer program product, including a computer program, which, when executed by a processor, implements the steps of the risk data identification method or model training method described in each embodiment of the present application: obtaining text data to be identified, wherein the text data to be identified is converted from voice data in a voice call; performing a preprocessing operation on the text data to be identified based on a priori knowledge vocabulary, wherein the priori knowledge vocabulary contains multiple keywords corresponding to risk characteristics, and the preprocessing operation is used to screen key phrases containing keywords in the text data to be identified; using a target model to identify the text data to be identified after the preprocessing operation, and obtain a recognition result, wherein the recognition result includes: there is an abnormal risk in the voice call corresponding to the text data to be identified, and the voice call corresponding to the text data to be identified is normal.

[0104] Alternatively, a basic corpus is obtained, and text data in the basic corpus whose text length exceeds a preset length threshold is determined, wherein the basic corpus contains a plurality of text data obtained by speech translation, and label information corresponding to the text data, and the label information includes: a first label for characterizing that the speech corresponding to the text data has an abnormal risk, and a second label for characterizing that the speech corresponding to the text data is normal; based on a priori knowledge vocabulary, a preprocessing operation is performed on the text data whose text length exceeds the preset length threshold, wherein the priori knowledge vocabulary contains a plurality of keywords corresponding to risk features, and the preprocessing operation is used to screen key phrases containing the keywords in the text data; the text data after the preprocessing operation is used to replace the corresponding text data in the basic corpus that has not been preprocessed, to obtain an updated basic corpus; based on the updated basic corpus, an initial natural language processing model is trained to obtain a target model, wherein the target model is used to recognize the text data to be recognized and obtain a recognition result, the text data to be recognized is obtained by converting the speech data in the voice call, and the recognition result includes: the voice call corresponding to the text data to be recognized has an abnormal risk, and the voice call corresponding to the text data to be recognized is normal.

[0105] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0106] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0107] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0108] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0109] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0110] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0111] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A risk data identification method, characterized in that: include: Acquire text data to be recognized, wherein the text data to be recognized is converted from voice data in a voice call; Preprocessing the text data to be identified based on a priori knowledge word library, wherein the priori knowledge word library contains a plurality of keywords corresponding to risk features, and the preprocessing operation is used to screen key phrases containing the keywords in the text data to be identified; The target model is used to identify the text data to be identified after the preprocessing operation to obtain a recognition result, wherein the recognition result includes: there is an abnormal risk in the voice call corresponding to the text data to be identified, and the voice call corresponding to the text data to be identified is normal.

2. The risk data identification method according to claim 1, characterized in that: The method further comprises: Acquire a basic corpus, wherein the basic corpus contains a plurality of text data obtained by speech translation, and label information corresponding to the text data, wherein the label information includes: a first label for indicating that the speech corresponding to the text data has an abnormal risk, and a second label for indicating that the speech corresponding to the text data is normal; Determine the text length of the text data of each piece of the tag information as the first tag, and determine the number of keywords planned to be extracted from the text data according to the text length; According to the number of keywords, segmented words used to characterize the risk characteristics are extracted from the text data, and the segmented words are determined as the keywords to obtain the prior knowledge vocabulary.

3. The risk data identification method according to claim 2, characterized in that: After obtaining the prior knowledge word library, the method further includes: Determine the first number of data items of the text data whose tag information including the keyword is the first tag in the basic corpus; Determine the total number of text data in the basic corpus whose label information is the first label as the second number of data; The ratio of the first data items to the second data items is determined as the appearance ratio of the keyword, and the keyword level corresponding to the keyword is determined based on the ratio interval range corresponding to the appearance ratio, wherein the keyword level is used to characterize the degree of influence of the keyword on determining whether there is an abnormal risk in the voice call corresponding to the text data.

4. The risk data identification method according to claim 3, characterized in that: The preprocessing operation of the text data to be recognized according to the prior knowledge word library includes: In the case where the text length of the text data to be recognized exceeds a preset length threshold, the text data to be recognized is divided into a plurality of sentence segments according to preset characters, wherein the preset characters include: punctuation characters; Based on the prior knowledge word library, determine the keywords contained in each of the sentence fragments, and determine the target retention probability corresponding to the sentence fragment and other sentence fragments adjacent to and before the sentence fragment in the text data to be recognized according to the keyword level corresponding to the keyword, wherein the target retention probability is used to represent the probability that the sentence fragment is retained in the text data to be recognized after the preprocessing operation; According to the target retention probability, the sentence fragments in the text data to be recognized are retained to obtain the text data to be recognized after the preprocessing operation, wherein the text length of the text data to be recognized after the preprocessing operation is smaller than the text length of the text data to be recognized that has not undergone the preprocessing operation, and the length difference ratio between the text length of the text data to be recognized after the preprocessing operation and the preset length threshold is smaller than the preset length difference ratio.

5. The risk data identification method according to claim 4, characterized in that: Determining the target retention probabilities corresponding to the sentence fragment and other sentence fragments adjacent to and before the sentence fragment in the text data to be recognized includes: Determining the text length of the to-be-recognized text data that has not undergone the preprocessing operation, and determining an initial retention probability based on the text length and the preset length threshold; Determine the highest keyword level among the keyword levels corresponding to the keywords included in the sentence fragment, and determine the adjustment coefficient corresponding to the highest keyword level, wherein the higher the keyword level, the greater the influence of the keyword on determining whether the voice call corresponding to the text data has an abnormal risk; Determining the target retention probability corresponding to the sentence fragment according to the adjustment coefficient and the initial retention probability; According to the target retention probability corresponding to the sentence fragment, a context influence increment is determined, and according to the context influence increment, the target retention probabilities corresponding to other sentence fragments adjacent to the sentence fragment in the text data to be recognized are adjusted.

6. The risk data identification method according to claim 2, characterized in that: The target model is trained based on the training data set, including: Determining the text data in the basic corpus whose text length exceeds a preset length threshold; According to the prior knowledge word library, the preprocessing operation is performed on the text data whose text length exceeds the preset length threshold, and the text data after the preprocessing operation is used to replace the corresponding text data in the basic corpus that has not been preprocessed, so as to obtain the updated basic corpus; Based on the updated basic corpus, the training data set is constructed, and the initial natural language processing model is trained using the training data set to obtain the target model, wherein the training data set includes multiple training samples and the label information corresponding to the training samples, and the natural language processing model includes: a text convolutional neural network model.

7. The risk data identification method according to claim 6, characterized in that: Based on the updated basic corpus, constructing the training data set includes: Performing word segmentation processing on each piece of the text data in the updated basic corpus to obtain an initial training sample and the label information corresponding to the training sample; Randomly replacing a preset proportion of words in the initial training sample with homophones to obtain a first training sample; Determine the pinyin data corresponding to the initial training sample and / or the first training sample to obtain a second training sample; The training data set is constructed according to the initial training sample, the first training sample and the second training sample.

8. A model training method, characterized in that: include: Acquire a basic corpus, and determine text data in the basic corpus whose text length exceeds a preset length threshold, wherein the basic corpus contains a plurality of text data obtained by speech translation, and label information corresponding to the text data, wherein the label information includes: a first label for indicating that the speech corresponding to the text data has an abnormal risk, and a second label for indicating that the speech corresponding to the text data is normal; According to a priori knowledge word library, a preprocessing operation is performed on the text data whose text length exceeds a preset length threshold, wherein the priori knowledge word library contains a plurality of keywords corresponding to risk characteristics, and the preprocessing operation is used to screen key phrases containing the keywords in the text data; Using the text data after the preprocessing operation to replace the corresponding text data in the basic corpus that has not undergone the preprocessing operation, to obtain the updated basic corpus; Based on the updated basic corpus, the initial natural language processing model is trained to obtain a target model, wherein the target model is used to recognize the text data to be recognized and obtain a recognition result, wherein the text data to be recognized is converted from voice data in a voice call, and the recognition result includes: there is an abnormal risk in the voice call corresponding to the text data to be recognized, and the voice call corresponding to the text data to be recognized is normal.

9. A risk data identification device, characterized in that: include: A data acquisition module, used to acquire text data to be recognized, wherein the text data to be recognized is obtained by converting voice data in a voice call; A long text processing module, used for performing a preprocessing operation on the text data to be identified based on a priori knowledge word library, wherein the priori knowledge word library contains a plurality of keywords corresponding to risk characteristics, and the preprocessing operation is used for screening key phrases containing the keywords in the text data to be identified; The risk identification module is used to use the target model to identify the text data to be identified after the preprocessing operation to obtain an identification result, wherein the identification result includes: the voice call corresponding to the text data to be identified has an abnormal risk, and the voice call corresponding to the text data to be identified is normal.

10. An electronic device, characterized in that: include: A memory and a processor, wherein the processor is used to run a program stored in the memory, wherein when the program is run, the risk data identification method described in any one of claims 1 to 7 or the model training method described in claim 8 is executed.

11. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the risk data identification method described in any one of claims 1 to 7 or the model training method described in claim 8 by running the computer program.

12. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the risk data identification method described in any one of claims 1 to 7 or the model training method described in claim 8 are implemented.

Images