Intelligent recommendation method and apparatus, device, medium

Abstract

The embodiment provides a kind of intelligent recommendation method and device, equipment, medium, belong to artificial intelligence technical field.The method comprises: obtaining user voice;The speech recognition processing is carried out to user voice, to obtain target speech text;At least one intent set is obtained by the NLP intent recognition model to the intent recognition processing of target speech text;According to intent set, obtain user intent list, and user intent list is used to provide at least one user intent to user;Receive the intent information that user feedbacks according to user intent;According to intent information, determine target recommendation rhetoric from pre-set rhetoric recommendation library.The scheme is cleaned by pre-set NLP intent recognition model to speech text, extracts intent set, and then obtains user intent list.To clarify intent to customer according to user intent list, to determine target recommendation rhetoric to carry out service sales, improve the conversion rate in service process.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an intelligent recommendation method, apparatus, device, and medium. Background Technology

[0002] Insurance companies typically provide customers with channels for conducting business, including online apps, telephone customer service, and offline branches. When conducting business through telephone customer service, the customer experience largely stems from interactions at after-sales service touchpoints. High-quality after-sales service can effectively improve customer satisfaction, thus turning the end of after-sales service into a new sales starting point. Currently, industry-standard telephone customer service assistants can retrieve relevant policy information or historical service records by inputting partial customer information, but they cannot proactively explore whether the customer's current service request reveals other business opportunities. Furthermore, they cannot provide effective information to potential customers (i.e., unfamiliar customers without policies), forcing customer service personnel to rely on generic marketing strategies to recommend products, resulting in low sales conversion rates. Summary of the Invention

[0003] The main objective of this disclosure is to provide an intelligent recommendation method, apparatus, device, and medium that can more effectively uncover customers' target needs, accurately provide high-quality services, and improve sales conversion rates during the service process.

[0004] To achieve the above objectives, a first aspect of this disclosure provides an intelligent recommendation method, comprising:

[0005] Get user voice;

[0006] The user's voice is subjected to speech recognition processing to obtain the target speech text;

[0007] The target speech text is input into a preset NLP intent recognition model, and the NLP intent recognition model performs intent recognition processing on the target speech text to obtain at least one intent set. Each intent set corresponds to a user intent, wherein the part of speech of the intent set includes at least subject, predicate, and object.

[0008] A user intent list is obtained based on the intent set, and the user intent list is used to provide the user with at least one of the user intents;

[0009] Receive intent information from the user based on the user's intent;

[0010] The target recommended script is determined from the preset script recommendation library based on the intent information.

[0011] According to the first aspect of the present disclosure, an intelligent recommendation method has at least the following beneficial effects: an intent set is extracted based on user voice, and an intent list is obtained based on the intent set, so that customer service personnel can obtain the intent information fed back by the user based on the intent list, and then output recommended products / services from a preset script recommendation library based on the intent information, which can fully explore the potential needs of customers and improve the sales conversion rate.

[0012] In some embodiments, performing speech recognition processing on the user's speech to obtain target speech text includes:

[0013] Background noise in the user's speech is removed according to the preset LMS algorithm to obtain a preliminary speech.

[0014] The initial speech is input into a preset ASR model, and the initial speech is processed by the ASR model to obtain the target speech text.

[0015] In some embodiments, the process of performing intent recognition processing on the target speech text using the NLP intent recognition model to obtain at least one intent set includes:

[0016] The target speech text is segmented into words to obtain the segmentation results;

[0017] The word segmentation results are then subjected to part-of-speech tagging to obtain the part-of-speech tagging results;

[0018] At least one intent set is extracted from the part-of-speech tagging results according to preset grammar rules.

[0019] In some embodiments, the step of performing word segmentation on the target speech text to obtain the segmentation result includes:

[0020] The target speech text is segmented according to a preset positive maximum matching algorithm to obtain a preliminary word segmentation list;

[0021] Each word in the initial word segmentation list is processed with a vector identifier to obtain a word segmentation vector set;

[0022] Each word in the word segmentation vector set is matched with a reference word in a preset dictionary, and the Euclidean distance between the word segment and the reference word is calculated. The matching result of the word segment is determined based on the Euclidean distance value.

[0023] If the Euclidean distance value is within the preset distance threshold, the matching result is successful, and the word segmentation is added to the word segmentation result.

[0024] In some embodiments, if a set of intents exists, obtaining a list of user intents based on the set of intents includes:

[0025] Multiple candidate intents are obtained from a preset intent library based on the intent set;

[0026] The matching score of the candidate intent is determined based on the degree of matching between the candidate intent and the intent set;

[0027] The candidate intents are sorted in reverse order based on the matching scores, and a user intent list is obtained based on the top N candidate intents after the reverse sorting, where N is a positive integer;

[0028] or,

[0029] A list of user intents is obtained based on the candidate intents whose matching scores are higher than a preset matching score threshold.

[0030] In some embodiments, if multiple intent sets exist, obtaining a list of user intents based on the intent sets includes:

[0031] Determine the frequency and position of the intent set within the word segmentation results, and obtain an intent score for the intent set based on the frequency and position;

[0032] The intent sets are sorted in reverse order based on the intent score, and a user intent list is obtained based on the first N intent sets after the reverse order, where N is a positive integer;

[0033] or,

[0034] A user intent list is obtained based on the set of intents whose intent scores are higher than a preset intent score threshold.

[0035] In some embodiments, the preset dictionary is obtained through the following steps:

[0036] Obtain historical user questions and extract historical intent sets from the historical user questions according to preset grammatical rules. Each historical intent set corresponds to a historical user intent. The parts of speech of the historical intent set include at least subject, predicate, and object.

[0037] A preset dictionary is obtained based on the set of historical user intents.

[0038] A second aspect of this disclosure provides an intelligent recommendation device, comprising:

[0039] The voice acquisition module is used to acquire the user's voice.

[0040] A speech recognition module is used to perform speech recognition processing on the user's speech to obtain target speech text;

[0041] An intent recognition module is used to input the target speech text into a preset NLP intent recognition model, and to perform intent recognition processing on the target speech text through the NLP intent recognition model to obtain at least one intent set, wherein one intent set corresponds to one user intent, and the part of speech of the intent set includes at least subject, predicate, and object.

[0042] The user intent list determination module is used to obtain a user intent list based on the intent set, wherein the user intent list is used to provide the user with at least one of the user intents;

[0043] The intent information determination module is used to receive intent information fed back by the user based on the user's intent;

[0044] The script recommendation module is used to determine the target recommended script from a preset script recommendation library based on the intent information.

[0045] A third aspect of this disclosure provides a computer device including a memory and a processor, wherein the memory stores a program, and when the program is executed by the processor, the processor is used to perform the method described in any one of the embodiments of the first aspect of this application.

[0046] A fourth aspect of this disclosure provides a storage medium that is a computer-readable storage medium storing computer-executable instructions for causing a computer to perform the method described in any one of the embodiments of the first aspect of this application.

[0047] The intelligent recommendation method, apparatus, device, and medium proposed in this disclosure involve: acquiring user voice; performing speech recognition processing on the user voice to obtain target speech text; inputting the target speech text into a preset NLP intent recognition model, and performing intent recognition processing on the target speech text through the NLP intent recognition model to obtain at least one intent set, where each intent set corresponds to a user intent, and the part-of-speech of the intent set includes at least a subject, predicate, and object; obtaining a user intent list based on the intent set, which is used to provide the user with at least one of the user intents; receiving intent information from the user based on the user intents; and determining target recommended dialogue from a preset dialogue recommendation library based on the intent information. This disclosure can extract intent sets based on user voice and obtain intent lists based on the intent sets, so as to obtain user-feedback intent information based on the intent lists, and then output recommended products / services based on the intent information, which can fully explore potential customer needs and improve sales conversion rates. Attached Figure Description

[0048] Figure 1This is a flowchart of the intelligent recommendation method provided in the embodiments of this disclosure;

[0049] Figure 2 yes Figure 1 The flowchart of step S130 in the process;

[0050] Figure 3 yes Figure 2 The flowchart of step S210 in the middle;

[0051] Figure 4 yes Figure 1 The first flowchart of step S140 in the process;

[0052] Figure 5 yes Figure 1 The second flowchart of step S140 in the process;

[0053] Figure 6 A block diagram of the module structure of the intelligent recommendation device provided in the embodiments of this disclosure;

[0054] Figure 7 This is a schematic diagram of the hardware structure of a computer device provided in an embodiment of this disclosure. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining this application and are not intended to limit this application.

[0056] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0058] First, let's analyze some of the terms used in this application:

[0059] Artificial intelligence (AI) is a new branch of computer science that studies, develops, and applies theories, methods, technologies, and systems to simulate, extend, and expand human intelligence. It aims to understand the essence of intelligence and produce intelligent machines that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thought. Furthermore, AI utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to achieve optimal results.

[0060] Natural Language Processing (NLP) is an important field within computer science and artificial intelligence. It studies the theories and methods for enabling effective communication between humans and computers using natural language. NLP is a science that integrates linguistics, computer science, and mathematics. Since the natural language involved in this field is the language people use in daily life, it is closely related to linguistic research. NLP technologies typically include text processing, semantic understanding, machine translation, chatbot question answering, and knowledge graphs. NLP aims to help computers understand sentences or words in human language. NLP aims to reduce user workload and fulfill the desire for human-computer interaction using natural language. Because users may be unfamiliar with machine language, NLP can help such users communicate with machines using natural language. NLP already has widespread applications in the financial sector, such as intelligent customer service—through NLP's intent recognition, question-answer similarity analysis, and semantic understanding, machines can replace human customer service representatives.

[0061] Automatic Speech Recognition (ASR): Also known as automated speech recognition, speech recognition is an interdisciplinary subfield of computational linguistics. Its development methods and technologies enable computers to recognize and translate spoken language. It integrates knowledge and research from linguistics, computer science, and electrical engineering.

[0062] The Least Mean Square (LMS) algorithm is an improvement on the steepest descent algorithm. This algorithm does not require known statistical characteristics of the input and desired signals; the weights at the current time step are obtained by adding a proportional term of the negative mean square error gradient to the weights at the previous time step. Its low computational complexity, good convergence in stationary signal environments, unbiased convergence of its expected value to the Wiener solution, and stationarity when implemented with finite precision make the LMS algorithm the most stable and widely used adaptive algorithm.

[0063] The forward maximum matching algorithm segments a given text from left to right, extracting the longest word at the current position. It's a dictionary-based word segmentation method whose principle is that the larger the granularity of a word, the more precise its meaning. The algorithm mainly involves: starting from the beginning of a string, selecting a segment with the longest possible word length; if the segment is shorter than the maximum word length, selecting the entire segment. First, it checks if the segment is in the dictionary; if so, it's counted as a segmented word. If not, it starts from the right, reducing the length by one character, and then checks if the shorter segment is in the dictionary, repeating this process until only one character remains.

[0064] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0065] The intelligent recommendation methods provided in this application can all be applied to artificial intelligence. Basic artificial intelligence technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. Artificial intelligence software technologies mainly include computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0066] With the development of the insurance business, insurance companies generally provide customers with channels for handling business, including online apps, telephone customer service, and offline branches (or agents). The elderly make up a significant portion of the insurance company's customer base. Based on their usage habits and convenience, telephone customer service accounts for a high percentage of transactions. Customer experience largely stems from interactions at after-sales service touchpoints. High-quality after-sales service can effectively improve customer satisfaction, turning the end of after-sales service into a new sales starting point. In an era where the insurance industry is increasingly focused on existing customers, acquiring new customers is becoming increasingly difficult. Effectively tapping into the consumption potential of existing customers and increasing their insurance spending is an inevitable choice for insurance companies. Currently, telephone customer service assistants can retrieve relevant policy information or historical service records by inputting partial customer information, but they cannot proactively explore whether the customer's current service request presents other business opportunities. Furthermore, they cannot provide effective information to potential customers (i.e., unfamiliar customers without policies), forcing customer service personnel to recommend products using generic marketing strategies, resulting in low sales conversion rates.

[0067] Based on this, the main objective of this disclosure is to propose an intelligent recommendation method, apparatus, device, and medium. This method involves acquiring user speech, performing speech recognition processing on the user speech to obtain target speech text, and then using an NLP intent recognition model to perform intent recognition processing on the target speech text to obtain at least one intent set. Each intent set corresponds to a user intent, and the part-of-speech tags of the intent set at least include a subject, predicate, and object. A user intent list is obtained based on the intent set, and this user intent list is used to provide the user with at least one user intent. The method also includes receiving intent information from the user based on their intents, and determining target recommended dialogue from a preset dialogue recommendation library based on the intent information. This disclosure extracts intent sets from user speech and obtains intent lists based on these intent sets, enabling the provision of user intents to the user based on the intent lists. Furthermore, it outputs recommended products / services based on the obtained user-feedback intent information, which can fully explore potential customer needs and improve sales conversion rates.

[0068] The intelligent recommendation method provided in this disclosure can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, or smartwatch, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the above method, but is not limited to the above forms.

[0069] This disclosure can be applied to numerous general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer computer devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. This application can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules can reside in local and remote computer storage media, including storage devices.

[0070] This disclosure provides intelligent recommendation methods, devices, and media, which are specifically illustrated through the following embodiments. First, the intelligent recommendation method in this disclosure is described.

[0071] Reference Figure 1 The intelligent recommendation method according to the embodiments of this disclosure includes, but is not limited to, steps S110 to S150.

[0072] S110, acquire user voice;

[0073] S120, performs speech recognition processing on the user's speech to obtain the target speech text;

[0074] S130, the target speech text is input into a preset NLP intent recognition model, and the NLP intent recognition model is used to perform intent recognition processing on the target speech text to obtain at least one intent set. One intent set corresponds to one user intent, wherein the part of speech of the intent set includes at least subject, predicate, and object.

[0075] S140, Obtain a user intent list based on the intent set. The user intent list is used to provide the user with at least one user intent.

[0076] S150, receive intent information from the user based on the user's intent;

[0077] S160, determine the target recommended script from the preset script recommendation library based on the intent information.

[0078] In step S110, the user's voice is acquired. The user's voice can refer to the voice generated in real time during the process of the user calling in to handle business, or it can be the voice input by the user through business systems, WeChat, or other channels.

[0079] In step S120, the user's voice is processed by speech recognition to obtain the target speech text. For example, the voice of a customer calling in via telephone will be fed into the ASR model in real time for recognition, and the recognized target speech text will be directly fed into the NLP intent recognition model for text processing.

[0080] In one embodiment, step S120 specifically includes the following steps:

[0081] Background noise in the user's speech is removed according to the preset LMS algorithm to obtain a preliminary speech.

[0082] The initial speech is input into a preset ASR model, which performs speech recognition processing on the initial speech to obtain the target speech text.

[0083] Specifically, considering that the accuracy requirements for recognition rate are not too stringent in telephone scenarios, the main focus is on quickly identifying the user's main intent. The user's voice undergoes initial processing, using the classic LMS algorithm to remove background noise, followed by ASR model processing to obtain the target speech text. This approach balances recognition efficiency and accuracy.

[0084] In step S130, the target speech text is processed by an NLP intent recognition model to obtain at least one intent set. Each intent set corresponds to a user intent, and the part-of-speech tag of the intent set includes at least a subject, predicate, and object. For example: I / check / insurance policy number, I / buy / product.

[0085] In step S140, a user intent list needs to be obtained based on the intent set. It should be noted that a single customer call may contain multiple business transactions or mention multiple intents, requiring the creation of a user intent list to provide the user with at least one user intent. Furthermore, since a single intent can correspond to multiple linguistic expressions, the object of the user's inquiry under that intent can be diverse. Therefore, a user intent list needs to be obtained from the extracted intent set to uncover potential customer intents. It should also be noted that the user intent appearing in the target speech text may not be accurate. For example, a user's original intent might be "I want to check my policy number," but their actual intent might be "I want to know how to change my policy number." Therefore, the user intent list obtained from the intent set may contain more than one intent set; it can be based on the original intent set to generate multiple candidate intent sets.

[0086] In steps S150 to S160, intent information fed back by the user is received, and a target recommended script is determined from a preset script recommendation library based on this intent information. The user intent list contains multiple intents, which customer service personnel use to confirm with the user, as the user's true intent may not be accurately expressed during the call. Obtaining the user's intent information after confirmation is similar to a search engine displaying a search page; the user's true intent may be the first intent listed on the user intent list, or it may not be. The purpose of this embodiment is to confirm with the user based on a user intent list composed of intent sets. During a phone service call, customer service personnel need to ask the user about the candidate intents on the user intent list to obtain the user's intent information, allowing customer service personnel to clearly understand the user's intent and recommend products / services accordingly.

[0087] In one embodiment, after receiving the user intent list from the model, customer service personnel will conduct a secondary confirmation with the customer regarding the desired service. Only after receiving this secondary confirmation can the customer be certain that the intent has been clarified. After clarifying the intent, a corresponding recommended script / response will be selected based on that intent. For example: how to recommend services to the customer, how to apply for a hospital expedited service, etc.

[0088] Through steps S110 to S160, after performing speech recognition on users calling customer service, and extracting user intent from the recognized text using an NLP intent recognition model, the intent of customer calls can be effectively identified. This embodiment of the disclosure can obtain a list of user intents based on the intent set, and then obtain the intent information fed back by the user, so as to recommend corresponding scripts based on the intent information. This embodiment of the disclosure provides customer service personnel with a massive script library, assisting them to output targeted recommended scripts to users more quickly and accurately based on user intent information. This embodiment of the disclosure can fully explore users' potential needs, meet users' personalized needs, provide high-quality services to customers, improve sales conversion rates during the service process, and enhance user experience.

[0089] In a specific example, consider this: A user receives the following voice text during an incoming call: "Please check my insurance policy number; I need to pay my bill soon." After filtering, segmenting, and tagging using an NLP intent recognition model, the extracted intent might be: [Intent missing - likely a typo]. Figure 1 I / check / policy number, meaning Figure 2 : I / Pay / (Missing). The first intent is matched against a pre-defined intent library, and a matching score is calculated. Based on the matching score, a user intent list is generated, which includes multiple candidate intents. Figure 1 For example: Candidate 1: How to find the policy number? Matching score for Candidate 1: 9.5 points; Candidate 2: How to find the policy customer number? Matching score for Candidate 2: 8 points; Candidate 3: Can the policy number be changed? Matching score for Candidate 3: 6.5 points. These candidate intents will be presented to customer service personnel in descending order of matching score. Customer service personnel will then conduct a secondary confirmation with the customer based on the user intent list. If the customer has a clear intent, say Candidate 1, the customer service personnel can click on the title of Candidate 1 to get the corresponding answer.

[0090] In some embodiments, such as Figure 2 As shown, step S130 specifically includes the following steps:

[0091] S210, Perform word segmentation on the target speech text to obtain the word segmentation results;

[0092] S220, perform part-of-speech tagging on the word segmentation results to obtain the part-of-speech tagging results;

[0093] S230, extract at least one intent set from the part-of-speech tagging results according to preset grammar rules.

[0094] In step S210, since customers often use a lot of interjections (such as "um" or "ah") or reverse word order during telephone communication, it is necessary to use the NLP intent recognition model to segment the target speech text recognized by the ASR model to obtain the segmentation results.

[0095] In step S220, after segmenting the target speech text, part-of-speech tagging is performed on each segment, and interjections irrelevant to the sentence are skipped. The subject, predicate, and object of the segmentation results are identified, and the segmentation results are filtered based on the identification results. The part-of-speech tagging results corresponding to the target speech text are obtained based on the filtering results. The subject, predicate, and object of the sentence are identified; segments not belonging to these three parts of speech are considered meaningless and filtered out. The remaining words are added to the part-of-speech tagging results. Intent recognition is performed based on the cleaned part-of-speech tagging results, reducing the difficulty of subsequent intent recognition.

[0096] In step S230, at least one intent set is extracted from the part-of-speech tagging results according to preset grammatical rules. For example, key elements such as "subject + predicate + object" can be extracted by referring to the rules of Chinese grammar, and the intent set "WHO (who) + DO (handle) + WHAT (business / product)" is output, as shown in Table 1. It should be noted that the key elements are used to indicate the semantics of the user's input speech expression, and the intent set to which the key elements belong is used to indicate the user's intent. The subject generally refers to a person or thing, including: personal name, place name, organization, group, other nouns, etc. The predicate generally refers to a simple predicate, mainly verbs here. The object generally refers to the object corresponding to the action, mainly the business or product to be handled here, and therefore is also mainly a noun.

[0097] Table 1

[0098] WHO (who) DO (Process) WHAT (Business / Product) I check Policy Number child Apply Hospital Green Channel spouse Processing Policy replacement

[0099] In some embodiments, such as Figure 3 As shown, step S210 specifically includes the following steps:

[0100] S310, the target speech text is segmented according to the preset positive maximum matching algorithm to obtain a preliminary word segmentation list;

[0101] S320, each word in the initial word segmentation list is processed with vector identification to obtain a word segmentation vector set;

[0102] S330: Match each word in the word segmentation vector set with the reference word in the preset dictionary, calculate the Euclidean distance between the word segment and the reference word, and determine the matching result of the word segmentation based on the Euclidean distance value.

[0103] S340, if the Euclidean distance value is within the preset distance threshold, the matching result is successful, and the word segmentation is added to the word segmentation result.

[0104] Specifically, in steps S310 to S340, considering the higher efficiency requirements of real-time telephone recognition scenarios, the forward maximum matching algorithm is used to segment the target speech text. The resulting preliminary word segmentation list needs to be vector-labeled to calculate the Euclidean distance between texts. Each word in the word segmentation vector set is matched with a reference word in a preset dictionary, and the Euclidean distance between the word segment and the reference word is calculated. The matching result of the word segmentation is determined based on the Euclidean distance value. If the Euclidean distance value is within a preset distance threshold, the matching result is successful, and the word segment is added to the word segmentation result. It should be noted that the Euclidean distance value is calculated based on the vector-labeled word segment and the vector-labeled reference word.

[0105] The preset dictionary in step S310 is obtained through the following steps, including:

[0106] Obtain historical user questions and extract historical intent sets from them according to preset grammatical rules. Each historical intent set corresponds to a historical user intent. The parts of speech of the historical intent set must include at least subject, predicate, and object.

[0107] A pre-defined dictionary is obtained based on a set of historical user intents.

[0108] Specifically, user questions from the dialogue recommendation library serve as dictionary input. The dictionary is manually annotated to obtain a user intent set (i.e., the model's dictionary). Manual annotation can be simply understood as tagging; the process involves marking out phrases and words that represent user intent, thus creating the dictionary. This allows the model to match user statements with the dictionary to identify user intent.

[0109] It should be noted that the preset dictionaries also include publicly available dictionaries such as Chinese dictionaries.

[0110] In some embodiments, if there exists a set of intents, such as Figure 4 As shown, step S140 specifically includes the following steps:

[0111] S410, obtain multiple candidate intents from the preset intent library based on the intent set;

[0112] S420, determine the matching score of the candidate intent based on the degree of matching between the candidate intent and the intent set;

[0113] S430: Sort multiple candidate intents in reverse order based on matching scores, and obtain a list of user intents based on the top N candidate intents after the reverse order, where N is a positive integer.

[0114] In steps S410 to S430, during a customer's phone call, a set of intents may be identified, or the customer may only be able to provide a complete intent within a preset time interval, or the customer may not know how to provide a specific intent. Therefore, it is necessary to search for candidate intents in a preset intent library based on the intent set, and then determine the matching score of the candidate intents based on their degree of matching with the intent set. Based on the matching score, it is determined which candidate intent can be added to the user intent list. For example, multiple candidate intents can be sorted in reverse order based on their matching scores, and the user intent list is obtained based on the top N candidate intents after the reverse order, where N is a positive integer. It is understandable that a candidate intent with a high matching score is not necessarily the user's true intent, just like a search engine displays a search page; the user's true intent may be the first candidate intent displayed, or it may not be. The purpose of this embodiment is to confirm with the user based on the user intent list composed of these candidate intents. During a phone service, the customer service representative needs to ask the user about the candidate intents on the user intent list, obtain the user's confirmation, and then make corresponding product / service recommendations based on the intent information provided by the user.

[0115] In some embodiments, if there is an intent set, multiple candidate intents are obtained from a preset intent library based on the intent set, the matching score of the candidate intents is determined based on the degree of matching between the candidate intents and the intent set, and a list of user intents is obtained based on the candidate intents whose matching scores are higher than a preset matching score threshold.

[0116] In some embodiments, if there are multiple sets of intents, such as Figure 5 As shown, step S140 specifically includes the following steps:

[0117] S510, determine the frequency and position of the intent set in the word segmentation results, and obtain the intent score of the intent set based on the frequency and position;

[0118] S520: Sort multiple intent sets in reverse order according to intent scores, and obtain a user intent list based on the first N intent sets after reverse sorting, where N is a positive integer;

[0119] In steps S510 to S520, multiple intent sets are identified in a single customer call, or multiple complete intent sets are provided by the customer within a preset time interval, indicating that the user's goal is relatively clear. When scoring the intent sets, the frequency and position of occurrence in the word segmentation results are used as factors for a comprehensive score. The scores are then updated and sorted in real-time from high to low, as shown in Table 2. Specifically, the frequency and position of the intent set within the word segmentation results are determined, and an intent score is obtained based on these factors. Then, multiple intent sets are sorted in reverse order based on their intent scores, and a user intent list is obtained from the top N intent sets in this reverse order. In one embodiment, the user intent list is obtained from intent sets whose intent scores are higher than a preset intent score threshold. It should be noted that the higher the frequency of occurrence in the word segmentation results, the higher the intent score. The later the occurrence in the word segmentation results, the higher the intent score.

[0120] Table 2

[0121] WHO (who) DO (Process) WHAT (Business / Product) Intent rating (highest to lowest) I check Policy Number 8.5 child Apply Hospital Green Channel 7.8 spouse Processing Policy replacement 6.5

[0122] In some embodiments, after a telephone customer service representative answers a customer's question, if the customer's satisfaction rating is higher than 8 points (out of 10), the relevant call recording will be automatically used as a positive sample input to the NLP intent recognition model. Conversely, ratings lower than 8 points will be used as negative samples. No ratings will not be included in the learning process. By using ratings after each service, a closed-loop input of positive and negative samples gradually improves the model's accuracy.

[0123] The intelligent recommendation method proposed in this disclosure involves: acquiring user voice; performing speech recognition processing on the user voice to obtain target speech text; inputting the target speech text into a preset NLP intent recognition model; performing intent recognition processing on the target speech text through the NLP intent recognition model to obtain at least one intent set, where each intent set corresponds to a user intent, and the part-of-speech tagging of the intent set includes at least a subject, predicate, and object; obtaining a user intent list based on the intent set, which is used to provide the user with at least one user intent; receiving intent information from the user based on the user intent; and determining target recommended dialogue from a preset dialogue recommendation library based on the intent information. This disclosure, by performing speech recognition on incoming customer service calls and then using an NLP intent recognition model to segment, tag, and extract user intent from the recognized text, can effectively identify the intent of incoming customers. This disclosure can sort the intent sets according to their scores and obtain a user intent list. This disclosure can clarify the intent to the user based on the user intent list to determine the user's intent information. This disclosure provides customer service personnel with a vast script library, assisting them in quickly and accurately generating targeted recommended scripts based on user feedback. This disclosure can fully uncover users' potential needs, meet their personalized requirements, provide high-quality service, improve sales conversion rates during the service process, and enhance user experience.

[0124] This disclosure also provides an intelligent recommendation device, such as... Figure 6 As shown, the above-mentioned intelligent recommendation method can be implemented using a device comprising: a voice acquisition module 610, a voice recognition module 620, an intent recognition module 630, a user intent list determination module 640, an intent information determination module 650, and a script recommendation module 660. Specifically, the voice acquisition module 610 acquires user voice; the voice recognition module 620 performs voice recognition processing on the user voice to obtain target voice text; the intent recognition module 630 inputs the target voice text into a preset NLP intent recognition model, performs intent recognition processing on the target voice text through the NLP intent recognition model to obtain at least one intent set, where each intent set corresponds to a user intent; the user intent list determination module 640 obtains a user intent list based on the intent set, which provides the user with at least one user intent; the intent information determination module 650 receives intent information fed back by the user based on the user intent; and the script recommendation module 660 determines target recommended scripts from a preset script recommendation library based on the intent information. The intelligent recommendation device of this disclosure is used to execute the intelligent recommendation method in the above embodiments. Its specific processing procedure is the same as that of the intelligent recommendation method in the above embodiments, and will not be described in detail here.

[0125] The intelligent recommendation device proposed in this embodiment implements the aforementioned intelligent recommendation method. After performing speech recognition on incoming customer service calls, it extracts user intent from the recognized text using an NLP intent recognition model. This effectively identifies the intent of incoming calls and generates a user intent list based on the intent set, thereby obtaining the user's feedback intent information for appropriate script recommendations. The device in this embodiment can also effectively identify the intent of incoming customer calls by performing speech recognition on the recognized text, performing word segmentation, part-of-speech tagging, and extracting user intent using an NLP intent recognition model. This embodiment can sort the intent sets according to their scores and obtain a user intent list. This embodiment can clarify the user's intent based on the user intent list to confirm the user's intent information. This embodiment provides customer service personnel with a massive script library, assisting them in quickly and accurately outputting targeted recommended scripts to users based on their feedback intent information. This embodiment can fully explore users' potential needs, meet their personalized requirements, provide high-quality services, improve sales conversion rates during the service process, and enhance user experience.

[0126] This disclosure also provides a computer device, including:

[0127] At least one processor, and,

[0128] A memory that is communicatively connected to at least one processor; wherein,

[0129] The memory stores instructions that are executed by at least one processor to cause the at least one processor to perform the method as described in any of the embodiments of this application when executing the instructions.

[0130] The computer device proposed in this disclosure, by implementing the aforementioned intelligent recommendation method, can effectively identify the intent of incoming customer calls by performing speech recognition on the customer's voice and then using an NLP intent recognition model to segment, tag, and extract user intent from the recognized text. This disclosure embodiment can sort the intent sets according to their scores to obtain a list of user intents. This disclosure embodiment can clarify the user's intent to the user based on the user intent list to confirm the user's intent information. This disclosure embodiment provides customer service personnel with a massive script library, assisting them in quickly and accurately outputting targeted recommended scripts to users based on the intent information provided by the user. This disclosure embodiment can fully explore the user's potential needs, meet the user's personalized requirements, provide high-quality service, improve sales conversion rates during the service process, and enhance the user experience.

[0131] The following is combined Figure 7The hardware structure of the computer device is described in detail. The computer device includes: a processor 701, a memory 702, an input / output interface 703, a communication interface 704, and a bus 705.

[0132] The processor 701 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this disclosure.

[0133] The memory 702 can be implemented in the form of ROM (Read Only Memory), static storage device, dynamic storage device, or RAM (Random Access Memory). The memory 702 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 702 and called and executed by the processor 701 using the intelligent recommendation method of the embodiments of this disclosure.

[0134] The input / output interface 703 is used to implement information input and output;

[0135] Communication interface 704 is used to enable communication and interaction between this device and other devices. Communication can be achieved via wired means (e.g., USB, Ethernet cable) or wireless means (e.g., mobile network, Wi-Fi, Bluetooth).

[0136] Bus 705 transmits information between various components of the device (e.g., processor 701, memory 702, input / output interface 703, and communication interface 704);

[0137] The processor 701, memory 702, input / output interface 703, and communication interface 704 are connected to each other within the device via bus 705.

[0138] This disclosure also provides a storage medium, which is a computer-readable storage medium storing computer-executable instructions for causing a computer to execute the intelligent recommendation method of this disclosure.

[0139] The storage medium of this embodiment cleanses and extracts intent sets from speech text using a preset NLP intent recognition model, thereby obtaining a user intent list. This allows for the determination of user-feedback intent information based on the user intent list, leading to the identification of target recommended sales scripts for service sales and improving sales conversion rates during the service process.

[0140] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0141] The embodiments described in this disclosure are for the purpose of more clearly illustrating the technical solutions of this disclosure and do not constitute a limitation on the technical solutions provided by this disclosure. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by this disclosure are also applicable to similar technical problems.

[0142] It will be understood by those skilled in the art that Figures 1 to 5 The technical solutions shown in the figures do not constitute a limitation on the embodiments of this disclosure. They may include more or fewer steps than those shown in the figures, or combine certain steps, or different steps.

[0143] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0144] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0145] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0146] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0147] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0148] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0149] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0150] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0151] The preferred embodiments of the present disclosure have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present disclosure. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and spirit of the present disclosure shall be within the scope of the claims of the present disclosure.

Claims

1. An intelligent recommendation method, characterized in that, include: Get user's phone voice recording; The user's telephone voice is processed by speech recognition to obtain the target voice text; The target speech text is input into a preset NLP intent recognition model. The NLP intent recognition model performs word segmentation on the target speech text to obtain word segmentation results. The word segmentation results are then tagged with parts of speech to obtain part-of-speech tagging results. At least one intent set is extracted from the part-of-speech tagging results according to preset grammar rules. Each intent set corresponds to a user intent. The intent set is used to indicate the semantics of the user's telephone voice expression. The parts of speech of the intent set include at least a subject, a predicate, and an object. The subject indicates a person or thing, the predicate indicates a verb, and the object indicates the service or product that the user needs to handle. If a set of intents exists, multiple candidate intents are obtained from a preset intent library based on the set of intents; a matching score for each candidate intent is determined based on the degree of matching between the candidate intents and the set of intents; the multiple candidate intents are sorted in reverse order based on the matching scores, and a user intent list is obtained based on the top N candidate intents after the reverse order; or, a user intent list is obtained based on the candidate intents whose matching scores are higher than a preset matching score threshold; where N is a positive integer. If multiple intent sets exist, determine the frequency and position of each intent set within the word segmentation result, and obtain an intent score for each intent set based on the frequency and position; sort the multiple intent sets in reverse order based on the intent scores, and obtain a user intent list based on the top N intent sets after the reverse order, where N is a positive integer; or, obtain a user intent list based on intent sets whose intent scores are higher than a preset intent score threshold, and use the user intent list to provide the user with at least one user intent; Receive intent information from the user based on the user's intent; Based on the intent information, a target recommended script is determined from a preset script recommendation library. The target recommended script is used to recommend products or services to users.

2. The method according to claim 1, characterized in that, The step of performing speech recognition processing on the user's telephone voice to obtain the target speech text includes: Background noise in the user's telephone voice is removed according to the preset LMS algorithm to obtain preliminary voice. The initial speech is input into a preset ASR model, and the initial speech is processed by the ASR model to obtain the target speech text.

3. The method according to claim 1 or 2, characterized in that, The step of segmenting the target speech text to obtain segmentation results includes: The target speech text is segmented according to a preset positive maximum matching algorithm to obtain a preliminary word segmentation list; Each word in the initial word segmentation list is processed with a vector identifier to obtain a word segmentation vector set; Each word in the word segmentation vector set is matched with a reference word in a preset dictionary, and the Euclidean distance between the word segment and the reference word is calculated. The matching result of the word segment is determined based on the Euclidean distance value. If the Euclidean distance value is within the preset distance threshold, the matching result is successful, and the word segmentation is added to the word segmentation result.

4. The method according to claim 3, characterized in that, The preset dictionary is obtained through the following steps: Obtain historical user questions and extract historical intent sets from the historical user questions according to preset grammatical rules. Each historical intent set corresponds to a historical user intent. The parts of speech of the historical intent set include at least subject, predicate, and object. A preset dictionary is obtained based on the set of historical user intents.

5. An intelligent recommendation device, characterized in that, include: The voice acquisition module is used to acquire user telephone voice messages; A speech recognition module is used to perform speech recognition processing on the user's telephone voice to obtain the target speech text; An intent recognition module is used to input the target speech text into a preset NLP intent recognition model, perform word segmentation on the target speech text through the NLP intent recognition model to obtain word segmentation results, perform part-of-speech tagging on the word segmentation results to obtain part-of-speech tagging results, and extract at least one intent set from the part-of-speech tagging results according to preset grammar rules. Each intent set corresponds to a user intent. The intent set is used to indicate the semantics of the user's telephone voice expression. The part of speech of the intent set includes at least a subject, a predicate, and an object. The subject indicates a person or thing, the predicate indicates a verb, and the object indicates the service or product that the user needs to handle. The user intent list determination module is used for: If a set of intents exists, multiple candidate intents are obtained from a preset intent library based on the set of intents; a matching score for each candidate intent is determined based on the degree of matching between the candidate intents and the set of intents; the multiple candidate intents are sorted in reverse order based on the matching scores, and a user intent list is obtained based on the top N candidate intents after the reverse order; or, a user intent list is obtained based on the candidate intents whose matching scores are higher than a preset matching score threshold; where N is a positive integer. If multiple intent sets exist, determine the frequency and position of each intent set within the word segmentation result, and obtain an intent score for each intent set based on the frequency and position; sort the multiple intent sets in reverse order based on the intent scores, and obtain a user intent list based on the top N intent sets after the reverse order, where N is a positive integer; or, obtain a user intent list based on intent sets whose intent scores are higher than a preset intent score threshold, and use the user intent list to provide the user with at least one user intent; The intent information determination module is used to receive intent information fed back by the user based on the user's intent; The script recommendation module is used to determine a target recommended script from a preset script recommendation library based on the intent information. The target recommended script is used to recommend products or services to users.

6. A computer device, characterized in that, The computer device includes a memory and a processor, wherein the memory stores a program, which the processor executes when the program is executed: The method as described in any one of claims 1 to 4.

7. A storage medium, said storage medium being a computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which are used to cause a computer to perform, for example: The method as described in any one of claims 1 to 4.

Images