Virtual human identification method, device and equipment, readable storage medium and program product

Abstract

The invention discloses a virtual human recognition method, device and equipment, a readable storage medium and a program product, and belongs to the technical field of artificial intelligence. The virtual human identification method comprises the following steps: acquiring a conversation text to be identified between a target user and a conversation opposite end; wherein the dialogue text to be recognized comprises first dialogue content of the target user and second dialogue content of the opposite communication end; the first dialogue content is input into n sessions generated by a pre-training model, n third dialogue content output by the n sessions is obtained, the n sessions are in one-to-one correspondence with preset n types of roles, and n is an integer larger than 1; and performing similarity calculation according to the to-be-identified conversation text and the n third conversation contents, and identifying whether the conversation opposite end is a virtual human or not according to a similarity calculation result. Therefore, whether the call opposite end is the virtual person can be accurately identified.

Description

Technical Field

[0001] The present application belongs to the field of artificial intelligence technology, and specifically relates to a virtual human recognition method, device, equipment, readable storage medium and program product. Background Art

[0002] With the development of artificial intelligence (AI), AI fraud is common. Some criminals often use AI technology to process fraudulent scripts into real voices and make fraudulent calls. In related technologies, voice recognition technology is mainly used to analyze the caller's voice to identify whether the caller is a virtual person, and then identify whether the corresponding call is a fraudulent call. However, with the development of AI technology, the voice synthesized by AI technology is increasingly difficult to distinguish from the voice of a real person, resulting in the inability to accurately identify whether the other party of the call is a virtual person. Summary of the invention

[0003] The purpose of the embodiments of the present application is to provide a virtual person recognition method, device, equipment, readable storage medium and program product to solve the problem in the related art that it is not possible to accurately identify whether the other party of the call is a virtual person.

[0004] In order to solve the above technical problems, this application is implemented as follows:

[0005] In a first aspect, a virtual person recognition method is provided, comprising:

[0006] Acquire a conversation text to be identified between a target user and a call partner; wherein the conversation text to be identified includes a first conversation content of the target user and a second conversation content of the call partner;

[0007] Inputting the first dialogue content into n conversations generated by a pre-trained model respectively, to obtain n third dialogue contents output by the n conversations, wherein the n conversations correspond one-to-one to n preset roles, and n is an integer greater than 1;

[0008] A similarity calculation is performed based on the conversation text to be identified and the n third conversation contents, and based on the result of the similarity calculation, it is identified whether the call partner is a virtual person.

[0009] In a second aspect, a virtual person recognition device is provided, comprising:

[0010] A first acquisition module is used to acquire a conversation text to be identified between a target user and a call partner; wherein the conversation text to be identified includes a first conversation content of the target user and a second conversation content of the call partner;

[0011] A processing module, configured to input the first dialogue content into n conversations generated by a pre-trained model respectively, to obtain n third dialogue contents output by the n conversations, wherein the n conversations correspond one-to-one to n preset roles, and n is an integer greater than 1;

[0012] The recognition module is used to perform similarity calculation based on the conversation text to be recognized and the n third conversation contents, and to recognize whether the call partner is a virtual person based on the result of the similarity calculation.

[0013] In a third aspect, an electronic device is provided, comprising a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method described in the first aspect.

[0014] In a fourth aspect, a readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented.

[0015] In a fifth aspect, a computer program product is provided, comprising computer instructions, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0016] In an embodiment of the present application, a conversation text to be identified between a target user and a call partner is obtained; the conversation text to be identified includes a first conversation content of the target user and a second conversation content of the call partner; the first conversation content is respectively input into n conversations generated by a pre-trained model to obtain n third conversation contents output by the n conversations, wherein the n conversations correspond one-to-one to n preset roles; similarity calculation is performed based on the conversation text to be identified and the n third conversation contents, and based on the result of the similarity calculation, whether the call partner is a virtual person can be identified based on the content of replies from different types of virtual people, and the degree of similarity between the conversation text to be identified and the conversation contents of different types of virtual people can be used to identify whether the call partner is a virtual person, thereby accurately identifying whether the call partner is a virtual person and improving the accuracy of identifying virtual people. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flow chart of a virtual person recognition method provided by an embodiment of the present application;

[0018] Figure 2 is a schematic diagram of a specific dialogue text in an embodiment of the present application;

[0019] Figure 3 is a schematic diagram of m different types of large models in an embodiment of the present application;

[0020] Figure 4 It is a schematic diagram of the model processing process in the embodiment of the present application;

[0021] Figure 5 It is a structural schematic diagram of a virtual person recognition device provided in an embodiment of the present application;

[0022] Figure 6 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are 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 are within the scope of protection of this application.

[0024] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0025] In order to solve the problem in related technologies that the other party of the call cannot be accurately identified as a virtual person, the solution in this application introduces the Turing test. The Turing test (which can be called: human-computer dialogue test) is a standard for testing whether artificial intelligence has intelligence. The principle of the Turing test is as follows: K (K>1) independent testers participate in the test, and each tester conducts a natural language dialogue with the computer, in which it is not known whether the other party is a person or a machine; if the tester believes that the other party is a real person rather than a machine, then the machine passes this test; K testers correspond to K tests. If the number of times the machine passes in these K tests reaches a certain proportion, it is considered that the machine has passed the Turing test and has communication capabilities that are indistinguishable from humans.

[0026] The following, in conjunction with the accompanying drawings, describes in detail the virtual human recognition method, device, electronic device and readable storage medium provided in the embodiments of the present application through specific embodiments and their application scenarios.

[0027] See also Figure 1 , Figure 11 is a flowchart of a virtual person recognition method provided by an embodiment of the present application. The method is applied to an electronic device, which may be a user terminal or a cloud. Figure 1 As shown, the method comprises the following steps:

[0028] Step 11: Acquire the conversation text to be identified between the target user and the other party of the call; the conversation text to be identified includes the first conversation content of the target user and the second conversation content of the other party of the call;

[0029] Step 12: inputting the first dialogue content into n conversations generated by the pre-training model respectively, and obtaining n third dialogue contents output by the n conversations, wherein the n conversations correspond one-to-one to the preset n types of roles, and n is an integer greater than 1;

[0030] Step 13: Perform similarity calculation based on the conversation text to be identified and the n third conversation contents, and identify whether the call partner is a virtual person based on the result of the similarity calculation.

[0031] In the embodiment of the present application, the dialogue text to be identified may be text data converted from the voice content of a call in a real scene. The dialogue text to be identified includes the dialogue content between the target user and the other end of the call. The target user can be understood as the current user of the call end, that is, the real user who interacts with the other end of the call by voice.

[0032] The pre-trained model may be a large model, such as a large language model (LLM), an open-based and fine-tuned chat model (such as LLaMA2), a BLOOM model, a YI 34B model, etc. The pre-trained model may be trained using a large amount of known fraud data, and the known fraud data may be dialogue text data converted from call voices in real scenarios. The model training method may adopt an existing method, which is not limited thereto.

[0033] The n types of roles can be set based on actual conditions, such as including but not limited to roles such as "parents", "wife", "husband", "children", "classmates", "neighbors", and "colleagues". The n conversations are n independent conversations (session). Each conversation is an independent conversation of the pre-trained model, and each conversation corresponds to a role, such as "parents" or "neighbors". That is, different conversations are isolated from each other, and the conversation content between each conversation will not affect each other. For example, if the first conversation content of the target user is S1, the second conversation content of the other end of the call is S2, and the third conversation content generated under a certain conversation is W, then the conversation text to be recognized of S1+S2 and the conversation sample text of S1+W can be found in Figure 2 shown.

[0034] The n third dialogue contents can be understood as the n dialogue contents of the n types of virtual people corresponding to the dialogue contents of the target user obtained when the target user respectively dialogues with the n types of virtual people (which correspond to the n types of roles, i.e., the n types of virtual people correspond to the n types of roles one by one). The n third dialogue contents correspond to the n types of roles one by one, i.e., each third dialogue content corresponds to one type of role.

[0035] Optionally, when similarity is calculated based on the conversation text to be identified and the n third conversation contents, text similarity or semantic similarity may be calculated, thereby determining whether the other end of the call is a virtual person based on similar situations.

[0036] Optionally, after determining that the other party on the call is a virtual person, the target user may be reminded, such as reminding the target user that the other party on the call is most likely an AI person.

[0037] Optionally, the applicable scenarios of the embodiments of the present application include but are not limited to voice call scenarios, video call scenarios, etc. During the actual call process, if the user-side program monitors that the incoming call is from an unfamiliar number, the voice call content (such as audio, text) between the user and the caller (i.e., the other end of the call) can be synchronously sent to the cloud, and the cloud will separate the user's conversation content from the caller's conversation content for subsequent virtual human recognition process. The specific recognition process can be executed by the cloud or the user end.

[0038] The identification method in the embodiment of the present application obtains a conversation text to be identified between a target user and a call partner; the conversation text to be identified includes a first conversation content of the target user and a second conversation content of the call partner; the first conversation content is respectively input into n conversations generated by a pre-trained model to obtain n third conversation contents output by the n conversations, wherein the n conversations correspond one-to-one to n preset roles; similarity calculation is performed based on the conversation text to be identified and the n third conversation contents, and whether the call partner is a virtual person is identified based on the result of the similarity calculation, and the content of replies from different types of virtual people can be used as a starting point to identify whether it is a virtual person by using the similarity between the conversation text to be identified and the conversation content of different types of virtual people, thereby accurately identifying whether the call partner is a virtual person and improving the accuracy of identifying virtual people.

[0039] Furthermore, with the help of the solution in this application, the possibility of users being defrauded by AI synthesized human voices can be avoided, thereby greatly improving the security of users' voice calls.

[0040] In the embodiment of the present application, in order to improve the model training efficiency, model training can be performed with the help of fine-tuning. The training process of the pre-trained model may include:

[0041] Obtain a first training data set and a second training data set; wherein the first training data set includes a plurality of real conversation texts, which may be obtained based on call speech conversion in a real scene; the second training data set includes n data subsets obtained based on a virtual person, the n data subsets correspond one-to-one to the n types of roles (such as "parents", "wife", "husband", "children", "classmates", "neighbors", "colleagues", etc.), and each data subset contains a virtual conversation text formed by the virtual person in one of the roles, that is, each data subset corresponds to one type of role;

[0042] Performing model training using the first training data set to obtain an initial training model; for example, the first training data set may be used to perform unsupervised model training;

[0043] Each data subset is used to fine-tune the initial training model (such as supervised fine-tuning) to obtain the pre-trained model; that is, each data subset in the n data subsets is used to fine-tune the initial training model, so that the pre-trained model can generate n conversations corresponding to n types of roles.

[0044] In this way, since it is easier to obtain a large amount of real dialogue texts in real scenes, but it is not easy to obtain a large amount of virtual dialogue texts of virtual people in each type of role, the model training efficiency can be improved by using fine-tuning operations, that is, first using multiple real dialogue texts for initial model training and then fine-tuning.

[0045] Optionally, the conversation text to be identified and the conversation contents of different types of virtual persons may be compared based on semantic similarity. The above-mentioned similarity calculation based on the conversation text to be identified and the n third conversation contents, and identifying whether the call counterpart is a virtual person based on the result of the similarity calculation may include:

[0046] Combining the first conversation content with each of the third conversation contents to obtain n conversation sample texts;

[0047] Calculate the semantic similarity between the to-be-recognized dialogue text and each of the dialogue sample texts to obtain n first similarity values; for example Figure 2 As shown, the semantic similarity between the dialogue text to be recognized of S1+S2 and the dialogue sample text of S1+W can be calculated;

[0048] Determine a first number of first similarity values ​​that are greater than or equal to a first threshold; for example, the first threshold may be preset based on actual needs; when a first similarity value is greater than or equal to the first threshold, it may indicate that the conversation sample text corresponding to the first similarity value is similar / extremely similar to the conversation text to be identified, that is, it indicates that the conversation sample text is very likely to be generated by a virtual person;

[0049] Calculating a ratio of the first number to the n to obtain a first ratio;

[0050] When the first ratio is greater than or equal to the second threshold, that is, the number of dialogue sample texts that meet the requirements (i.e., similar to the dialogue text to be recognized) among the n dialogue sample texts reaches a certain proportion, it is determined that the call counterpart is a virtual person; or, when the first ratio is less than the second threshold, that is, the number of dialogue sample texts that meet the requirements (i.e., similar to the dialogue text to be recognized) among the n dialogue sample texts does not reach a certain proportion, it is determined that the call counterpart is not a virtual person. The second threshold can be preset based on actual needs.

[0051] In this way, we can use the principle of the Turing test to consider the overall context. When the proportion of conversation sample texts of different types of virtual people that are similar to the conversation text to be identified reaches a certain level, it can be determined that the other party on the call is a virtual person, thereby improving the accuracy of identifying virtual people.

[0052] Optionally, the calculation process of the semantic similarity may include:

[0053] Calculate the first semantic vector of each sentence in the dialogue text to be identified, and calculate the second semantic vector of each sentence in each dialogue sample text; wherein, in order to ensure the accuracy of the vector representation, before calculating the semantic vector of each sentence, the corresponding text may be preprocessed by word segmentation, stop word removal, part-of-speech tagging, etc., to extract keywords and sentence structure information; then, a pre-trained word embedding model, such as Word2Vec, GloVe, FastText, etc., may be used to convert each word obtained by the preprocessing into a dense vector representation, and the word vectors of each sentence are averaged to obtain the semantic vector representation of each sentence;

[0054] Calculating the semantic similarity between the to-be-recognized dialogue text and a corresponding sentence in each of the dialogue sample texts according to the first semantic vector and the second semantic vector;

[0055] The semantic similarities between the dialogue text to be recognized and the corresponding sentences in each of the dialogue sample texts are weighted averaged to obtain the n first similarity values.

[0056] For example, the following formula can be used to calculate the similarity between two sentence vectors using cosine similarity:

[0057] sim(Sen1,Sen2)=dot(v1,v2)÷(norm(v1)×norm(v2))

[0058] Among them, Sen1 and Sen2 represent two sentences whose similarity is to be calculated, v1 is the semantic vector of sentence Sen1, and v2 is the semantic vector of sentence Sen2; dot(v1, v2) represents the dot product of vectors v1 and v2, and the dot product is calculated by multiplying the corresponding elements of the two vectors and then summing them; norm(v1) represents the length of vector v1, and norm(v2) represents the length of vector v2.

[0059] Optionally, the conversation text to be identified and the conversation contents of different types of virtual persons may be compared based on text similarity. The above-mentioned similarity calculation based on the conversation text to be identified and the n third conversation contents, and identifying whether the call counterpart is a virtual person based on the result of the similarity calculation may include:

[0060] Calculating the text similarity between the second dialogue content in the dialogue text to be identified and each of the third dialogue contents respectively to obtain n second similarity values; for example, the text similarity between the second dialogue content and each of the third dialogue contents may be calculated respectively using vector representation, such as by calculating the similarity between the vector representation of the second dialogue content and the vector representation of each of the third dialogue contents to obtain the text similarity between the two;

[0061] Determine a second number of the second similarity values ​​that are greater than or equal to a third threshold; for example, the third threshold may be preset based on actual needs; when a second similarity value is greater than or equal to the first threshold, it may indicate that the second conversation text corresponding to the second similarity value is similar / extremely similar to the third conversation content, that is, it indicates that the corresponding second conversation content is highly likely to be generated by a virtual person;

[0062] Calculating a ratio of the second number to the n to obtain a second ratio;

[0063] When the second ratio is greater than or equal to a fourth threshold, that is, the third conversation content similar to the second conversation content in the n third conversation contents reaches a certain proportion, it is determined that the call counterpart is a virtual person; or when the second ratio is less than the fourth threshold, that is, the third conversation content extremely similar to the second conversation content in the n third conversation contents does not reach a certain proportion, it is determined that the call counterpart is not a virtual person. The fourth threshold can be preset based on actual needs.

[0064] In this way, we can use the principle of the Turing test and consider it from the perspective of text similarity. When the conversation content of different types of virtual people that is similar to the conversation content of the other party reaches a certain proportion, it is determined that the other party is a virtual person, thereby improving the accuracy of virtual person recognition.

[0065] In the embodiment of the present application, in order to further improve the recognition accuracy of the virtual person, multiple different types of pre-trained models can be used to complete the virtual person recognition. The above-mentioned inputting the first conversation content into the n conversations generated by the pre-trained model respectively to obtain the n third conversation contents output by the n conversations may include:

[0066] Generate n sessions in each of the m pre-trained models; wherein the m pre-trained models have different model architectures, and the m pre-trained models implement the same function, and m is an integer greater than 1; and the n sessions correspond one-to-one to the preset n types of roles;

[0067] The first conversation content is respectively input into the n conversations generated by each pre-trained model to obtain n third conversation contents output by the n conversations of each pre-trained model; that is, under m pre-trained models, m*n conversations can be obtained, and m*n third conversation contents can be obtained accordingly.

[0068] In this way, through m different types of pre-training models, the conversation content based on different types of virtual people can be increased, that is, the statistical test results can be increased, and then when the Turing test principle is used to identify virtual people, the recognition accuracy can be improved, and the recognition accuracy of virtual people can be improved.

[0069] It should be noted that the processing functions to be implemented by the m different types of pre-trained models are the same, which is to obtain the dialogue content under the corresponding role. The m different types of pre-trained models can be obtained by using different model architectures, such as LLM, LLaMA2, BLOOM model, YI 34B model, etc. For each of the m different types of pre-trained models, it can be obtained by using the above-mentioned model training process, which will not be repeated here.

[0070] In an optional implementation, if Figure 3 As shown, the same training data set can be used to train m different types of models, such as large model 1, large model 2, ..., large model m.

[0071] In an optional implementation, if Figure 4 As shown, n conversations (corresponding to n types of roles) can be generated in each of the m different types of pre-trained models, and the first conversation content (such as Figure 2 S1) are respectively input into the n conversations generated by each pre-training model to obtain the n third dialogue contents (such as Figure 3 Middle W).

[0072] In an optional implementation, if m different types of pre-trained models are used to complete virtual human recognition, since each conversation corresponds to a third conversation content W, there are a total of m models and each model has n conversations, so n*m third conversation contents W can be obtained, and then n*m conversation sample texts S1+W can be obtained; then, the semantic similarity between the conversation text to be identified S1+S2 and each S1+W can be calculated respectively to obtain n*m semantic similarity values ​​X; if a certain X is greater than a preset similarity threshold k1, it means that the corresponding two texts are similar; after obtaining n*m semantic similarity values ​​X, the number F of similarity values ​​X greater than k1 can be determined, that is, the number F of conversations with semantic similarity values ​​X greater than k1 is obtained; then, according to the Turing test principle, if the ratio of F to n*m is greater than or equal to the preset threshold k2 (for example, 30%), it can be determined that S2 is most likely an AI reply, and then it can be determined that the other end of the call is most likely an AI person, and the user can be reminded that the person talking to him is most likely an AI person.

[0073] It should be noted that the virtual person identification method provided in the embodiment of the present application can be executed by a virtual person identification device, or a control module in the virtual person identification device for executing the virtual person identification method. In the embodiment of the present application, the virtual person identification device executing the virtual person identification method is taken as an example to illustrate the virtual person identification device provided in the embodiment of the present application.

[0074] See also Figure 5 , Figure 5 is a schematic diagram of the structure of a virtual person recognition device provided in an embodiment of the present application, and the device is applied to electronic devices such as Figure 5 As shown, the virtual human recognition device 50 includes:

[0075] A first acquisition module 51 is used to acquire a conversation text to be identified between a target user and a call partner; wherein the conversation text to be identified includes a first conversation content of the target user and a second conversation content of the call partner;

[0076] A processing module 52 is used to input the first dialogue content into n conversations generated by a pre-trained model respectively, to obtain n third dialogue contents output by the n conversations, wherein the n conversations correspond one-to-one to n preset roles, and n is an integer greater than 1;

[0077] The recognition module 53 is used to perform similarity calculation based on the conversation text to be recognized and the n third conversation contents, and to recognize whether the call partner is a virtual person based on the result of the similarity calculation.

[0078] Optionally, the processing module 52 is specifically used for:

[0079] Generate n conversations in each of m pre-trained models, respectively; wherein the m pre-trained models have different model architectures, and the m pre-trained models implement the same functions, and m is an integer greater than 1; input the first conversation content into the n conversations generated by each pre-trained model, respectively, to obtain n third conversation contents output by the n conversations of each pre-trained model.

[0080] Optionally, the identification module 53 includes:

[0081] A first calculation unit is used to combine the first dialogue content with each of the third dialogue contents to obtain n dialogue sample texts, and calculate the semantic similarity between the dialogue text to be identified and each of the dialogue sample texts to obtain n first similarity values;

[0082] A first determining unit, configured to determine a first number of the first similarity values ​​that are greater than or equal to a first threshold;

[0083] a second calculation unit, configured to calculate a ratio of the first number to the n to obtain a first ratio;

[0084] The second determination unit is used to determine that the call counterpart is a virtual person when the first ratio is greater than or equal to a second threshold; or to determine that the call counterpart is not a virtual person when the first ratio is less than the second threshold.

[0085] Optionally, the first calculation unit is specifically used to: calculate the first semantic vector of each sentence in the dialogue text to be recognized, and calculate the second semantic vector of each sentence in each of the dialogue sample texts; calculate the semantic similarity between the dialogue text to be recognized and the corresponding sentence in each of the dialogue sample texts based on the first semantic vector and the second semantic vector; and perform weighted averaging of the semantic similarities between the dialogue text to be recognized and the corresponding sentence in each of the dialogue sample texts to obtain the n first similarity values.

[0086] Optionally, the identification module 53 includes:

[0087] A third calculation unit is used to respectively calculate the text similarity between the second dialogue content in the dialogue text to be recognized and each of the third dialogue contents to obtain n second similarity values;

[0088] a third determining unit, configured to determine a second number of said second similarity values ​​that are greater than or equal to a third threshold;

[0089] a fourth calculating unit, configured to calculate a ratio of the second number to the n to obtain a second ratio;

[0090] The fourth determination unit is used to determine that the call counterpart is a virtual person when the second ratio is greater than or equal to a fourth threshold; or to determine that the call counterpart is not a virtual person when the second ratio is less than the fourth threshold.

[0091] Optionally, the virtual human recognition device 50 further includes:

[0092] A second acquisition module is used to acquire a first training data set and a second training data set; wherein the first training data set includes a plurality of real dialogue texts, and the second training data set includes n data subsets obtained based on a virtual person, wherein the n data subsets correspond one-to-one to the n types of roles, and each of the data subsets contains a virtual dialogue text formed by the virtual person in one of the types of roles;

[0093] A training module, used to perform model training using the first training data set to obtain an initial training model;

[0094] The adjustment module is used to use each of the data subsets to fine-tune the initial training model to obtain the pre-trained model.

[0095] The virtual person recognition device 50 of the embodiment of the present application can realize the above Figure 1 The various processes of the method embodiment shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0096] Optional, such as Figure 6 As shown, the embodiment of the present application also provides an electronic device 60, including a processor 61, a memory 62, and a program or instruction stored in the memory 62 and executable on the processor 61. When the program or instruction is executed by the processor 61, each process of the above-mentioned virtual human recognition method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0097] The present application also provides a computer program product, including computer instructions, which can achieve the above-mentioned Figure 1 The various processes of the virtual human recognition method embodiment shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0098] The embodiment of the present application also provides a readable storage medium on which a program or instruction is stored. When the program or instruction is executed by a processor, the various processes of the above-mentioned virtual human recognition method embodiment can be implemented and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0099] Computer readable media include permanent and non-permanent, removable and non-removable media, and can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0100] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0101] 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.

[0102] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a service classification device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0103] 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 virtual human recognition method, characterized in that: include: Acquire a conversation text to be identified between a target user and a call partner; wherein the conversation text to be identified includes a first conversation content of the target user and a second conversation content of the call partner; Inputting the first dialogue content into n conversations generated by a pre-trained model respectively, to obtain n third dialogue contents output by the n conversations, wherein the n conversations correspond one-to-one to n preset roles, and n is an integer greater than 1; A similarity calculation is performed based on the conversation text to be identified and the n third conversation contents, and based on the result of the similarity calculation, it is identified whether the call partner is a virtual person.

2. The method according to claim 1, characterized in that: The step of inputting the first conversation content into n conversations generated by a pre-trained model respectively to obtain n third conversation contents output by the n conversations includes: Generate n sessions in each of m pre-trained models respectively; wherein the m pre-trained models have different model architectures, and the m pre-trained models implement the same functions, and m is an integer greater than 1; The first conversation content is respectively input into the n conversations generated by each pre-trained model to obtain n third conversation contents output by the n conversations of each pre-trained model.

3. The method according to claim 1, characterized in that The performing similarity calculation based on the to-be-identified conversation text and the n third conversation contents, and identifying whether the call counterpart is a virtual person based on the result of the similarity calculation, includes: Combining the first conversation content with each of the third conversation contents to obtain n conversation sample texts; Calculating the semantic similarity between the to-be-recognized dialogue text and each of the dialogue sample texts to obtain n first similarity values; determining a first number of said first similarity values ​​that are greater than or equal to a first threshold; Calculating a ratio of the first number to the n to obtain a first ratio; When the first ratio is greater than or equal to a second threshold, it is determined that the call counterpart is a virtual person; or, when the first ratio is less than the second threshold, it is determined that the call counterpart is not a virtual person.

4. The method according to claim 3, characterized in that The calculating the semantic similarity between the to-be-recognized dialogue text and each of the dialogue sample texts to obtain n first similarity values ​​includes: Calculating a first semantic vector for each sentence in the to-be-recognized dialogue text, and calculating a second semantic vector for each sentence in each of the dialogue sample texts; Calculating the semantic similarity between the to-be-recognized dialogue text and a corresponding sentence in each of the dialogue sample texts according to the first semantic vector and the second semantic vector; The semantic similarities between the dialogue text to be recognized and the corresponding sentences in each of the dialogue sample texts are weighted averaged to obtain the n first similarity values.

5. The method according to claim 1, characterized in that The performing similarity calculation based on the to-be-identified conversation text and the n third conversation contents, and identifying whether the call counterpart is a virtual person based on the result of the similarity calculation, includes: respectively calculating the text similarity between the second dialogue content in the dialogue text to be identified and each of the third dialogue contents to obtain n second similarity values; determining a second number of said second similarity values ​​that are greater than or equal to a third threshold; Calculating a ratio of the second number to the n to obtain a second ratio; When the second ratio is greater than or equal to a fourth threshold, it is determined that the call counterpart is a virtual person; or, when the second ratio is less than the fourth threshold, it is determined that the call counterpart is not a virtual person.

6. The method according to claim 1, characterized in that The method further comprises: Acquire a first training data set and a second training data set; wherein the first training data set includes a plurality of real dialogue texts, and the second training data set includes n data subsets obtained based on a virtual person, the n data subsets correspond one-to-one to the n types of roles, and each of the data subsets contains a virtual dialogue text formed by the virtual person in one of the types of roles; Performing model training using the first training data set to obtain an initial training model; Each of the data subsets is used to fine-tune the initial training model to obtain the pre-training model.

7. A virtual person recognition device, characterized in that: include: A first acquisition module is used to acquire a conversation text to be identified between a target user and a call partner; wherein the conversation text to be identified includes a first conversation content of the target user and a second conversation content of the call partner; A processing module, configured to input the first dialogue content into n conversations generated by a pre-trained model respectively, to obtain n third dialogue contents output by the n conversations, wherein the n conversations correspond one-to-one to n preset roles, and n is an integer greater than 1; The recognition module is used to perform similarity calculation based on the conversation text to be recognized and the n third conversation contents, and to recognize whether the call partner is a virtual person based on the result of the similarity calculation.

8. An electronic device, characterized in that: The method comprises a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method according to any one of claims 1 to 6.

9. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 6.

Images