Information processing method, computer equipment and computer program product

Abstract

The invention discloses an information processing method, computer equipment and a computer program product. The information processing method comprises the following steps: acquiring a dialogue text corresponding to a call record; determining a first detection result for the call recording based on a matching result of the dialogue text and a target keyword; based on the dialogue text, a second detection result for the call recording is obtained through a large language model, and the large language model is obtained after adjustment is conducted according to sample data of the field to which the call belongs; and based on the first detection result and the second detection result, obtaining a quality inspection result for the call recording. According to the method, the accuracy of quality inspection of the call recording can be improved.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and more specifically, to an information processing method, a computer device, and a computer program product. Background Art

[0002] In all walks of life, as the number of user groups grows, voice calls are usually made with users in order to conduct business or provide better services to users. In addition, in order to ensure compliance and service quality during voice calls, quality inspection of voice conversations is usually performed. However, in related technologies, the accuracy of quality inspection of voice conversations needs to be improved. Summary of the invention

[0003] The present application proposes an information processing method, a computer device and a computer program product, which can improve the accuracy of quality inspection of call recordings.

[0004] In a first aspect, an embodiment of the present application provides an information processing method, the method comprising: obtaining a conversation text corresponding to a call recording; determining a first detection result for the call recording based on a matching result between the conversation text and a target keyword; obtaining a second detection result for the call recording based on the conversation text and through a large language model, wherein the large language model is obtained after adjustment based on sample data of the field to which the call belongs; and obtaining a quality inspection result for the call recording based on the first detection result and the second detection result.

[0005] In the second aspect, the embodiment of the present application provides an information processing device, including: a text acquisition module, a first detection module, a second detection module, and a quality inspection result acquisition module. The text acquisition module is used to acquire the conversation text corresponding to the call recording; the first detection module is used to determine the first detection result for the call recording based on the matching result between the conversation text and the target keyword; the second detection module is used to acquire the second detection result for the call recording based on the conversation text and through a large language model, and the large language model is obtained after adjustment based on sample data of the field to which the call belongs; the quality inspection result determination module is used to obtain the quality inspection result for the call recording based on the first detection result and the second detection result.

[0006] In a third aspect, an embodiment of the present application provides an electronic device, comprising: one or more processors; a memory; one or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more applications are configured to execute the information processing method provided in the first aspect above.

[0007] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which a program code is stored. The program code can be called by a processor to execute the information processing method provided in the first aspect above.

[0008] In a fifth aspect, an embodiment of the present application provides a computer program product, which includes a computer program, and the computer program is stored in a storage medium; a processor of a computer device reads the computer program from the storage medium, and the processor executes the computer program, so that the computer device executes the information processing method provided in the first aspect above.

[0009] The solution provided by the present application obtains the conversation text corresponding to the call recording, determines the first detection result for the call recording based on the matching result between the conversation text and the target keyword, obtains the second detection result for the call recording based on the conversation text and through a large language model, and the large language model is obtained after adjustment based on the sample data of the field to which the above call belongs, and obtains the quality inspection result for the call recording based on the first detection result and the second detection result. As a result, when the call recording is quality inspected, the call recording is not only detected by matching with the target keyword, but also detected by the large language model obtained after adjustment based on the sample data of the field to which the call belongs, thereby improving the accuracy of the determined quality inspection result. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0011] Figure 1 A schematic diagram of an application environment provided by an embodiment of the present application is shown.

[0012] Figure 2 A schematic flow chart of an information processing method according to an embodiment of the present application is shown.

[0013] Figure 3 A flow chart of an information processing method according to another embodiment of the present application is shown.

[0014] Figure 4 A schematic diagram of an interface provided in an embodiment of the present application is shown.

[0015] Figure 5 A schematic flow chart of an information processing method according to another embodiment of the present application is shown.

[0016] Figure 6 A flow chart of an information processing method according to another embodiment of the present application is shown.

[0017] Figure 7 A schematic flow chart of an information processing method according to yet another embodiment of the present application is shown.

[0018] Figure 8 A flowchart of an information processing method according to yet another embodiment of the present application is shown.

[0019] Fig. 9 A schematic diagram showing the architecture layers involved in the information processing method provided in an embodiment of the present application.

[0020] Fig.10 A schematic diagram showing the technical architecture involved in the information processing method provided in an embodiment of the present application.

[0021] Fig.11 A block diagram of an information processing device according to an embodiment of the present application is shown.

[0022] Fig.12 It is a block diagram of a computer device according to an embodiment of the present application for executing an information processing method according to an embodiment of the present application.

[0023] Fig.13 It is a storage unit of an embodiment of the present application for storing or carrying a program code for implementing an information processing method according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] 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.

[0025] Quality inspection of customer service conversations is a very important part of the business operation process. It is an important part of the business to improve service quality, enhance customer relationship management, optimize internal processes and ensure compliance. Especially in the financial industry, post-loan collection by phone is an important part of financial institutions' risk management. With the increase in financial industry supervision, the compliance requirements for post-loan collection behavior are becoming more stringent. Enterprises must strictly control and identify behaviors in the collection process to ensure that the business complies with relevant laws, regulations and industry standards.

[0026] When conducting quality inspections on customer service call recordings, they mainly rely on manual monitoring and evaluation. Faced with massive call data, it is impossible for humans to achieve 100% coverage, and they can only conduct sampling inspections according to a certain proportion. This is not only inefficient, but the accuracy of the quality inspection results is also affected by the personal judgment of the quality inspectors, and is easily disturbed by emotions, experience and prejudice, resulting in inaccurate results.

[0027] Of course, there are also semi-automatic speech quality inspection solutions, which are based on Automatic Speech Recognition (ASR) technology to convert speech into text, and then combine sensitive word matching rules and manual review for quality inspection. However, ASR to text has a certain recognition error rate. For example, there will be many proper nouns, accents, incoherent speech, and recording of side sounds, which will lead to low quality of the generated text, resulting in deviations in quality inspection results when using keyword matching, and the accuracy of the text content cannot be fully guaranteed. In this case, keyword matching will result in lower accuracy, and it is impossible to deeply analyze the quality characteristics of the text content, making it difficult to fully evaluate the text quality.

[0028] In response to the above problems, the inventors have proposed the information processing method, device, electronic device and storage medium provided in the embodiments of the present application, which can realize the quality inspection of call recordings, not only by matching with target keywords, but also by using a large language model obtained by adjusting sample data of the field to which the call belongs, thereby improving the accuracy of the determined quality inspection results. The specific information processing method is described in detail in the subsequent embodiments.

[0029] The following first introduces the scenarios involved in the embodiments of the present application.

[0030] like Figure 1 As shown, in Figure 1 The scene shown includes a server 10 and an electronic device 20. The server 10 may be a server (network access server), a server cluster (cloud server) consisting of several servers, or a cloud computing center (database server). The electronic device 20 may be any electronic device with communication and storage functions, including but not limited to a PC (Personal Computer), a PDA (tablet computer), a smart TV, a smart phone, a smart wearable device, or other smart communication devices with network connection functions.

[0031] The server 10 and the electronic device 20 can interact through the network to receive or send information. In an embodiment of the present application, the server 10 can obtain the conversation text corresponding to the call recording of the customer service call, determine the first detection result for the call recording based on the matching result between the conversation text and the target keyword, and obtain the second detection result for the call recording based on the conversation text and through the large language model. The large language model is obtained after adjustment based on the sample data of the field to which the above call belongs, and obtain the quality inspection result for the call recording based on the first detection result and the second detection result. After obtaining the quality inspection result for the call recording, the server 10 can send the quality inspection result to the electronic device 20 corresponding to the quality inspector so that the quality inspector can view and review the quality inspection result.

[0032] The information processing method provided in the embodiment of the present application is described in detail below with reference to the accompanying drawings.

[0033] See also Figure 2 , Figure 2 The following is a flow chart of an information processing method provided by an embodiment of the present application. Figure 2 The process shown is described in detail, and the information processing method may specifically include the following steps:

[0034] Step S110: Obtain the conversation text corresponding to the call recording.

[0035] In an embodiment of the present application, a computer device may obtain a conversation text corresponding to a call recording so as to perform a quality check on the call recording based on the obtained conversation text.

[0036] In some embodiments, the above call recording may be a recording of any call that needs to be quality checked. For example, when a customer service call needs to be quality checked, the above call recording may be a recording of a call when a customer service staff member provides customer service; for another example, when a call between a teacher and a student needs to be quality checked, the above call recording may be a recording of a call when the teacher and the student communicate.

[0037] In some implementations, the computer device may store call recordings that need to be quality checked. When performing the quality check on the call recordings, the computer device may obtain the call recordings that currently need to be quality checked from the stored call recordings.

[0038] In a possible implementation, the quality inspection scenario for call recordings may be a scenario for quality inspection of call recordings during customer service. The computer device may store call recordings of each customer service staff when providing customer service. When quality inspection is required for any customer service staff, the computer device may obtain the call recording of the customer service staff from the stored call recordings. In this implementation, when each customer service staff uses a call device to provide customer service, the customer staff's call device may record the call in real time, and send the call recording to the computer device after the call ends.

[0039] Of course, call recordings that need to be quality checked may also be stored in other devices. When the computer device performs quality check on the call recordings, the computer device may obtain the call recordings that currently need to be quality checked from other devices.

[0040] For example, in the above scenario of quality inspection of call recordings during customer service, when each customer service staff provides customer service, the call recordings, when any customer service staff needs to be quality inspected, the computer device can obtain the call recordings of the customer service staff from other devices. For example, the above computer device providing quality inspection services is the first server, and the device storing the call recordings of each customer service staff is the second server, then the first server can obtain the above call recordings from the second server. In this implementation, when each customer service staff uses the call device to provide customer service, the customer staff's call device can record in real time, and send the call recordings to the above other devices after the call ends.

[0041] In some implementations, call recordings that require quality inspection may be stored in the call devices of the persons having the call, and when the computer device needs to perform quality inspection on the call recordings, it may obtain the call recordings that currently require quality inspection from these call devices.

[0042] For example, in the above scenario of quality inspection of call recordings during customer service, the call recordings of each customer service staff can be stored in the call devices of each customer service staff. When the computer device needs to perform quality inspection on any customer service staff, the computer device can obtain the call recording from the call device of the customer service staff. Among them, the call device of the customer service staff can actively turn on the recording function and record the call process when it detects that the customer service staff is providing customer service.

[0043] In some implementations, after acquiring the above call recording, the computer device converts the call recording into text using automatic speech recognition technology to obtain the above conversation text. Considering that the quality inspection mainly needs to detect the speech content, the speech content of different users can be distinguished when converting the above call recording into conversation text to generate a conversation text of multiple rounds of conversations sorted in chronological order.

[0044] Step S120: Determine a first detection result for the call recording based on the matching result between the conversation text and the target keyword.

[0045] In the embodiment of the present application, after the computer device obtains the conversation text corresponding to the above call recording, it can match the conversation text with the target keyword, and determine the detection result (as the first detection result) for the call recording based on the matching result of the conversation text and the target keyword. Among them, the target keyword can be a pre-set keyword corresponding to the target behavior to be detected. The target behavior to be detected in different quality inspection scenarios is different, so the target keyword in different quality inspection scenarios can be different.

[0046] In addition, for the same quality inspection scenario, the target keywords in different fields can also be different. For example, in the quality inspection scenarios of customer service calls in different fields, the target keywords corresponding to different fields can be different. For example, in the field of loan collection in the financial industry, the target keywords can be sensitive words such as lawyer and contact village committee, which may indicate illegal collection behavior. In the field of business promotion, the target keywords can be sensitive words such as best, most excellent, and first brand, which may induce users to buy.

[0047] In some implementations, the computer device may traverse each target keyword for the above conversation text, check whether each target keyword appears in the conversation text, and then obtain a matching result between the conversation text and the target keyword. The computer device matches the target keyword for the above conversation text, and the matching may be performed by algorithms such as word vector matching and BERT model, and the specific matching algorithm may not be limited.

[0048] In some implementations, the above conversation text includes the speech content of different speakers in the call, and the speech content of different speakers can be matched with different keywords. In other words, different keywords can be set for different speakers.

[0049] Exemplarily, in the scenario of quality inspection of call recordings of customer service, the above conversation text may include the speech content of the customer service staff and the speech content of the user. For the speech content of the customer service staff and the speech content of the user, the speech content can be matched with different keywords. In other words, different target keywords can be determined for customer service staff and users respectively. It is understandable that in some cases, the customer service staff may not explicitly say the target keywords set above, but say other words that violate the rules. In this case, it can be determined whether there is a violation based on the content of the user's reply during the call. Therefore, target keywords can also be set for the user's speech content, so as to more comprehensively detect the call recording and improve the accuracy of quality inspection of the call recording.

[0050] In some embodiments, when the computer device determines the first detection result for the call recording based on the above matching results, it can determine whether the call recording passes the detection (i.e., whether the target behavior exists) and other results as the above first detection result. It can be understood that when the target keyword exists in the above conversation text, the vocabulary of the target behavior exists in the conversation text, so it can be determined that the target behavior exists in the call recording; when the target keyword does not exist in the above conversation text, the vocabulary of the target behavior does not exist in the conversation text, so it can be determined that the target behavior does not exist in the call recording. Of course, the determined first detection result can also include the original words where the target keyword in the conversation text is located.

[0051] Step S130: Based on the conversation text, a second detection result for the call recording is obtained through a large language model, where the large language model is adjusted according to sample data of the field to which the call belongs.

[0052] In an embodiment of the present application, for the call text corresponding to the above call recording, the detection result for the above call recording (as the second detection result) can also be obtained through a large language model. Among them, the large language model is a model based on deep learning and natural language processing technology. It uses a large amount of text data for training to learn language understanding and generation capabilities. The large language model can handle complex natural language tasks, such as text classification, question and answer, dialogue, etc. The above large language model can be obtained after adjustment based on the sample data of the field to which the above call belongs, thereby enabling the large language model to adapt to the detection of target behaviors in the field to which the above call belongs. Since the large language model has strong understanding and analysis capabilities, it can extract rich features for the text content, and then accurately identify whether there is target behavior in the call recording.

[0053] In some implementations, the above large language model can be deployed in the above computer device. When the computer device needs to detect the call recording, it can detect the above conversation text by calling the locally deployed large language model.

[0054] In a possible implementation, when a call recording is detected by a large language model, the above conversation text and prompt words (prompt) can be input into the large language model to obtain the detection result output by the large language model, and the detection result output by the large language model is used as the second detection result. Among them, the prompt word refers to the input text or instruction provided to the model, which is used to guide the large language model to generate a specific type of response. The prompt word converts the natural language text into machine-readable intent and embedding vectors, so that the large model can understand and execute human instructions. Specifically, the prompt word can be a question, a description, a task description, or even a part of the conversation history record, etc. Its function is to guide the model to generate an expected response or complete a specific task. In this implementation, the prompt word can be used to prompt the large language model to detect the target behavior for the input text content. For example, it can be "detect whether there is a target behavior in this conversation."

[0055] In some implementations, the above large language model can also be deployed in other servers. When the computer device needs to detect the call recording, the computer device can generate a request for requesting the server to perform detection based on the above conversation text, and send the generated request to the server; correspondingly, the server can receive the request, respond to the request, and detect the conversation text carried in the request through the deployed large language model.

[0056] In some implementations, the adjustment of the large language model may be further fine-tuning of the pre-trained large language model. Fine-tuning refers to the process of further training the model using a new, task-specific data set based on a pre-trained large deep learning model, and in the embodiment of the present application, the large language model is fine-tuned based on the sample data of the field to which the above call belongs, so that the large language model obtained after fine-tuning can adapt to the detection of target behaviors in call recordings in this field.

[0057] Step S140: Based on the first detection result and the second detection result, obtain a quality inspection result for the call recording.

[0058] In the embodiment of the present application, after the computer device obtains the above first detection result and the second detection result, it can obtain the quality inspection result for the call recording based on the above first detection result and the second detection result. Therefore, based on the detection results obtained by the two methods of keyword matching and large language model recognition, the final quality inspection result of the above call recording is determined, which can improve the accuracy of quality inspection of customer call recordings.

[0059] In some embodiments, when the quality inspection result for the call recording is obtained based on the first detection result and the second detection result, it can be determined whether the target behavior exists in the call recording according to the first detection result and the second detection result as the quality inspection result for the call recording. If the first detection result indicates that the target behavior exists in the call recording (or indicates that the target keyword exists in the above conversation text), and the second detection result indicates that the target behavior exists in the call recording, then it can be determined that the target behavior exists in the above call recording; if the first detection result indicates that the target behavior does not exist in the call recording, and / or the second detection result indicates that the target behavior does not exist in the call recording, then it can be determined that the target behavior does not exist in the above call recording.

[0060] In some embodiments, when obtaining the quality inspection result of the call recording based on the first test result and the second test result, the quality inspection result for the above call recording can also be generated directly based on the first test result and the second test result, and the generated quality inspection result may include the above first test result and the second test result.

[0061] The information processing method provided in the embodiment of the present application, when performing quality inspection on call recordings, not only detects the call recordings by matching with target keywords, but also detects the call recordings by using a large language model obtained by adjusting sample data of the field to which the call belongs, thereby improving the accuracy of the determined quality inspection results.

[0062] See also Figure 3 , Figure 3 A schematic diagram of a flow chart of an information processing method provided by another embodiment of the present application is shown. Figure 3 The process shown is described in detail, and the information processing method may specifically include the following steps:

[0063] Step S210: Obtain the conversation text corresponding to the call recording.

[0064] Step S220: Based on the matching result between the conversation text and the target keyword, determine a first detection result for the call recording, wherein the first detection result includes whether the target keyword exists in the conversation text and the position of the target keyword in the conversation text.

[0065] Step S230: Based on the conversation text, a second detection result for the call recording is obtained through a large language model, wherein the large language model is adjusted according to sample data of the field to which the call belongs, and the second detection result includes a target score for the call recording and a target behavior in the call recording, wherein the target score is used to characterize the probability of the existence of the target behavior.

[0066] In the embodiment of the present application, steps S210 to S230 may refer to the contents of other embodiments and will not be described in detail here.

[0067] Step S240: Based on the first detection result and the second detection result, a quality inspection report of the call recording is generated, the quality inspection report including whether the target keyword exists in the conversation text, the position of the target keyword in the conversation text, the target score and behavior information of the existing target behavior.

[0068] In an embodiment of the present application, the above first detection result determined based on the matching result between the conversation text and the target keyword may include whether the target keyword exists in the conversation text and the position of the target keyword in the conversation text. That is to say, after matching the conversation text with the target keyword, it can be determined based on the matching result whether the target keyword exists in the conversation text; if the target keyword exists in the conversation text, the position of the target keyword in the conversation text can also be determined, for example, the original content of the target keyword can be determined.

[0069] In addition, the second detection result for the call recording obtained through the large language model may include the target score of the call recording and the behavior information of the target behavior existing in the call recording. That is to say, the large language model can be adjusted in advance, and the prompt words are used to constrain when the large language model is used for detection, so that the large language model can output the target score for the input dialogue text to characterize the probability of the existence of the target behavior and the behavior information of the target behavior existing. Among them, there are multiple target behaviors in the voice call that need to be detected, and the behavior information of the target behavior in the above second detection result can be at least one of the multiple target behaviors. The behavior information of the target behavior in the above second detection result can also be understood as a behavior type, that is, multiple target behaviors can be regarded as multiple behavior types, and the second detection result can include what behavior type the target behavior existing in the call recording belongs to.

[0070] In the embodiment of the present application, when the quality inspection result for the above call recording is determined based on the first test result and the second test result, a quality inspection report for the call recording can be generated based on the first test result and the second test result, and the generated quality inspection report can include whether the target keyword exists in the above conversation text, the location of the target keyword in the conversation text, the target score, and the behavior information of the target behavior. In this way, the quality inspection result finally determined can not only indicate whether the target behavior exists in the call recording, but also include detailed behavior information of the target behavior, and the target score represents the probability of the existence of the target behavior, so that accurate and detailed quality inspection results can be obtained.

[0071] In some embodiments, when the computer device generates a quality inspection report for the call recording based on the first test result and the second test result, it can also compare the above target score with the score threshold, which is used as a basis for judging whether the target behavior exists in the conversation recording. If the target score reaches the above score threshold (i.e., greater than or equal to the score threshold), it can be determined that the target behavior exists in the conversation recording; if the target score is less than the above score threshold, it can be determined that the target behavior does not exist in the conversation. The generated quality inspection report can include the result of whether the target behavior exists in the conversation recording determined based on the second test result.

[0072] In some embodiments, after obtaining the above quality inspection report, the computer device can also send the quality inspection report to the electronic device of the quality inspector; after receiving the quality inspection report, the electronic device of the quality inspector can display the quality inspection report on the screen. Since the quality inspection report can include whether the target keyword exists in the above conversation text, the position of the target keyword in the conversation text, the target score and the existing target behavior, it is convenient for the quality inspector to understand the quality inspection results of the above call recording in detail.

[0073] For example, in the scenario of quality inspection of call recordings of customer service, please refer to Figure 4 , Figure 4 The display interface A1 of the quality inspection report is shown. In the display interface A1, it can be displayed whether the target keyword is included, that is, Figure 4 The "pass" or "fail" in the column where the "sensitive word results" are located is shown in , "pass" means that the target keyword does not exist, and "fail" means that the target keyword exists; the display interface A1 may also include the location of the target keyword, for example Figure 4 In the column where “Sensitive words hit” is shown in , “In the 8th round of dialogue, sensitive words [village committee] appeared”. In addition, the display interface A1 can also display whether there is a target behavior in the dialogue recording determined by the target score obtained by the large voice model, for example Figure 4In the column of "AI Detection Result" shown in the figure, "Pass" means that the target behavior does not exist, and "Fail" means that the target behavior exists. In addition, the target score and the behavior information of the target behavior can also be displayed, for example Figure 4 As shown in the figure, "The customer service mentioned 'we will let the village committee contact him' which is a serious violation and the score is 100 points". Of course, the content that can be displayed in the display interface is not limited to this. For example, the number of historical violations of the customer service staff corresponding to the conversation recording can also be displayed.

[0074] In some embodiments, after obtaining the above first detection result and the second detection result, it can be determined whether the target keyword exists in the conversation text based on the first detection result, and it can be determined whether the target score is less than the target threshold based on the second detection result. If the target keyword exists in the conversation text and the target score is less than the target threshold, it means that the target behavior may exist in the call recording by matching the conversation text with the keyword, and the probability of determining the target behavior in the call recording by the large language model is low. In this case, a prompt message can be output to prompt the quality inspector to manually review the call recording, thereby further ensuring the accuracy of the quality inspection.

[0075] The information processing method provided in the embodiment of the present application is used to implement quality inspection of call recordings. The call recordings are not only tested by matching with target keywords, but also tested by a large language model obtained by adjusting sample data based on the field to which the call belongs, thereby improving the accuracy of the determined quality inspection results. In addition, by matching with the target keywords, it can be determined whether the target keywords exist in the conversation text and the position of the target keywords in the conversation text. The target score of the call recording and the behavior information of the target behavior in the call recording can be determined through the large language model, and then a detailed quality inspection report can be generated, which can better provide reference for relevant personnel.

[0076] See also Figure 5 , Figure 5 A flowchart of an information processing method provided by another embodiment of the present application is shown below. Figure 5 The process shown is described in detail, and the information processing method may specifically include the following steps:

[0077] Step S310: Obtain the conversation text corresponding to the call recording.

[0078] Step S320: Determine a first detection result for the call recording based on the matching result between the conversation text and the target keyword.

[0079] In the embodiment of the present application, steps S310 to S320 may refer to the contents of other embodiments and will not be described in detail here.

[0080] Step S330: If the first detection result indicates that the target keyword exists in the call recording, a second detection result for the call recording is obtained based on the conversation text and through a large language model, where the large language model is adjusted based on sample data of the field to which the call belongs.

[0081] In an embodiment of the present application, when combining a quality inspection method of matching a conversation text with a target keyword and a quality inspection method using a pre-trained large language model, a first detection result for the call recording can be determined based on the matching result between the conversation text and the target keyword. After obtaining the first detection result, if the first detection result indicates that the target keyword exists in the call recording, a second detection result for the call recording can be obtained based on the conversation text and using a large language model.

[0082] In addition, if the first detection result indicates that the target keyword does not exist in the call recording, then it can be directly determined that the target behavior does not exist in the above call recording. That is to say, after the quality inspection method of matching the conversation text with the target keyword is used to preliminarily determine that the target behavior may exist in the call recording, the call recording is obtained through the large language model for detection. It can be understood that when the caller is on the phone, if there is a target behavior, the target keyword corresponding to the target behavior will be spoken. Therefore, after preliminarily determining that the target behavior may exist in the call recording, the call recording is obtained through the large language model for detection. This can not only effectively improve the accuracy of the quality inspection of the call recording, but also reduce the invalid processing performed by the computer equipment.

[0083] Step S340: Based on the first detection result and the second detection result, obtain a quality inspection result for the call recording.

[0084] In the embodiment of the present application, step S340 can refer to the contents of the aforementioned embodiment and will not be repeated here.

[0085] The information processing method provided in the embodiment of the present application, when quality checking the call recording, not only detects the call recording by matching with the target keywords, but also detects the call recording by a large language model obtained by adjusting the sample data of the field to which the call belongs, thereby improving the accuracy of the determined quality inspection results; in addition, after preliminarily determining that the target behavior may exist in the call recording, the call recording is detected through the large language model, which can not only effectively improve the accuracy of the quality inspection of the call recording, but also reduce the invalid processing performed by the computer equipment.

[0086] See also Figure 6 , Figure 6 FIG. 1 is a flow chart showing an information processing method provided by another embodiment of the present application. Figure 6 The process shown is described in detail, and the information processing method may specifically include the following steps:

[0087] Step S410: Obtain the conversation text corresponding to the call recording.

[0088] In the embodiment of the present application, step S410 can refer to the contents of other embodiments and will not be repeated here.

[0089] Step S420: Match the conversation text with keywords in a keyword library.

[0090] In the embodiment of the present application, after the computer device obtains the conversation text corresponding to the above call recording, when determining the matching result between the conversation text and the target keyword, it can match the conversation text with the keywords in the keyword library. The keyword library includes a first keyword corresponding to the target behavior and a second keyword corresponding to the first keyword, and the second keyword is used as an exemption keyword for determining the absence of the target behavior when the first keyword exists in the conversation text.

[0091] Step S430: If the first keyword in the keyword library exists in the conversation text, the conversation text is matched with the second keyword corresponding to the first keyword in the keyword library.

[0092] Step S440: If the second keyword does not exist in the conversation text, it is determined that a target keyword exists in the conversation text, and the target keyword in the conversation text is the first keyword.

[0093] Step S450: If the second keyword exists in the conversation text, it is determined that the target keyword does not exist in the conversation text.

[0094] In an embodiment of the present application, after matching the conversation text with the keywords in the keyword library, if the conversation text contains the first keyword in the keyword library, the computer device can match the conversation text with the second keyword corresponding to the first keyword in the keyword library to determine whether the conversation text contains the corresponding exemption keyword; if the conversation text does not contain the second keyword in the keyword library, it means that the target behavior exists in the above call recording, so it can be determined that the matching result of the conversation text and the target keyword is that the target keyword exists in the conversation text, and the determined target keyword is the first keyword that can be matched, so that the detection result can be determined as the target behavior exists in the call recording based on the matching result of the conversation text and the target keyword; if the conversation text contains the second keyword in the keyword library, it means that the target behavior does not exist in the above call recording, so it can be determined that the matching result of the conversation text and the target keyword is that the target keyword does not exist in the conversation text. In addition, if the conversation text contains the first keyword that does not exist in the keyword library, it means that the target behavior does not exist in the above call recording, so it can be determined that the matching result of the conversation text and the target keyword is that the target keyword does not exist in the conversation text.

[0095] It is understandable that in some scenarios, when the keywords corresponding to the target behavior appear in the call, the keywords corresponding to the target behavior may be spoken in a specific scenario. For example, in the quality inspection scenario of the call recording of the customer service service, when the keywords corresponding to the target behavior appear in the customer service conversation, the keywords corresponding to the target behavior may be spoken in a specific scenario. For example, when the keyword "contact the village committee" appears, "verify the fund user" also appears in the sentence. In this case, the target behavior of the customer service staff can be exempted, that is, it can be determined that the target behavior does not exist in the call recording, so it can be determined that the matching result of the conversation text and the target keyword is that the target keyword does not exist in the conversation text. In this way, the accuracy of the detection result for the call recording can be improved by determining the matching result based on the conversation text and the target keyword.

[0096] In some implementations, for all first keywords corresponding to the target behavior in the keyword library, some of the first keywords in the keyword library may have corresponding second keywords, while some of the first keywords may not have corresponding second keywords. In the case where it is determined that the first keyword exists in the conversation text, the computer device may first determine whether the second keyword corresponding to the first keyword exists in the keyword library. If the second keyword corresponding to the first keyword exists in the keyword library, the conversation text may be further matched with the second keyword corresponding to the first keyword. If the second keyword corresponding to the first keyword does not exist in the keyword library, the matching result between the conversation text and the target keyword may be directly determined as the target keyword exists in the conversation text, and the determined target keyword is the matched first keyword.

[0097] Step S460: Determine a first detection result for the call recording based on the matching result between the conversation text and the target keyword.

[0098] Step S470: Based on the conversation text, a second detection result for the call recording is obtained through a large language model.

[0099] The large language model is obtained after being adjusted according to sample data of the field to which the call belongs.

[0100] Step S480: Based on the first detection result and the second detection result, obtain a quality inspection result for the call recording.

[0101] In the embodiment of the present application, steps S460 to S480 may refer to the contents of other embodiments and will not be described in detail here.

[0102] The information processing method provided in the embodiment of the present application, when quality checking the call recording, not only detects the call recording by matching with the target keyword, but also detects the call recording by using a large language model obtained by adjusting the sample data of the field to which the call belongs, thereby improving the accuracy of the determined quality inspection results; in addition, when determining the matching result between the conversation text and the target keyword, the keyword corresponding to the target behavior is matched with its corresponding exemption keyword. When the exemption keyword does not exist in the conversation text, it is finally determined that the target keyword exists in the conversation text, thereby improving the accuracy of the detection result for the call recording determined based on the matching result between the conversation text and the target keyword.

[0103] See also Figure 7 , Figure 7 FIG. 1 is a flow chart showing an information processing method provided by another embodiment of the present application. Figure 7The process shown is described in detail, and the information processing method may specifically include the following steps:

[0104] Step S510: Separate the call recording into a first recording and a second recording, wherein the first recording and the second recording correspond to different speakers.

[0105] In an embodiment of the present application, when the computer device determines the conversation text corresponding to a call recording, it can extract the recordings corresponding to different speakers, that is, separate the call recording into a first recording and a second recording, and the first recording and the second recording correspond to different speakers.

[0106] In some embodiments, the computer device can use the voiceprint features of each speaker to identify and segment different speakers in the call recording, and then separate it into the above first recording and second recording. Among them, features such as Mel frequency cepstral coefficients can be extracted from the recording, and then the audio can be divided into multiple speaker segments based on the extracted features, and the timestamps of each speaker segment can be recorded.

[0107] In some implementations, the computer device may also separate the call recording into segments of different speakers and timestamps of each segment using a pre-trained sound separation model, thereby obtaining the first recording and the second recording. The sound separation model may be a convolutional neural network, a recurrent neural network, a deep neural network, etc., and the specific model type may not be limited.

[0108] Step S520: Obtain a first text corresponding to the first recording and a first timestamp of the text in the first text.

[0109] In an embodiment of the present application, after obtaining the above-mentioned first recording and second recording, speech-to-text processing can be performed on the first recording, so as to obtain a first text corresponding to the first recording and a first timestamp of the text in the first text. The first timestamp can be understood as the timestamp corresponding to each sentence of the speaker, so that a conversion result with text and timestamp can be obtained.

[0110] In some embodiments, when performing speech-to-text processing on the first recording, punctuation processing may also be performed on the text, so that the obtained text content may include words and punctuation. For example, text conversion, punctuation of the text, and determination of the timestamp of the text may be achieved by a model based on a functional automatic speech recognition toolkit (FunASR).

[0111] Step S530: Obtain a second text corresponding to the second recording and a second timestamp of the text in the second text.

[0112] In an embodiment of the present application, the computer device can also perform speech-to-text processing on the second recording, thereby obtaining a second text corresponding to the second recording and a second timestamp of the text in the second text. The second timestamp can be understood as the timestamp corresponding to each sentence of the speaker, thereby obtaining a conversion result with text and timestamp.

[0113] Step S540: Generate a conversation text corresponding to the call recording based on the first text, the first timestamp, the second text, and the second timestamp.

[0114] In an embodiment of the present application, after obtaining the above-mentioned first text, the timestamp corresponding to the first text, the second text, and the timestamp corresponding to the second text, the conversation text corresponding to the call recording can be generated based on the first text, the first timestamp, the second text, and the second timestamp, thereby obtaining the conversation text of multiple rounds of conversations that distinguishes the speech content of different users and is sorted in chronological order from first to last.

[0115] In some implementations, the computer device may sort the text contents in the first text and the second text from earliest to latest according to the first timestamp and the second timestamp, thereby obtaining the above conversation text.

[0116] Step S550: Determine a first detection result for the call recording based on the matching result between the conversation text and the target keyword.

[0117] Step S560: Based on the conversation text, a second detection result for the call recording is obtained through a large language model, where the large language model is adjusted according to sample data of the field to which the call belongs.

[0118] Step S570: Based on the first detection result and the second detection result, obtain a quality inspection result for the call recording.

[0119] In the embodiment of the present application, steps S550 to S570 may refer to the contents of other embodiments and will not be described in detail here.

[0120] The information processing method provided in the embodiment of the present application, when quality checking call recordings, not only detects the call recordings by matching with target keywords, but also detects the call recordings through a large language model obtained by adjusting sample data based on the field to which the call belongs, thereby improving the accuracy of the determined quality inspection results; in addition, the embodiment of the present application also provides a method for determining the conversation text for the call recording, by separating the call recording into recordings of different speakers and then generating the conversation text, so that the content of different speakers can be distinguished in the conversation text, thereby improving the accuracy of subsequent detection of target behaviors using the conversation text.

[0121] See also Figure 8 , Figure 8 FIG. 1 is a flow chart showing an information processing method provided by yet another embodiment of the present application. Figure 8 The process shown is described in detail, and the information processing method may specifically include the following steps:

[0122] Step S610: Obtain sample data of the field to which the call belongs.

[0123] In the embodiment of the present application, a training process of a large language model for detecting target behaviors based on conversation texts for call recordings is also provided. When training a large language model for detecting target behaviors, sample data of the field to which the call to be detected belongs can be obtained, so that the large language model can be fine-tuned based on the obtained sample data. The above fields are not limited, for example, they can be the loan collection field of the financial industry, the after-sales service field of the home appliance industry, etc.

[0124] In some implementations, when obtaining the above sample data, the conversation text corresponding to the call recording generated in the above field can be annotated and processed, so as to construct sample data of multiple target behaviors in the field. When annotating and processing the conversation text, the annotation and processing can be performed according to the compliance requirements of the field.

[0125] For example, the sample data obtained in the field of debt collection can be: the question is "When the debt collector communicates with the borrower about the loan information, he tells the borrower that if he does not repay the loan, he will contact the village committee and neighbors to inform the borrower of the loan information." The answer to this question is "When the debt collector communicates with the borrower about the loan information, if he threatens to contact the village committee and neighbors to inform the borrower of the loan information, it involves violations of the treaty: privacy protection, personal information protection law, code of conduct for debt collection, harassment and intimidation, etc. Specifically, privacy protection: the borrower's loan information is personal privacy. If the debt collector discloses this information to others without consent, it may infringe the borrower's privacy rights. Personal information protection law: According to the Personal Information Protection Law, the collection and use of personal information requires explicit consent, and the behavior of the debt collector may violate this provision. Code of conduct for debt collection: Many countries and regions have laws and regulations on debt collection behavior, which stipulate the code of conduct that debt collectors should follow and prohibit the use of inappropriate means such as threats and harassment. Harassment and intimidation: If a debt collector uses threats (such as contacting neighbors) to force a borrower to repay the loan, it may be regarded as harassment or intimidation, which is generally not allowed by law."

[0126] The sample data obtained in the field of debt collection may be: the question is "When the debt collector is negotiating repayment with the borrower, he mentions to the borrower that if the loan is not repaid, the credit rating will be affected, which will affect the children's schooling and the civil service examination", and the answer is "When the debt collector is negotiating repayment with the borrower, if he mentions that non-repayment will affect the borrower's credit rating, and further claims that this will affect his children's schooling or the civil service examination, it involves violations of relevant laws and regulations: improper debt collection behavior, threats and intimidation. Specifically, improper debt collection behavior: The "Judicial Interpretation on Private Lending" clearly stipulates that debt collection activities shall be legal and in compliance with regulations, and threats, intimidation and other improper means shall not be used. Threats and intimidation: Linking non-repayment with important personal matters such as children's schooling and civil service examinations may be regarded as a psychological threat to the borrower, which is improper debt collection behavior."

[0127] Step S620: fine-tune the large language model according to the sample data and through the LoRA algorithm to obtain a trained large language model.

[0128] In the embodiments of the present application, taking into account the limited number of computing resources and training samples, a low-rank adaptation (LoRA) model algorithm can be used to fine-tune the large language model, that is, to fine-tune some layers and parameters of the large language model, thereby reducing the requirement for the amount of training data of the large language model used for detecting target behaviors in the above fields, thereby reducing the implementation cost and complexity.

[0129] In some implementations, since the amount of data is relatively small when fine-tuning a large language model, the learning rate used when training the large language model may be: 1e-4, i.e., 1*10 -4 = 0.0001. Among them, the learning rate is a key parameter that controls the learning speed of the model. It determines the step size of the model moving towards the direction of minimizing the loss function each time the parameters are updated.

[0130] In some embodiments, the detection results output by the large language model for detecting the target behavior may include behavior classification (i.e., the behavior information of the specific target behavior in the aforementioned embodiment). In such a case, a classification layer may be added to the large language model to identify the behavior classification. Moreover, by adding the above classification layer, it can help the large language model to learn and optimize more robustly during the fine-tuning stage, and can more effectively capture and integrate features related to content recognition, thereby improving the stability and accuracy of the model when processing detection tasks; the classification layer can also output a suitable format according to the requirements of the task and perform some appropriate regularization to reduce the complexity of the model and help prevent overfitting.

[0131] In some embodiments, when the large language model is fine-tuned by the LoRA algorithm, a low-rank adaptation (LoRA) model can be added to the large language model, and the LoRA model is trained with the above sample data, so that the large language model can better detect the target behavior in the above fields. Among them, the LoRA model is obtained by fine-tuning the cross-attention layer in the UNet module in the large language model; the LoRA model is applied to the cross-attention layer in the large language model, and the above sample data is used for training while freezing the parameters of the large language model, so that a LoRA model that can better detect the target behavior in the above fields in this scenario can be obtained. When used, the parameters of the LoRA model are injected into the large language model, so that the large language model outputs the detection result.

[0132] It is worth noting that the training of the above large language model can be carried out in advance based on sample data. Subsequently, each time the target behavior needs to be detected, the trained large language model can be used for detection, without the need to train the large language model every time a detection is required.

[0133] Step S630: Obtain the conversation text corresponding to the call recording.

[0134] Step S640: Determine a first detection result for the call recording based on the matching result between the conversation text and the target keyword.

[0135] Step S650: Based on the conversation text, a second detection result for the call recording is obtained through a large language model, where the large language model is adjusted according to sample data of the field.

[0136] Step S660: Based on the first detection result and the second detection result, obtain a quality inspection result for the call recording.

[0137] In the embodiment of the present application, steps S630 to S660 can refer to the contents of the aforementioned embodiment and will not be described in detail here.

[0138] The information processing method provided in the embodiment of the present application, when quality checking the call recording, not only detects the call recording by matching with the target keywords, but also detects the call recording by using a large language model obtained after adjustment based on sample data of the field to which the call belongs, thereby improving the accuracy of the determined quality inspection results; in addition, the embodiment of the present application also provides a training method for a large language model for detecting target behaviors in the field to which the call belongs, by using sample data of the field and fine-tuning the large language model using the LoRA algorithm, so as to obtain the above large language model, which can not only use the large language model to accurately detect the target behavior, but also reduce the requirement for the amount of training data of the large language model used for detecting target behaviors in the above fields, thereby reducing the implementation cost and complexity.

[0139] The following introduces the architecture involved in the information processing method provided in the above embodiment in conjunction with a scenario of quality inspection of call recordings for customer service.

[0140] See also Fig. 9, the architecture layers involved in the information processing method provided in the aforementioned embodiment include user layer, access layer, application layer, service layer, data layer and deployment. The user layer is the part directly facing the user. It provides an interface between the user and the system, so that the user can access and use various functions and services of the system through the interface. In the embodiment of the present application, the user layer is mainly for quality inspectors, customer service personnel and quality inspection policy administrators; the access layer can refer to the middle layer between the interface (user layer) for direct user interaction and the application layer that provides core functions or services. It serves as a bridge between the user layer and the application layer, and is responsible for processing user requests, forwarding them to the application layer, receiving responses from the application layer, and returning them to the user layer. The SSO AUTH in the access layer can refer to single sign-on (SSO, Single The application layer can be used to provide services such as recording management of customer service conversations, manual review, handling of violations, quality inspection reports, and customer service staff's speech library; the service layer is used to provide voice services, rule engines, large model bases, etc. Among them, voice services are mainly used for recording format conversion, automatic speech recognition services, sentiment analysis, text generation, etc.; the rule engine is mainly used for the management of illegal keywords, rule classification of illegal keyword matching, rule execution engine, rule compilation, etc.; the large model base is used to provide prompt word writing, large language model interface, agent (Agent) and retrieval-augmented generation (RAG), model library, etc.; the data layer mainly involves database management systems (such as MySQL), databases (such as Redis), data interaction (such as OCS), distributed messaging systems (such as Kafka), etc.; deployment mainly involves the deployment of architecture, for example, the architecture can be deployed in the cloud, etc.

[0141] The technical architecture involved in the information processing method provided in the above embodiment mainly includes the platform layer, the interaction layer and the algorithm layer. The platform layer includes the customer service system and the automatic voice bar system. The interaction layer includes the data collection layer, the voice processing layer, the reasoning and judgment layer and the quality inspection report generation layer. The algorithm layer includes the text matching algorithm and the self-built open source large language model. Please refer to Fig.10, the embodiment of the present application mainly realizes the quality inspection of the call recording of the customer service call through the voice processing layer and the reasoning and judgment layer, not only by matching with the illegal keywords to detect the call recording for violation, but also by fine-tuning the large language model obtained based on the sample violation data of the field to which the customer service call belongs to detect the call recording for violation. Among them, the voice processing layer can obtain the conversation text after processing the call recording through channel separation, sampling and format standardization, automatic speech recognition, text synthesis, dialogue text reading, etc.; and in the reasoning and judgment layer, through the text matching algorithm (i.e. matching with keywords) and through the self-built large language model to detect violations, it is possible to achieve accurate quality inspection of the call recording.

[0142] See also Fig.11 , which shows a structural block diagram of an information processing device 700 provided in an embodiment of the present application. The information processing device 700 applies the above-mentioned computer device, and the information processing device 700 includes: a text acquisition module 710, a first detection module 710, a second detection module 730 and a quality inspection result acquisition module 740. The text acquisition module 710 is used to obtain the conversation text corresponding to the call recording; the first detection module 710 is used to determine the first detection result for the call recording based on the matching result between the conversation text and the target keyword; the second detection module 730 is used to obtain the second detection result for the call recording based on the conversation text and through a large language model, and the large language model is obtained after adjustment based on the sample data of the field to which the call belongs; the quality inspection result determination module is used to obtain the quality inspection result for the call recording based on the first detection result and the second detection result.

[0143] In some embodiments, the first detection result includes whether the target keyword exists in the conversation text and the position of the target keyword in the conversation text; the second detection result includes the target score of the call recording and behavior information of the target behavior existing in the call recording, and the target score is used to characterize the probability of the existence of the target behavior.

[0144] In one possible implementation, the quality inspection result determination module is used to generate a quality inspection report for the call recording based on the first detection result and the second detection result, and the quality inspection report includes whether the target keyword exists in the conversation text, the position of the target keyword in the conversation text, the target score, and behavioral information of the existing target behavior.

[0145] In a possible implementation, the information processing device 700 may further include an information prompt module. The information prompt module is configured to output a prompt message after obtaining the second detection result for the call recording based on the conversation text and through the large language model, if the target keyword exists in the conversation text and the target score is less than the target threshold, and the prompt message is used to prompt manual review of the call recording.

[0146] In some implementations, the second detection module 730 may be specifically configured to: if the first detection result indicates that the target keyword exists in the call recording, obtain a second detection result for the call recording based on the conversation text and through a large language model.

[0147] In some embodiments, the information processing device 700 may also include a keyword matching module, which is used to match the conversation text with keywords in a keyword library before determining the first detection result for the call recording based on the matching result of the conversation text and the target keyword; if the conversation text contains a first keyword in the keyword library, the conversation text is matched with a second keyword corresponding to the first keyword in the keyword library; if the conversation text does not contain the second keyword, it is determined that the conversation text contains a target keyword, and the target keyword in the conversation text is the first keyword; if the conversation text contains the second keyword, it is determined that the conversation text does not contain the target keyword.

[0148] In some embodiments, the text acquisition module 710 can be specifically used to: separate the call recording into a first recording and a second recording, the first recording and the second recording corresponding to different speakers; obtain a first text corresponding to the first recording and a first timestamp of the text in the first text; obtain a second text corresponding to the second recording and a second timestamp of the text in the second text; and generate a conversation text corresponding to the call recording based on the first text, the first timestamp, the second text, and the second timestamp.

[0149] In some implementations, the information processing device 700 may further include a model training module. The model training module may be used to obtain sample data of the field to which the call belongs, and fine-tune the large language model according to the sample data and through the LoRA algorithm to obtain a trained large language model.

[0150] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and modules can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here.

[0151] In several embodiments provided in the present application, the coupling between modules may be electrical, mechanical or other forms of coupling.

[0152] In addition, each functional module in each embodiment of the present application can be integrated into a processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or software functional modules.

[0153] In summary, the solution provided by the present application obtains the conversation text corresponding to the call recording, determines the first detection result for the call recording based on the matching result between the conversation text and the target keyword, obtains the second detection result for the call recording based on the conversation text and through the large language model, and the large language model is obtained after adjustment based on the sample data of the field to which the call belongs, and then obtains the quality inspection result for the call recording based on the first detection result and the second detection result. As a result, when quality inspection of the call recording is performed, the call recording is not only detected by matching with the target keyword, but also detected by the large language model obtained after adjustment based on the sample data of the field to which the call belongs, thereby improving the accuracy of the determined quality inspection result.

[0154] Please refer to Fig.12 , which shows a structural block diagram of a computer device provided in an embodiment of the present application. The computer device 100 may be a server. The computer device 100 in the present application may include one or more of the following components: a processor 110, a memory 120, and one or more application programs, wherein the one or more application programs may be stored in the memory 120 and configured to be executed by one or more processors 110, and the one or more application programs are configured to execute the method described in the aforementioned method embodiment.

[0155] The processor 110 may include one or more processing cores. The processor 110 uses various interfaces and lines to connect various parts of the entire computer device 100, and executes various functions and processes data of the computer device 100 by running or executing instructions, programs, code sets or instruction sets stored in the memory 120, and calling data stored in the memory 120. Optionally, the processor 110 can be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), and programmable logic array (PLA). The processor 110 can integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. Among them, the CPU mainly processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing display content; and the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 110, but may be implemented separately through a communication chip.

[0156] The memory 120 may include a random access memory (RAM) or a read-only memory (ROM). The memory 120 may be used to store instructions, programs, codes, code sets or instruction sets. The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the following various method embodiments, etc. The data storage area may also store data (such as a phone book, audio and video data, chat record data) created by the computer device 100 during use.

[0157] Please refer to Fig.13 , which shows a structural block diagram of a computer-readable storage medium provided in an embodiment of the present application. The computer-readable medium 800 stores program codes, which can be called by a processor to execute the method described in the above method embodiment.

[0158] The computer readable storage medium 800 may be an electronic memory such as a flash memory, an EEPROM (electrically erasable programmable read-only memory), an EPROM, a hard disk, or a ROM. Optionally, the computer readable storage medium 800 includes a non-transitory computer-readable storage medium. The computer readable storage medium 800 has storage space for program code 810 that performs any method step in the above method. These program codes can be read from or written to one or more computer program products. The program code 810 can be compressed, for example, in an appropriate form.

[0159] An embodiment of the present application also provides a computer program product, which includes a computer program, and the computer program is stored in a storage medium; a processor of a computer device reads the computer program from the storage medium, and the processor executes the computer program, so that the computer device executes the information processing method provided in the above embodiment.

[0160] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An information processing method, characterized in that: The method comprises: Get the conversation text corresponding to the call recording; Determining a first detection result for the call recording based on a matching result between the conversation text and the target keyword; Based on the conversation text, a second detection result for the call recording is obtained by using a large language model, where the large language model is adjusted according to sample data of the field to which the call belongs; Based on the first detection result and the second detection result, a quality inspection result for the call recording is obtained.

2. The method according to claim 1, characterized in that The first detection result includes whether the target keyword exists in the conversation text and the position of the target keyword in the conversation text; the second detection result includes the target score of the call recording and the behavior information of the target behavior existing in the call recording, and the target score is used to characterize the probability of the existence of the target behavior.

3. The method according to claim 2, characterized in that The obtaining, based on the first detection result and the second detection result, a quality inspection result for the call recording includes: Based on the first detection result and the second detection result, a quality inspection report of the call recording is generated, the quality inspection report including whether the target keyword exists in the conversation text, the position of the target keyword in the conversation text, the target score, and behavior information of the existing target behavior.

4. The method according to claim 2, characterized in that: After obtaining a second detection result for the call recording based on the conversation text and through a large language model, the method further includes: If the target keyword exists in the conversation text and the target score is less than the target threshold, a prompt message is output, where the prompt message is used to prompt a manual review of the call recording.

5. The method according to claim 1, characterized in that The obtaining a second detection result for the call recording based on the conversation text and using a large language model includes: If the first detection result indicates that the target keyword exists in the call recording, a second detection result for the call recording is obtained based on the conversation text and through a large language model.

6. The method according to claim 1, characterized in that Before determining the first detection result for the call recording based on the matching result between the conversation text and the target keyword, the method further includes: Matching the conversation text with keywords in a keyword library; If the first keyword in the keyword library exists in the conversation text, matching the conversation text with a second keyword corresponding to the first keyword in the keyword library; If the second keyword does not exist in the conversation text, determining that a target keyword exists in the conversation text, and the target keyword in the conversation text is the first keyword; If the second keyword exists in the conversation text, it is determined that the target keyword does not exist in the conversation text.

7. The method according to any one of claims 1 to 6, characterized in that: The obtaining of the conversation text corresponding to the call recording includes: Separating the call recording into a first recording and a second recording, wherein the first recording and the second recording correspond to different speakers; Obtaining a first text corresponding to the first recording and a first timestamp of a word in the first text; Obtaining a second text corresponding to the second recording and a second timestamp of the text in the second text; Based on the first text, the first timestamp, the second text, and the second timestamp, a conversation text corresponding to the call recording is generated.

8. The method according to any one of claims 1 to 6, characterized in that: The large language model is trained in the following way: Obtain sample data of the field to which the call belongs; According to the sample data, the large language model is fine-tuned through the LoRA algorithm to obtain a trained large language model.

9. A computer device, characterized in that: include: one or more processors; Memory; One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method according to any one of claims 1 to 8.

10. A computer program product, characterized in that The computer program product comprises a computer program, wherein the computer program is stored in a storage medium; The processor of the computer device reads the computer program from the storage medium, and the processor executes the computer program, so that the computer device executes the method according to any one of claims 1 to 8.

Images