Information processing method, apparatus, device, storage medium, and program product

Abstract

The application discloses an information processing method, device and equipment, a storage medium and a program product. The method comprises the following steps: obtaining customer service text of a customer service role in historical recording data; determining a target deep semantic vector corresponding to the customer service text by using a pre-trained language model according to the customer service text; searching for target customer service text corresponding to the target deep semantic vector from the customer service text according to the target deep semantic vector, and constructing a customer service dialogue library by using the target customer service text and the target deep semantic vector; adding a new customer service text obtained to the customer service dialogue library, so as to update the customer service dialogue library; and the customer service dialogue library is used for supporting role matching and providing a basis for voice quality inspection. The application can simply and conveniently identify a role, and the accuracy of the identification is relatively high, thereby providing a relatively accurate basis for voice quality inspection, and automatically updating the customer service dialogue library, so that the workload of manually updating rules by business personnel is greatly reduced.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and in particular to an information processing method, apparatus, device, storage medium, and program product. Background Technology

[0002] Quality inspection is an indispensable part of risk control, widely used in scenarios such as bank electronic verification and debt collection. Intelligent quality inspection through recorded conversations can effectively alleviate the problems of low efficiency and limited coverage associated with manual sampling and recording of quality inspection recordings. In most cases, quality inspection requires role information to specify the objects to be inspected. Therefore, role recognition is a crucial component of intelligent quality inspection systems.

[0003] Currently, existing character recognition methods involve splitting a mono recording into two channels, then using a designed rule model and a binary classifier to classify the text in each channel, thereby achieving text matching.

[0004] However, in essence, this method only stays at the superficial level of text rule matching. It requires the formulation of many rules to cover as many types of text as possible, which is cumbersome and the accuracy of role recognition is low. Summary of the Invention

[0005] The main objective of this invention is to provide an information processing method, apparatus, device, storage medium, and program product that can easily identify roles with high accuracy, thereby providing a more accurate basis for voice quality inspection.

[0006] To achieve the above objectives, the present invention provides an information processing method, the method comprising:

[0007] Retrieve the customer service text belonging to the customer service role from historical recording data;

[0008] Based on the customer service text, the target deep semantic vector corresponding to the customer service text is determined through a pre-trained language model;

[0009] Based on the target depth semantic vector, the target customer service text corresponding to the target depth semantic vector is found from the customer service text, and the target customer service text and the target depth semantic vector are used to form a customer service script library;

[0010] The newly acquired customer service text is added to the customer service script library to update the customer service script library;

[0011] The customer service script library is used to support role matching and to provide a basis for voice quality inspection.

[0012] Optionally, determining the target deep semantic vector corresponding to the customer service text using a pre-trained language model based on the customer service text includes:

[0013] The customer service text is input into the pre-trained language model to generate the first deep semantic vector for each sentence in the customer service text;

[0014] Clustering is performed on each of the first deep semantic vectors to determine the target deep semantic vector.

[0015] Optionally, clustering each of the first deep semantic vectors to determine the target deep semantic vector includes:

[0016] Cluster each of the first deep semantic vectors to obtain multiple clusters, each cluster containing at least one first deep semantic vector;

[0017] Based on the number of first-depth semantic vectors in each cluster, the multiple clusters are filtered to obtain at least one target cluster;

[0018] The first depth semantic vector in the at least one target cluster is used as the target depth semantic vector.

[0019] Optionally, the newly acquired customer service text is added to the customer service script library to update the customer service script library, including:

[0020] Obtain each channel to be identified from the audio recording data to be identified, and each channel to be identified includes at least one text to be identified;

[0021] Using the pre-trained language model, the role corresponding to each of the channels to be identified is determined based on each text to be identified in each channel and each target deep semantic vector in the customer service script library. The role includes at least a customer service role and a customer role.

[0022] The text in the channel with the role of customer service is added to the customer service script library, and then the customer service script library is updated by clustering and filtering.

[0023] In this context, the text in the channel for the role of customer service is the new customer service text.

[0024] Optionally, determining the role corresponding to each of the channels to be identified using the pre-trained language model, based on each text to be identified in each channel and each target deep semantic vector in the customer service script database, includes:

[0025] For each channel to be identified, the following steps are performed: Based on each text to be identified, the pre-trained language model is used to determine the deep semantic vector to be identified corresponding to each text to be identified; for each deep semantic vector to be identified, the semantic matching score between the deep semantic vector to be identified and each target deep semantic vector in the customer service script library is calculated; based on the semantic matching score corresponding to each deep semantic vector to be identified, the role matching score corresponding to the channel to be identified is determined.

[0026] The role corresponding to each of the channels to be identified is determined based on the role matching score.

[0027] Optionally, calculating the semantic matching score between the deep semantic vector to be identified and each target deep semantic vector in the customer service script database includes:

[0028] Calculate the cosine similarity between the deep semantic vector to be identified and each target deep semantic vector in the customer service script database;

[0029] Obtain a first preset number of target cosine similarities from each cosine similarity, and calculate the average value of the first preset number of target cosine similarities;

[0030] The average of the preset number of target cosine similarities is the semantic matching score.

[0031] Optionally, determining the role matching score corresponding to the channel to be identified based on the semantic matching score corresponding to each of the depth semantic vectors to be identified includes:

[0032] Based on the semantic matching score corresponding to each of the depth semantic vectors to be identified, a second preset number of target semantic matching scores are obtained, and the average value of the second preset number of target semantic matching scores is calculated; the average value of the second preset number of target semantic matching scores is the role matching score corresponding to the channel to be identified.

[0033] Accordingly, determining the role corresponding to each of the channels to be identified based on the role matching score includes:

[0034] The role matching scores corresponding to each of the channels to be identified are compared, and the role with the highest role matching score is selected as the customer service role.

[0035] The present invention also provides an information processing apparatus, the apparatus comprising:

[0036] The acquisition module is used to retrieve the customer service text belonging to the customer service role in the historical recording data;

[0037] The first processing module is used to determine the target deep semantic vector corresponding to the customer service text by using a pre-trained language model.

[0038] The second processing module is used to find the target customer service text corresponding to the target deep semantic vector from the customer service text based on the target deep semantic vector, and to construct a customer service script library from the target customer service text and the target deep semantic vector.

[0039] The third processing module is used to add the acquired new customer service text to the customer service script library in order to update the customer service script library;

[0040] The customer service script library is used to support role matching and to provide a basis for voice quality inspection.

[0041] The present invention also provides an information processing device, the information processing device comprising: a memory, a processor, and an information processing program stored in the memory and executable on the processor, wherein the information processing program, when executed by the processor, implements the steps of the information processing method as described in any of the preceding claims.

[0042] The present invention also provides a computer-readable storage medium storing an information processing program, which, when executed by a processor, implements the steps of the information processing method as described in any of the preceding claims.

[0043] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the method described in any of the preceding claims.

[0044] In this invention, a customer service script library belonging to specific customer service roles is constructed to achieve role matching, thereby providing a basis for voice quality inspection. Specifically, firstly, customer service text belonging to specific customer service roles is obtained from historical recording data. Then, using the semantic analysis capabilities of a pre-trained language model, a deep representation of the customer service text, namely a deep semantic vector, is generated. Based on the deep semantic vector, through filtering and other operations, the retained deep semantic vector, namely the target deep semantic vector, is determined. This target deep semantic vector and the corresponding customer service text form the customer service script library. During the use of the customer service script library, new customer service texts can be added to the library to update it. Therefore, by using pre-trained language models for semantic analysis, a customer service script library containing effective texts carrying customer service roles can be generated. Semantic matching using this script library eliminates the need for continuous manual annotation, automatically annotating similar customer service scripts to achieve role matching. Furthermore, semantic matching differs from shallow text rule matching, eliminating the need to formulate a variety of rules to cover all possible statements. The process is simple and the role matching accuracy is high. At the same time, it enables automatic updates to the customer service script library, greatly reducing the workload of business personnel manually updating rules. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of an application scenario provided by an embodiment of the present invention;

[0046] Figure 2 A flowchart illustrating an information processing method provided in an embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram illustrating an application scenario provided by another embodiment of the present invention;

[0048] Figure 4 A flowchart illustrating an information processing method according to another embodiment of the present invention;

[0049] Figure 5 This is a schematic diagram of the structure of an information processing device provided in an embodiment of the present invention;

[0050] Figure 6 This is a schematic diagram of the structure of an information processing device provided in an embodiment of the present invention.

[0051] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0052] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the invention to those skilled in the art.

[0053] Currently, existing character recognition methods split mono recordings into two channels, then use a designed rule model and classifier to classify the text in each channel, thereby achieving text matching. However, since the input to the character recognition module is the result of a speech recognition system, it contains speech recognition errors, and some sentences may even become difficult to read. In this case, the accuracy of rule-based or classifier-based character recognition is low, forcing different business units to invest significant manpower and time in refining the model rules. Therefore, this method essentially only reaches a superficial level of text rule matching, requiring the development of numerous rules to cover various text types as much as possible, resulting in cumbersome steps and low accuracy in character recognition.

[0054] To address this issue, the technical concept of this invention is to utilize a pre-trained language model for semantic analysis, thereby generating a customer service script library containing valid texts representing customer service roles. Semantic matching is then performed using this script library, eliminating the need for continuous manual annotation and automatically annotating similar customer service scripts to achieve role matching. Furthermore, semantic matching differs from shallow text rule matching, eliminating the need to develop numerous rules to cover all possible statements. The process is simple and boasts high accuracy in role matching. Simultaneously, adding new customer service texts to the script library enables automatic updates, significantly reducing the workload of manual rule updates by business personnel.

[0055] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Where there is no conflict between the embodiments, the following embodiments and features thereof can be combined with each other.

[0056] It should be noted that the term "multiple" in this application is used to mean two or more, and the specific number is not specifically limited here.

[0057] Figure 1 This is a schematic diagram illustrating an application scenario provided by an embodiment of the present invention. For example... Figure 1 As shown, the executing entity of this information processing method can be an information processing device, which can be a terminal or a server, without specific limitations. Among them, voice quality inspection is achieved through role recognition, which can be applied to scenarios such as bank electronic verification and debt collection.

[0058] For example, taking a server as an example, in the application scenario of bank electronic verification, electronic devices with recording capabilities (such as communication devices or smart terminals with voice interaction capabilities) can record the conversation between customer service representatives and customers. This recorded conversation data (such as recordings of bank statement verification) can be stored in local files on the electronic device or uploaded to a cloud server. To accurately assess the voice quality during the verification process, the server can first distinguish between the customer role and the customer service role in the recorded conversation, preventing the quality assessment from being negatively influenced by the customer's wording and leading to false positives.

[0059] Specifically, role matching can be achieved by constructing a customer service script database, and then performing similarity matching between the acquired real-time recording data or the recording data to be detected and the content of the customer service script database to identify roles. The construction and updating process of the customer service script database can be as follows: First, collect all customer service text from manually annotated (related to telephone core) recording data, denoted as set M; then, the server obtains customer service text from historical telephone core recordings through electronic devices, uses a pre-trained language model to generate a deep representation (deep semantic vector) of each customer service text, and performs clustering, filtering out clusters with a small number of sentences (only retaining general customer service scripts); then, based on this information processing equipment, the implementation is run to obtain the role recognition results for the latest time period, and all customer service texts corresponding to the role recognition results are added to M, and the above steps are repeated for iterative optimization.

[0060] Based on the pre-built customer service script database, the server acquires real-time or target audio recordings via electronic devices. These recordings are then divided into two channels: one voiceprint per channel, designated as the customer service channel and the other as the target customer channel. A pre-trained language model generates deep semantic vectors for the text in each channel. These deep semantic vectors are then compared with the target deep semantic vectors in the customer service script database to determine the role for each channel. The channel with the highest similarity score is designated as the customer service channel, i.e., the customer service role. The text in the channel designated as the customer service role is added to the customer service script database. Through clustering and filtering, the database is automatically updated, significantly reducing the workload of manual rule updates by business personnel.

[0061] Therefore, this invention employs a semantic matching mechanism for role recognition. Since customer service scripts, questions, and language styles are similar in most scenarios, clustering methods can be used to automatically label similar customer service scripts, building a large-scale customer service script database. Then, the text from each recording channel is semantically matched against the text in the database. The channel with the highest matching score is the customer service channel, and the other channel is the customer channel. By using the customer service script database for semantic matching, continuous manual labeling is unnecessary; similar customer service scripts can be automatically labeled, thus achieving role matching. Furthermore, semantic matching differs from shallow text rule matching, eliminating the need to formulate numerous rules to cover all possible statements. The process is simple and the role matching accuracy is high. Simultaneously, it enables automatic updates to the customer service script database, significantly reducing the workload of manual rule updates by business personnel.

[0062] Figure 2 This is a flowchart illustrating an information processing method provided in an embodiment of the present invention. Figure 2 As shown, the method in this embodiment may include:

[0063] S201. Obtain the customer service text of the customer service role in the historical recording data.

[0064] Recording data can be obtained either directly from the local storage of an electronic device with recording capabilities, or from cloud storage via that device. Based on the obtained historical recording data, a speech recognition system parses it to obtain text. The text is then manually labeled with the corresponding roles, and customer service text for the customer service role is selected from the text.

[0065] S202. Based on the customer service text, determine the target deep semantic vector corresponding to the customer service text through a pre-trained language model.

[0066] Specifically, based on the pre-annotated customer service text, a pre-trained language model is used to generate deep semantic vectors corresponding to the customer service text, thereby achieving semantic differentiation of roles, rather than simply relying on shallow text rule matching. The pre-trained language model can map the customer service text to semantic vectors through matrix mapping relationships. Then, the deep semantic vectors corresponding to each (or sentence) text are pre-processed to filter out information irrelevant to professional or business needs (such as casual conversation unrelated to the business scenario), resulting in the target deep semantic vector corresponding to the customer service text. One sentence of text corresponds to one deep semantic vector.

[0067] S203. Based on the target depth semantic vector, find the target customer service text corresponding to the target depth semantic vector in the customer service text, and construct a customer service script library from the target customer service text and the target depth semantic vector.

[0068] The customer service script library is used to support role matching and to provide a basis for voice quality inspection.

[0069] Specifically, based on the target deep semantic vector, the text corresponding to the target deep semantic vector is found from the customer service text (which can be denoted as the target customer service text). Then, the target customer service text and the target deep semantic vector are combined into a set M, which serves as the customer service script library, to achieve role recognition or matching, thereby achieving high-level and high-accuracy voice quality inspection.

[0070] S204. Add the acquired new customer service text to the customer service script library to update the customer service script library. In this embodiment, through automatically iteratively updated customer service scripts, the latest business scripts can be automatically updated to the recognition system (such as information processing equipment), greatly reducing the workload of business personnel manually updating rules.

[0071] By acquiring customer service text belonging to specific customer service roles from historical recording data, and then utilizing the semantic analysis capabilities of a pre-trained language model, a deep semantic vector (DMS) is generated from the customer service text. Based on this DMS, filtering and other operations are performed to determine the retained DMS, which is the target DMS. This target DMS and its corresponding customer service text form a customer service script library. During the use of this script library, new customer service text can be added to update it. Therefore, by using a pre-trained language model for semantic analysis, a customer service script library containing valid text carrying customer service roles is generated. Semantic matching using this script library eliminates the need for continuous manual annotation, automatically annotating similar customer service scripts to achieve role matching. Furthermore, semantic matching differs from shallow text rule matching, eliminating the need to formulate numerous rules to cover all possible statements. The process is simple and boasts high accuracy in role matching.

[0072] Optionally, the target deep semantic vector corresponding to the customer service text can be determined by using a pre-trained language model, which can be achieved through the following steps:

[0073] Step a1: Input the customer service text into the pre-trained language model to generate the first deep semantic vector for each sentence in the customer service text.

[0074] Step a2: Cluster each of the first deep semantic vectors to determine the target deep semantic vector.

[0075] Specifically, since the acquired historical recording data consists of multiple entries, and each entry can include at least one or more sentences (each sentence can be considered as or correspond to a text), the acquired customer service texts are at least one or more. Taking multiple customer service texts as an example, the text is first input into a pre-trained language model. Through semantic analysis of the customer service texts, a first deep semantic vector is obtained for each text or sentence. To ensure the accuracy of the constructed customer service script library, cluster analysis can be performed on all generated first deep semantic vectors. This allows for the automatic labeling of similar customer service scripts using clustering methods, which is then used to semantically match the text of the recording channel to be detected with the text in the customer service script library, thereby achieving role differentiation.

[0076] Optionally, clustering is performed on each of the first deep semantic vectors to determine the target deep semantic vector. The method can also be implemented through the following steps:

[0077] Step b1: Cluster each of the first deep semantic vectors to obtain multiple clusters, each cluster containing at least one first deep semantic vector.

[0078] Step b2: Filter the multiple clusters according to the number of first-depth semantic vectors in each cluster to obtain at least one target cluster.

[0079] Step b3: Use the first depth semantic vector in the at least one target cluster as the target depth semantic vector.

[0080] Specifically, firstly, all first-depth semantic vectors are clustered, with similar semantic vectors forming a cluster. Therefore, clustering all first-depth semantic vectors can form multiple clusters. To ensure that the text and corresponding semantic vectors in the customer service script library are closely related to professional or business needs and are closer to real text, multiple clusters can be filtered and deleted: clusters with a small number of sentences are filtered out (only general customer service scripts are retained), thus retaining at least one target cluster. The first-depth semantic vectors in this target cluster are the target depth semantic vectors. Then, the target depth semantic vectors and the corresponding target customer service texts are combined to form the customer service script library, meaning the customer service script library includes the target depth semantic vectors and the corresponding target customer service texts.

[0081] For example, a large-scale customer service script library can be built in three steps: First, collect all customer service texts from the recording data of manually annotated roles, denoted as set M; Second, use a pre-trained language model to generate a deep representation (i.e., the first deep semantic vector) of each customer service text, and perform clustering to filter out clusters with a small number of sentences (only retaining general customer service scripts); Third, run the program on the information processing device for a period of time (e.g., one month) to obtain the latest role recognition results, add all customer service texts to M, and re-execute the first step of iterative optimization, continuously clustering and filtering to update the customer service script library, thereby achieving automatic iterative expansion of the customer service script library.

[0082] Optionally, this embodiment, based on the above embodiments, provides a detailed explanation of how to use the customer service script library to achieve role recognition and automatic updating of the customer service script library. The step of adding the acquired new customer service text to the customer service script library to update it can be achieved through the following steps:

[0083] Step c1: Using the pre-trained language model, determine the role corresponding to each of the channels to be identified based on each text to be identified in each channel and each target deep semantic vector in the customer service script library. The role includes at least a customer service role and a customer role.

[0084] Step c2: Add the Chinese text of the channel with the role of customer service to the customer service script library, and update the customer service script library by clustering and filtering.

[0085] In this context, the text in the channel for the role of customer service is the new customer service text.

[0086] Specifically, in combination Figure 3 As shown, Figure 3 This is a schematic diagram of an application scenario provided by another embodiment of the present invention. First, the recording data to be identified (such as new recording data generated in the past month) is obtained through an electronic device, such as a smartphone with recording function. The acquisition method can be similar to the method of acquiring historical recording data described above, and will not be repeated here. Then, the recording data is divided into two recording channels, namely the channels to be identified, through voiceprint separation technology. In order to distinguish the roles of the channels to be identified, a pre-trained language model can be used. The text of each channel to be identified is input into the pre-trained language model to generate a deep semantic vector of each sentence of each channel to be identified. Then, each sentence (i.e., each sentence text) of each channel to be identified is semantically matched with each target customer service text in the customer service script library.

[0087] The specific process of semantic matching involves semantically matching the deep semantic vector of each sentence in each channel to be identified with the target deep semantic vector of each target customer service text in the customer service script database, thereby determining whether the role corresponding to each channel to be identified is a customer service role or a customer role. Simultaneously, the latest role identification results are fed back to the customer service script database, and then, through the aforementioned semantic analysis, clustering, and filtering steps, the customer service script database is automatically iterated and expanded, greatly reducing the workload of business personnel manually updating rules.

[0088] Optionally, this embodiment provides a detailed explanation of how to achieve role recognition based on the above embodiments. Using the pre-trained language model, the role corresponding to each of the channels to be recognized is determined based on the text to be recognized in each channel and the target deep semantic vector in the customer service script database. This can be achieved through the following steps:

[0089] Step d1: For each channel to be identified, perform the following steps:

[0090] Step d11: Based on each of the texts to be identified, determine the deep semantic vector corresponding to each of the texts to be identified using the pre-trained language model;

[0091] Step d12: For each of the depth semantic vectors to be identified, calculate the semantic matching score between the depth semantic vector to be identified and each target depth semantic vector in the customer service script library;

[0092] Step d13: Determine the role matching score corresponding to the channel to be identified based on the semantic matching score corresponding to each of the depth semantic vectors to be identified;

[0093] Step d2: Determine the role corresponding to each of the channels to be identified based on the role matching score.

[0094] Specifically, each text to be identified for each channel is input into a pre-trained language model to generate a deep semantic vector for each text. Then, the similarity of each deep semantic vector is calculated with each target deep semantic vector in the customer service script database to obtain a semantic matching score. Next, for each channel, based on the semantic matching scores of all its deep semantic vectors, a role matching score is calculated for that channel, such as by averaging or averaging the top-ranked scores. The channel with the highest score is then identified as the customer service channel, and its role is designated as the customer service role.

[0095] Therefore, by calculating semantic matching score and role matching score, the role can be directly determined, which is a simple process and less prone to errors.

[0096] Optionally, a similarity score is used to calculate the semantic matching score, and the average value is taken to ensure the accuracy of the calculation. Specifically, the semantic matching score between the deep semantic vector to be identified and each target deep semantic vector in the customer service script database can be calculated through the following steps:

[0097] Step e1: Calculate the cosine similarity between the deep semantic vector to be identified and each target deep semantic vector in the customer service script database.

[0098] Step e2: Obtain a first preset number of target cosine similarities from each cosine similarity, and calculate the average value of the first preset number of target cosine similarities.

[0099] The average of the preset number of target cosine similarities is the semantic matching score.

[0100] Optionally, determining the role matching score corresponding to the channel to be identified based on the semantic matching score corresponding to each of the depth semantic vectors to be identified can be achieved through the following steps:

[0101] Step e3: Based on the semantic matching score corresponding to each of the depth semantic vectors to be identified, obtain a second preset number of target semantic matching scores, and calculate the average value of the second preset number of target semantic matching scores;

[0102] Step e4: The average of the second preset number of target semantic matching scores is the role matching score corresponding to the channel to be identified;

[0103] Optionally, determining the role corresponding to each of the channels to be identified based on the role matching score can be achieved through the following steps:

[0104] Step e5: Compare the role matching scores corresponding to each of the channels to be identified, and select the role with the highest role matching score as the customer service role.

[0105] Specifically, the semantic matching scores corresponding to a preset number of deep semantic vectors to be identified are taken, and the average score is calculated. Then, the average scores for each channel are compared, and the one with the higher score is identified as a customer service role. Since the customer service script database belongs to customer service roles, a higher score indicates a closer match to the database content, thus identifying the role. Role recognition is achieved through simple calculations.

[0106] For example, given a customer service script library M, channels P and Q to be identified, P = {P1, P2, ..., PQ} n} and Q = {Q1, Q2, ..., Q} mThe text consists of n and m sentences respectively. The goal is to identify which role P and Q belong to by analyzing the sentences in P and Q. The first step is to use a pre-trained language model to generate a deep semantic vector for each sentence in P and Q, and then calculate the cosine similarity with the deep semantic vector of each sentence in the customer service script database M. Taking channel P as an example, let s be the cosine similarity. <P i , j >=cosine(P i M j The second step is to work on P. i The first preset number of target semantic matching scores (e.g., for P) i The average of the top 10 (ranked by score) matching scores is used as P. i Similarly, the semantic matching score for each sentence in the Q channel can be obtained. The third step involves averaging the semantic matching scores of the second preset number of target sentences in the P channel (e.g., the top 10 matching scores in the P channel) to obtain the final score for the P channel. The final score for the Q channel can be obtained similarly. The fourth step compares the scores in the P and Q channels; sentences with higher matching scores are identified as customer service representatives, and those with lower scores are identified as customers. Based on this role recognition result, all customer service texts are added to M, and the iterative optimization of the customer service script database is re-executed.

[0107] Combination Figure 4 As shown, the labeled customer service scripts are input into a pre-trained language model to generate deep language vectors for the customer service text. Then, by clustering all deep speech vectors and filtering small clusters, a customer service script library (or customer service script semantic vector library) is generated. For the obtained text to be classified (i.e. the text to be recognized in the recognition channel), it is matched with the deep semantic vector of each sentence in the customer service script library, and the matching score is calculated to determine which channel belongs to the customer service channel. The new script is then periodically updated to the previously generated customer service script library (e.g., the original customer service corpus).

[0108] This invention introduces a pre-trained language model into role recognition, significantly improving the algorithm's semantic modeling capabilities. Simultaneously, the adoption of a semantic matching framework enhances the model's fault tolerance. Through automatically iteratively updated customer service scripts, the latest business scripts can be automatically updated into the recognition system (such as information processing equipment), greatly reducing the workload of business personnel manually updating rules.

[0109] Figure 5 This is a schematic diagram of the structure of an information processing device provided in an embodiment of the present invention. Figure 5 As shown, the information processing device may include:

[0110] The acquisition module 501 is used to acquire the customer service text belonging to the customer service role in the historical recording data;

[0111] The first processing module 502 is used to determine the target deep semantic vector corresponding to the customer service text by using a pre-trained language model.

[0112] The second processing module 503 is used to find the target customer service text corresponding to the target deep semantic vector from the customer service text according to the target deep semantic vector, and to form a customer service script library by combining the target customer service text and the target deep semantic vector.

[0113] The third processing module 504 is used to add the acquired new customer service text to the customer service script library in order to update the customer service script library.

[0114] The customer service script library is used to support role matching and to provide a basis for voice quality inspection.

[0115] Optionally, the first processing module includes a first processing unit and a second processing unit; the first processing unit is used to input the customer service text into the pre-trained language model to generate a first deep semantic vector for each sentence in the customer service text; the second processing unit is used to cluster each of the first deep semantic vectors to determine the target deep semantic vector.

[0116] Optionally, the second processing unit is specifically used for:

[0117] Cluster each of the first deep semantic vectors to obtain multiple clusters, each cluster containing at least one first deep semantic vector;

[0118] Based on the number of first-depth semantic vectors in each cluster, the multiple clusters are filtered to obtain at least one target cluster;

[0119] The first depth semantic vector in the at least one target cluster is used as the target depth semantic vector.

[0120] Optionally, the acquisition module is also used to: acquire each channel to be identified in the recording data to be identified, each channel to be identified including at least one text to be identified;

[0121] The fourth processing module includes the third processing unit and the fourth processing unit.

[0122] The third processing unit is used to determine the role corresponding to each of the channels to be identified by using the pre-trained language model, based on each of the texts to be identified in each channel to be identified and each target deep semantic vector in the customer service script library. The role includes at least a customer service role and a customer role.

[0123] The fourth processing unit is used to add the channel text with the role of customer service to the customer service script library, and update the customer service script library through clustering and filtering; wherein, the channel text with the role of customer service is the new customer service text.

[0124] Optionally, the third processing unit is specifically used for:

[0125] For each channel to be identified, the following steps are performed: Based on each text to be identified, the pre-trained language model is used to determine the deep semantic vector to be identified corresponding to each text to be identified; for each deep semantic vector to be identified, the semantic matching score between the deep semantic vector to be identified and each target deep semantic vector in the customer service script library is calculated; based on the semantic matching score corresponding to each deep semantic vector to be identified, the role matching score corresponding to the channel to be identified is determined.

[0126] The role corresponding to each of the channels to be identified is determined based on the role matching score.

[0127] Optionally, the third processing unit is specifically used for:

[0128] Calculate the cosine similarity between the deep semantic vector to be identified and each target deep semantic vector in the customer service script database;

[0129] Obtain a first preset number of target cosine similarities from each cosine similarity, and calculate the average value of the first preset number of target cosine similarities;

[0130] The average of the preset number of target cosine similarities is the semantic matching score.

[0131] Optionally, the third processing unit is specifically used for:

[0132] Based on the semantic matching score corresponding to each of the depth semantic vectors to be identified, a second preset number of target semantic matching scores are obtained, and the average value of the second preset number of target semantic matching scores is calculated.

[0133] The average of the second preset number of target semantic matching scores is the role matching score corresponding to the channel to be identified;

[0134] Correspondingly, the third processing unit is also specifically used for:

[0135] The role matching scores corresponding to each of the channels to be identified are compared, and the role with the highest role matching score is selected as the customer service role.

[0136] The information processing apparatus provided in any of the foregoing embodiments is used to execute the technical solutions of any of the foregoing method embodiments. Its implementation principle and technical effect are similar, and will not be repeated here.

[0137] Figure 6 This is a schematic diagram of the structure of an information processing device provided in an embodiment of the present invention. Figure 6 As shown, the device may include: a memory 601, a processor 602, and an information processing program stored in the memory 601 and executable on the processor 602. When the information processing program is executed by the processor 602, it implements the steps of the information processing method as described in any of the foregoing embodiments.

[0138] Alternatively, the memory 601 can be either standalone or integrated with the processor 602.

[0139] The implementation principle and technical effects of the device provided in this embodiment can be found in the foregoing embodiments, and will not be repeated here.

[0140] This invention also provides a computer-readable storage medium storing an information processing program, which, when executed by a processor, implements the steps of the information processing method as described in any of the foregoing embodiments.

[0141] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the methods described in any of the foregoing embodiments.

[0142] In the several embodiments provided by this invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.

[0143] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of the present invention.

[0144] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.

[0145] The memory may include high-speed RAM, and may also include non-volatile storage (NVM), such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk or optical disc, etc.

[0146] The aforementioned storage medium can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium accessible to general-purpose or special-purpose computers.

[0147] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor. Both the processor and the storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and storage medium can exist as discrete components in an electronic device or host device.

[0148] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0149] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0150] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0151] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. An information processing method, characterized in that, The method includes: Retrieve the customer service text belonging to the customer service role from historical recording data; Based on the customer service text, the target deep semantic vector corresponding to the customer service text is determined through a pre-trained language model; Based on the target depth semantic vector, the target customer service text corresponding to the target depth semantic vector is found from the customer service text, and the target customer service text and the target depth semantic vector are used to form a customer service script library; The newly acquired customer service text is added to the customer service script library to update the customer service script library; The customer service script library is used to support role matching and to provide a basis for voice quality inspection. The step of determining the target deep semantic vector corresponding to the customer service text using a pre-trained language model includes: The customer service text is input into the pre-trained language model to generate the first deep semantic vector for each sentence in the customer service text; Cluster each of the first deep semantic vectors to obtain multiple clusters, each cluster containing at least one first deep semantic vector; Based on the number of first-depth semantic vectors in each cluster, the multiple clusters are filtered to obtain at least one target cluster. The filtering is used to filter out clusters with a small number of sentences and retain only the general customer service scripts. The first depth semantic vector in the at least one target cluster is used as the target depth semantic vector.

2. The method according to claim 1, characterized in that, The step of adding the acquired new customer service text to the customer service script library to update the customer service script library includes: Obtain each channel to be identified from the audio recording data to be identified, and each channel to be identified includes at least one text to be identified; Using the pre-trained language model, the role corresponding to each of the channels to be identified is determined based on each text to be identified in each channel and each target deep semantic vector in the customer service script library. The role includes at least a customer service role and a customer role. The text in the channel with the role of customer service is added to the customer service script library, and then the customer service script library is updated by clustering and filtering. In this context, the text in the channel for the role of customer service is the new customer service text.

3. The method according to claim 2, characterized in that, The step of determining the role corresponding to each of the channels to be identified using the pre-trained language model, based on each text to be identified in each channel and each target deep semantic vector in the customer service script database, includes: For each channel to be identified, the following steps are performed: Based on each text to be identified, the pre-trained language model is used to determine the deep semantic vector to be identified corresponding to each text to be identified; for each deep semantic vector to be identified, the semantic matching score between the deep semantic vector to be identified and each target deep semantic vector in the customer service script library is calculated; based on the semantic matching score corresponding to each deep semantic vector to be identified, the role matching score corresponding to the channel to be identified is determined. The role corresponding to each of the channels to be identified is determined based on the role matching score.

4. The method according to claim 3, characterized in that, The calculation of the semantic matching score between the deep semantic vector to be identified and each target deep semantic vector in the customer service script database includes: Calculate the cosine similarity between the deep semantic vector to be identified and each target deep semantic vector in the customer service script database; Obtain a first preset number of target cosine similarities from each cosine similarity, and calculate the average value of the first preset number of target cosine similarities; The average of the preset number of target cosine similarities is the semantic matching score.

5. The method according to claim 3, characterized in that, The step of determining the role matching score corresponding to the channel to be identified based on the semantic matching score corresponding to each of the depth semantic vectors to be identified includes: Based on the semantic matching score corresponding to each of the depth semantic vectors to be identified, a second preset number of target semantic matching scores are obtained, and the average value of the second preset number of target semantic matching scores is calculated. The average of the second preset number of target semantic matching scores is the role matching score corresponding to the channel to be identified; Accordingly, determining the role corresponding to each of the channels to be identified based on the role matching score includes: The role matching scores corresponding to each of the channels to be identified are compared, and the role with the highest role matching score is selected as the customer service role.

6. An information processing device, characterized in that, The device includes: The acquisition module is used to retrieve the customer service text belonging to the customer service role in the historical recording data; The first processing module is used to determine the target deep semantic vector corresponding to the customer service text by using a pre-trained language model. The second processing module is used to find the target customer service text corresponding to the target deep semantic vector from the customer service text based on the target deep semantic vector, and to construct a customer service script library from the target customer service text and the target deep semantic vector. The third processing module is used to add the acquired new customer service text to the customer service script library in order to update the customer service script library; The customer service script library is used to support role matching and to provide a basis for voice quality inspection. The first processing module includes a first processing unit and a second processing unit; The first processing unit is used to input the customer service text into the pre-trained language model and generate a first deep semantic vector for each sentence in the customer service text; The second processing unit is used to cluster each of the first deep semantic vectors to obtain multiple clusters, each cluster containing at least one first deep semantic vector. Based on the number of first-depth semantic vectors in each cluster, the multiple clusters are filtered to obtain at least one target cluster. The filtering is used to filter out clusters with a small number of sentences and retain only the general customer service scripts. The first depth semantic vector in the at least one target cluster is used as the target depth semantic vector.

7. An information processing device, characterized in that, The information processing device includes: a memory, a processor, and an information processing program stored in the memory and executable on the processor, wherein when the information processing program is executed by the processor, it implements the steps of the information processing method as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an information processing program, which, when executed by a processor, implements the steps of the information processing method as described in any one of claims 1-5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-5.

Images