WO2022095378A1 - 基于人工智能的培训方法、装置、计算机设备及存储介质 - Google Patents

基于人工智能的培训方法、装置、计算机设备及存储介质 Download PDF

Info

Publication number
WO2022095378A1
WO2022095378A1 PCT/CN2021/091287 CN2021091287W WO2022095378A1 WO 2022095378 A1 WO2022095378 A1 WO 2022095378A1 CN 2021091287 W CN2021091287 W CN 2021091287W WO 2022095378 A1 WO2022095378 A1 WO 2022095378A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
customer
question
video
training
Prior art date
Application number
PCT/CN2021/091287
Other languages
English (en)
French (fr)
Inventor
满园园
侯晓龙
陈闽
许闻笳
宋思宇
高毅
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2022095378A1 publication Critical patent/WO2022095378A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/3331—Query processing
    • G06F16/334—Query execution
    • G06F16/3344—Query execution using natural language analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35—Clustering; Classification
    • G06F16/353—Clustering; Classification into predefined classes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/284—Lexical analysis, e.g. tokenisation or collocates
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/20—Education
    • G06Q50/205—Education administration or guidance
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Definitions

  • the present application relates to the field of artificial intelligence technology, and belongs to the application scenario of intelligent online training for service personnel in smart cities, and particularly relates to an artificial intelligence-based training method, device and computer equipment.
  • the embodiments of the present application provide an artificial intelligence-based training method, device, computer equipment, and storage medium, which aim to solve the problem of low training efficiency in the prior art methods when training service personnel.
  • an artificial intelligence-based training method which includes:
  • a virtual question video corresponding to the question information in the target training information is generated according to the virtual video generation model and sent to the user terminal to obtain the answer video information fed back by the user terminal, wherein the answer video information includes at least one segment reply video;
  • Response scoring information corresponding to the answering video information is acquired according to the preset scoring model and the target training information, and sent to the user terminal.
  • an embodiment of the present application provides an artificial intelligence-based training device, which includes:
  • the customer classification information acquisition unit is configured to, if receiving the historical business handling records input by the administrator, classify and count the customers in the historical business handling records according to the preset customer information quantification model and the preset customer classification model to obtain customer classification information;
  • a classified sample data information acquisition unit configured to split the historical business handling records according to the preset business scenario process and the customer classification information to obtain classified sample data information matching the customer classification information;
  • a target training information obtaining unit configured to obtain target training information that matches the training request information in the classified sample data information if the training request information from the user terminal is received;
  • a virtual question video sending unit configured to generate a virtual question video corresponding to the question information in the target training information according to the virtual video generation model and send it to the user terminal, so as to obtain the reply video information fed back by the user terminal, wherein,
  • the reply video information includes at least one reply video
  • a reply scoring information sending unit configured to acquire reply scoring information corresponding to the reply video information according to a preset scoring model and the target training information, and send it to the user terminal.
  • an embodiment of the present application further provides a computer device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer
  • the program implements the artificial intelligence-based training method described in the first aspect above.
  • an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when executed by a processor, the computer program causes the processor to execute the above-mentioned first step.
  • the embodiments of the present application provide an artificial intelligence-based training method, apparatus, computer equipment, and storage medium.
  • the customer classification information and business scenario process obtained by statistics the historical business processing records are split to obtain the classified sample data information, the target training information matching the training request information is obtained from the classified sample data information, and the corresponding virtual questioning video is generated. , obtain the reply video information fed back by the user, score the reply video information, and send the reply score information to the user terminal.
  • the classified sample data information containing a large amount of question and answer information is obtained based on the historical business processing records containing multiple customers, and the reply scoring information of the reply video information is obtained according to the scoring model and the target training information matching the training request information, so as to improve the The coverage of online training and the accurate scoring of response video information can greatly improve the training efficiency of online training.
  • FIG. 1 is a schematic flowchart of an artificial intelligence-based training method provided by an embodiment of the present application
  • FIG. 2 is a schematic diagram of an application scenario of an artificial intelligence-based training method provided by an embodiment of the present application
  • FIG. 3 is a schematic diagram of a sub-flow of an artificial intelligence-based training method provided in an embodiment of the present application
  • FIG. 4 is a schematic diagram of another sub-flow of the artificial intelligence-based training method provided by the embodiment of the present application.
  • FIG. 5 is a schematic diagram of another sub-flow of the artificial intelligence-based training method provided by the embodiment of the present application.
  • FIG. 6 is a schematic diagram of another sub-flow of the artificial intelligence-based training method provided by the embodiment of the present application.
  • FIG. 7 is another schematic flowchart of the artificial intelligence-based training method provided by the embodiment of the present application.
  • FIG. 8 is a schematic diagram of another sub-flow of the artificial intelligence-based training method provided by the embodiment of the present application.
  • FIG. 9 is a schematic block diagram of an artificial intelligence-based training device provided by an embodiment of the present application.
  • FIG. 10 is a schematic block diagram of a computer device provided by an embodiment of the present application.
  • FIG. 1 is a schematic flowchart of an artificial intelligence-based training method provided by an embodiment of the present application
  • FIG. 2 is a schematic diagram of an application scenario of the artificial intelligence-based training method provided by an embodiment of the present application.
  • the intelligent training method is applied in the management server 10, the method is executed by the application software installed in the management server 10, the management server 10 communicates with at least one user terminal 20, and the user terminal 20 can send training request information to the management server 10 to obtain a virtual question video, the user feeds back the reply video information to the management server 10 according to the virtual question video, and the management server 10 scores the answer video information to obtain the answer score information and feeds it back to the user terminal 20.
  • the administrator is the use of the management server 10.
  • the user of the user terminal 20 may be a service person of an enterprise.
  • the management server 10 is the server side used to execute the artificial intelligence-based training method to carry out intelligent online training for service personnel. Terminal devices that communicate, such as laptops, tablets, or mobile phones.
  • FIG. 2 only illustrates that one user terminal 20 performs information transmission with the management server 10 .
  • the management server 10 can also perform information transmission with multiple user terminals 20 at the same time. As shown in FIG. 1, the method includes steps S110-S150.
  • the customers in the historical business handling records are classified according to the preset customer information quantification model and the preset customer classification model to obtain customer classification information.
  • the administrator is the user of the management server, and the historical business handling records include business handling information of multiple customers who have completed business handling, as well as personal information of corresponding customers.
  • the customer information quantification model is a model that quantifies the personal information of customers, which can convert each customer's personal information into quantitative customer information for quantitative representation
  • the customer classification model is a model that classifies customers based on customer quantitative information.
  • the neural network model can classify and process the quantitative customer information according to the customer classification model to obtain the customer category to which each customer belongs, obtain the customer classification information, and obtain the customer classification information by counting the customers included in each customer category, then the customer classification information contains multiple customers that match each customer category.
  • the customer's personal information includes the customer's name, gender, age, occupation, hobbies, monthly income, marital status, childbearing status and other information
  • the business processing information is the text information that the customer communicates with the service personnel during the business processing stage.
  • step S110 includes sub-steps S111 , S112 and S113 .
  • the personal information of the customer in the historical business processing record is quantified to obtain customer quantitative information corresponding to each customer.
  • the customer information quantification model includes multiple quantitative items, and the number of quantitative items may be equal to or less than the number of information items contained in the customer's personal information. Converted to a vector value for representation, the multiple vector values corresponding to each customer's personal information are combined into the customer's quantitative customer information, and the customer quantitative information can be expressed as a multi-dimensional feature vector.
  • the range of the quantized value obtained by quantizing the item information is [0, 1].
  • the data corresponding to the keyword matching the non-numerical value in the quantization item is directly obtained as the quantized value corresponding to the non-numerical value.
  • the quantitative item of customer information quantitative model and marital status contains two keywords, "married” and “unmarried”, the data corresponding to "married” is "1", and the data corresponding to "unmarried” is " 0", the marital status in a customer's personal information is married, and the corresponding quantitative value is "1".
  • the corresponding quantification rule in the customer information quantification model is an activation function and an intermediate value. Get the corresponding quantized value.
  • the activation function can be expressed as Wherein, x is an item of information corresponding to the quantization item, and v is an intermediate value corresponding to the quantization item.
  • S112 input the customer quantitative information into the customer classification model in turn to obtain a customer category corresponding to each customer quantitative information;
  • S113 perform statistics on the customers included in each customer category to obtain customer classification information.
  • the customer classification model is constructed based on neural network.
  • the customer classification model consists of multiple input nodes, multiple output nodes and fully connected layers. Each input node corresponds to a vector value of one dimension in the customer feature vector. Each node corresponds to a customer category, and multiple output nodes corresponding to multiple customer categories can be configured in the customer classification model according to the classification purpose.
  • a fully-connected layer is included between the input node and the output node, and the fully-connected layer includes multiple feature units.
  • a first formula group is set between the input node and the fully-connected layer, and a second formula is set between the output node and the fully-connected layer. Group.
  • the first formula group includes formulas from all input nodes to all feature units, the formulas in the first formula group use the input node value as input value and the feature unit value as output value, and the second formula group includes all output nodes to all
  • the formula of the feature unit, the formulas in the second formula group all take the feature unit value as the input value and the output node value as the output value, and each formula included in the customer classification model has corresponding parameter values.
  • the output node value is also the matching probability between the customer quantitative information and the customer category corresponding to the output node. According to the matching probability between the customer quantitative information of a customer and each output node, the output node with the highest matching probability is selected.
  • the corresponding customer category is a classification result corresponding to the quantitative information of the customer, wherein the customer category may include urban white-collar workers, young students, housewives, pregnant mothers, and the like. According to the obtained classification results, the customers included in each customer category are counted to obtain customer classification information.
  • step S112 before step S112, it further includes: if the input training data set is received, performing iterative training on the customer classification model according to a preset gradient descent training model and the training data set to obtain the trained data set. Customer classification model.
  • the customer classification model is iteratively trained according to the preset gradient descent training model and the training data set to obtain a trained customer classification model.
  • the input of the training data set can be the administrator of the management server.
  • the customer classification model can be iteratively trained before using the customer classification model. That is, the parameter values in the first formula group and the second formula group of the customer classification model are adjusted, and the customer classification model obtained after training can more accurately classify customer quantitative information.
  • the gradient descent training model is a model for training the customer classification model.
  • the gradient descent training model includes the loss value calculation formula and the gradient calculation formula.
  • the training data set contains multiple pieces of training data.
  • the data contains the customer quantitative information of a customer and the classification label of the customer; input the customer quantitative information of a customer into the customer classification model to obtain the matching probability corresponding to multiple output nodes, and calculate the loss value according to the formula and classification label.
  • the corresponding loss value can be obtained by calculating the matching probability corresponding to the output node.
  • the updated value corresponding to each parameter in the first formula group and the second formula group can be calculated and obtained by updating the value.
  • the parameter value corresponding to each parameter is updated, and the process of updating the parameter value is the specific process of training the customer classification model.
  • the loss value calculation formula can be expressed as Among them, f s is the matching probability of an output node corresponding to the classification label, f n is the matching probability of the nth output node, and the value ranges of f s and f n are both [0, 1].
  • the updated value of each parameter in the customer classification model is calculated according to the gradient calculation formula, the loss value and the calculated value of the customer classification model. Specifically, the calculated value obtained by calculating the quantitative information of a customer with a parameter in the customer classification model is input into the gradient calculation formula, and combined with the above loss value, the updated value corresponding to the parameter can be calculated. Computed for gradient descent.
  • the gradient calculation formula can be expressed as:
  • ⁇ x is the original parameter value of the parameter x
  • ⁇ is the preset learning rate in the gradient calculation formula
  • the parameter value of the corresponding parameter in the customer classification model is updated according to the updated value of each parameter, so as to train the customer classification model.
  • the parameter value of each parameter in the customer classification model is updated correspondingly, that is, a training process of the customer classification model is completed.
  • another piece of training data in the training data set is calculated and processed again, and the above training process is repeated to implement iterative training of the customer classification model; when the calculated loss value is less than the preset value After the loss threshold or the training data in the training data set are used for training, the training process is terminated to obtain the trained customer classification model.
  • the historical business handling records are split according to the preset business scenario process and the customer classification information to obtain classified sample data information matching the customer classification information.
  • the business scenario process also includes the business handling information of each customer, and each piece of business handling information contains at least one question and answer message. Each question and answer message contains the question information entered by the customer and the answer information explained and guided by the service personnel.
  • the business scenario process includes multiple business scenarios, each business scenario involves a specific scenario for handling a business, and each business scenario includes multiple corresponding process nodes.
  • a customer's business handling information is classified, and then the business handling information of each customer category is split according to the process nodes corresponding to multiple business scenarios to obtain classified sample data information, then the classified sample data information involves multiple business scenarios.
  • Each business scenario corresponds to multiple Q&A messages for each customer category, and each Q&A message corresponds to a process node.
  • step S120 includes sub-steps S121 and S122.
  • the business handling information of the historical business handling record is classified according to the customer classification information to obtain business handling classification information. Specifically, the business handling information of the corresponding customer in the historical business handling record is obtained according to the customer's business handling information contained in each customer category in the customer classification information, and the business handling classification information is obtained, and the business handling classification information includes the business handling information corresponding to each customer category. .
  • S122 Acquire question and answer information in which the business handling information of each customer category in the business handling classification information matches each of the process nodes according to the process nodes included in the business scenario process, as the classification sample Data information.
  • each business scenario in the business scenario process obtain the question and answer information that matches the business handling information of each customer category with each process node, and obtain the classified sample data information, then classify the question and answer information in the sample data information It is classified into process nodes corresponding to multiple business scenarios, and each process node contains question and answer information corresponding to multiple customer categories.
  • step S123 is further included after step S122 .
  • S123 Perform deduplication processing on the question and answer information of the same customer category in each process node according to a preset deduplication rule, and use the deduplicated question and answer information of each process node as the classified sample data information.
  • the question and answer information of the same customer category in each process node is deduplicated according to a preset deduplication rule, and the deduplicated question and answer information of each process node is obtained.
  • Multiple question and answer information of the same customer category in a process node may be duplicated, and the question and answer information of the same customer category in each process node needs to be deduplicated through the deduplication rule.
  • the deduplication rule is that the question and answer information can be removed. Specific rules for reprocessing. Specifically, it can be determined whether two question information or two reply information in the question and answer information of the same customer category are repeated according to the deduplication rule, and if they are repeated, one of them is selected to be retained;
  • the deduplication rule can be set to determine whether the character repetition rate of the two pieces of text information is not less than 90%. If the rate is not less than 90%, it is judged that the two pieces of text information are repeated, otherwise it is judged that the two pieces of text information are not repeated.
  • the target training information in the sample data information that matches the training request information is acquired.
  • the service personnel can send training request information to the management server through the user terminal, and the training request information includes business scenario information, and the service personnel can send the training request information to the problem corresponding to the specific scene of a certain business.
  • the target training information that matches the training request information can be obtained from the sample data information.
  • the target training information includes multiple Q&A information corresponding to multiple process nodes in the business scenario information, and multiple Q&A information of the target training information. The information all belong to the same customer category.
  • step S130 includes sub-steps: randomly selecting question and answer information of a customer category in the classified sample data information as the target category question and answer information; obtaining the target category according to the business scenario information in the training request information A piece of question and answer information in the question and answer information that matches each process node of the business scenario information is combined to obtain target training information that matches the training request information.
  • the classified sample data information includes process nodes corresponding to multiple business scenarios, and each process node contains question and answer information corresponding to multiple customer categories. You can select a customer category as the target customer category, and obtain the target customer in each process node. The question and answer information that matches the category is used as the question and answer information of the target category.
  • the training request information includes business scenario information. According to the business scenario information in the training request information, the question and answer information that matches each process node of the business scenario information in the target category question and answer information is obtained, and the question and answer information that matches each process node is obtained. One question and answer information is randomly selected and combined from multiple question and answer information, and the target training information matching the training request information can be obtained.
  • S140 Generate a virtual question video corresponding to the question information in the target training information according to the virtual video generation model and send it to the user terminal, so as to obtain the reply video information fed back by the user terminal.
  • a virtual question video corresponding to the question information in the target training information is generated according to the virtual video generation model and sent to the user terminal to obtain the answer video information fed back by the user terminal, wherein the answer video information includes at least one segment Answer video.
  • the corresponding virtual questioning video can be generated through the virtual video generation model.
  • the virtual questioning video contains a virtual video corresponding to each process node, and each virtual video contains a process node corresponding to The service personnel receive the virtual question video through the user terminal, and perform video responses for each virtual video in the virtual question video to obtain the corresponding reply video information and feed it back to the management server.
  • step S140 includes sub-steps S141 , S142 and S143 .
  • S141 Acquire a category template in the virtual video generation model that matches the customer category of the target training information.
  • the virtual video generation model includes multiple category templates, and each category template matches a customer category, then a category template in the virtual video generation model that matches the customer category of the target training information can be obtained.
  • the category template contains a voice dictionary, and the question information recorded in the target training information can be converted into voice question information through the voice dictionary.
  • the voice question information contains the voice information corresponding to each question information, that is, a question information.
  • the speech dictionary contains multiple phrases and multiple words, each phrase corresponds to a phrase pronunciation, and each word corresponds to a word pronunciation, obtain the question information corresponding to a process node in the target training information, and use the question information in the question information.
  • the characters in the phonetic dictionary are matched with the phrases in the phonetic dictionary, and the phrase pronunciation of the corresponding phrase is obtained, and the remaining characters in the question information that do not match the phrase are matched with the words in the phonetic dictionary, and the corresponding word
  • a piece of voice information corresponding to the question information can be obtained by combining the pronunciation of the phrase and the pronunciation of the word according to the sequence of the characters in the question information.
  • a question message is "I want to know about this major disease risk”
  • the phrase corresponding to "want” in the phonetic dictionary is pronounced “xi ⁇ ng, yào”
  • the corresponding pronunciation of "understand” is “li ⁇ o, ji ⁇ ”
  • the corresponding pronunciation of "this” is “zhè, gè”
  • the corresponding pronunciation of "major disease” is “zh ⁇ ng, dà, j ⁇ , b ⁇ ng”
  • the word corresponding to "I” is pronounced “w ⁇ ”
  • the word corresponding to "risk” is pronounced “w ⁇ ”.
  • the corresponding word is pronounced as “xi ⁇ n”
  • a corresponding piece of voice information can be obtained by combining the above pronunciations.
  • the portrait model is a virtual portrait that matches a customer category, and the voice question information and video information can be combined.
  • a piece of virtual video corresponding to a piece of question information is obtained, and a virtual video corresponding to each piece of question information is obtained as a virtual question video.
  • step S1510 is included before step S150 .
  • the business corpus database of a business scenario can train the initialized neural network to obtain a neural network matching the business scenario, and multiple business corpus databases can respectively train the initialized neural network to obtain a neural network containing multiple neural networks.
  • the neural network set of the network, each neural network in the neural network set is matched with a business scenario.
  • the training rules include proportional values, loss function calculation formulas, and gradient calculation formulas.
  • the training rules are the rule information for training the initialized neural network.
  • the process of training the initialized neural network based on a business corpus database of a business scenario includes steps ( 1) to (6):
  • Part of the corpus data corresponding to the ratio value is randomly selected from one of the business corpus databases as the target corpus data.
  • Each piece of corpus data in the business corpus database is a complete sentence, and each piece of corpus data is composed of multiple characters.
  • the training rule is also set with a proportional value, which can be randomly selected from a business corpus database according to the proportional value.
  • a corresponding amount of corpus data is used as the target corpus data, for example, the proportion value can be set to 10-90%.
  • Each corpus is composed of multiple characters, and any character in each corpus can be masked to obtain corpus processing data including the masked characters.
  • a target corpus data is "want to learn about property insurance”
  • the corpus data obtained after random masking processing is "want to solve property insurance with X", where "X" represents the masked character.
  • step (2) may further include: randomly replacing the covered characters in the part of the corpus processing data corresponding to the ratio value in the corpus processing data.
  • the masked characters in part of the corpus processing data can also be randomly replaced with other characters according to the above ratio value.
  • a certain corpus processing data is “want to understand property insurance”, and the corpus processing data obtained after random replacement is "want to understand property insurance”.
  • the target corpus data and the corpus processing data are respectively converted to obtain a corpus feature vector and a corpus processing feature vector.
  • Each character can be matched to a corresponding feature code in the conversion dictionary, and the characters contained in the target corpus data can be converted according to the conversion dictionary to obtain a corpus feature vector of size (1, M), which represents the corpus.
  • the feature vector is 1 row and M columns, and the feature code of the target corpus data is used as the value to fill the corpus feature vector, and the unfilled value is marked as "0".
  • a piece of corpus processing data corresponding to the target corpus data is converted by the same conversion method to obtain a corpus processing feature vector.
  • the feature code corresponding to "Think” in the conversion dictionary is “2318”; the feature code of “Li” is “6522”, the feature code of “Solution” is “7351”, and the feature code of "One” is “0100” , the feature code of "Xia” is “8631”, the feature code of "Cai” is “3621”, the feature code of "Production” is “1531”, the feature code of "Bao” is “4280”, and the feature code of "Insurance” The code is “6634”. "101” represents the start feature code of the sentence, and “102” represents the end feature code of the sentence. Then the corresponding combination of "I want to understand property insurance” can be expressed as [101, 2318, 6522, 7351, 0100, 8631, 3621, 1531, 4280, 6634, 102, ..., 0].
  • the initialized neural network consists of an input layer, multiple intermediate layers, and an output layer.
  • the input layer and the intermediate layer, between the intermediate layer and other intermediate layers, and between the intermediate layer and the output layer are all related by association formulas.
  • the number of input nodes contained in the input layer corresponds to the length of the first feature vector, then each vector value in the first feature vector corresponds to one input node, and the first feature vector is input into the initialized neural network for calculation, that is,
  • the output result can be obtained from its output layer, and the output result is represented by an array (M, N).
  • the output result corresponding to the corpus feature vector is a corpus array, and the size of the corpus array is M rows and N columns.
  • the corpus processing feature vector is input into the recognition template for calculation, and the corpus processing array is obtained. value range.
  • the loss value can be used to quantify the difference between the corpus array and the corpus processing array. Specifically, calculating the loss value between the corpus array S 1 and the corpus processing array S 2 can be obtained by calculating the loss function calculation formula Among them, Ls is the calculated loss value, a xy is the value of the xth row and the yth column of the corpus array S1, bxy is the value of the xth row and the yth column of the corpus processing array S2, and M is the corpus array S. 1 is the total number of rows, and N is the total number of columns in the corpus array S1.
  • the update value of each parameter in the initialized neural network of the recognition template is calculated to update the parameter of the parameter value.
  • the calculated value obtained by calculating the corpus feature vector with a parameter in the initialized neural network is input into the gradient calculation formula, and combined with the above loss value, the update value corresponding to the parameter can be calculated.
  • This calculation process also That is, the gradient descent calculation, and the process of updating the parameter value is the same as the above process, which will not be repeated here.
  • Response scoring information corresponding to the reply video information is obtained according to a preset scoring model and the target training information, and sent to the user terminal, wherein the scoring model includes speech conversion rules, node conversion keywords, and feature dictionaries. , neural network set and scoring calculation formula, the neural network in the neural network set is constructed based on BERT (Bidirectional Encoder Representations from Transformers) network.
  • the response video information can be analyzed according to the scoring model to obtain the response score information and send it to the user terminal, and the service personnel can obtain the corresponding response score through the user terminal to obtain the training result.
  • the reply video information includes multiple reply videos, and each reply video matches a process node in the business scenario information, and the score values of the reply videos in the reply video information can be obtained separately as the reply score information.
  • step S150 includes sub-steps S151 , S152 , S153 , S154 , S155 and S156 .
  • the speech conversion rules include an acoustic model, a pinyin dictionary, and a semantic parsing model.
  • the voice information contained in a reply video is composed of phonemes of the pronunciation of a plurality of characters, and the phonemes of a character include the frequency and timbre of the pronunciation of the character.
  • the acoustic model contains the phonemes of all character pronunciations. By matching the phonetic information with all the phonemes in the acoustic model, the phonemes of a single character in the phonetic information can be segmented, and the answer is finally obtained through segmentation. phoneme.
  • the pinyin dictionary contains the phoneme information corresponding to all character pinyin.
  • the phoneme of a single character can be converted into the phoneme matching the phoneme in the pinyin dictionary. Convert all phonemes contained in the voice information of the reply video into pinyin information.
  • the semantic analysis model includes the mapping relationship between the pinyin information and the text information, and the obtained pinyin information can be semantically analyzed through the mapping relationship included in the semantic analysis model to convert the pinyin information into the corresponding text information.
  • the obtained text information is a piece of text information corresponding to the reply video.
  • the text information corresponding to the pinyin "bàn, l ⁇ " in the semantic parsing model is "handle”.
  • the node conversion keyword contains the conversion keyword corresponding to each process node, and the conversion keyword of the process node corresponding to the text information in the node conversion keyword can be obtained, and whether the text information matches the conversion keyword can be determined. Whether the text information contains the conversion keyword is used to obtain a keyword judgment result of whether the text information matches the conversion keyword. If the text information matches the conversion keyword, it means that the reply video corresponding to the text information meets the core demands of the corresponding customer category, which can trigger the corresponding conversion node in the business scenario process and promote the business processing process.
  • the business After triggering the corresponding conversion node, the business is processed It can be smoothly transferred from the current process node to the next process node; if the text information does not match the conversion keyword, it means that the reply video corresponding to the text information does not meet the core demands of the corresponding customer category.
  • S153 Convert the text information and the reply information of the process node according to the feature dictionary to obtain a first feature vector and a second feature vector.
  • the conversion dictionary is a dictionary that converts characters. Each character can be matched with a corresponding feature code in the conversion dictionary. Then, the characters contained in the text information can be converted according to the conversion dictionary, and the corresponding The feature codes are combined to obtain the first feature vector, and the obtained first feature vector is to represent the features of the text information in a vector manner.
  • the first eigenvector is filled, and the unfilled values in the first eigenvector are marked as "0".
  • the reply information of the process node matching the text information is converted by the same conversion method to obtain the second feature vector.
  • a target neural network in the neural network set that matches the business scenario information is determined.
  • the neural network set contains multiple neural networks, all of which are constructed based on the BERT network.
  • Each neural network matches a business scenario, that is, a neural network is suitable for a matching business scenario. specific locale.
  • the initialized neural network is pre-trained by using the corpus data of a business scenario to obtain a neural network suitable for the language environment of the business scenario, and the neural networks corresponding to multiple business scenarios are combined into the neural network set.
  • the first feature vector and the second feature vector can be scored based on the language environment of the target neural network, and a more accurate scoring result can be obtained.
  • S155 Input the first feature vector and the second feature vector into the target neural network for calculation to obtain a first array and a second array, respectively.
  • the process of calculating the first feature vector and the second feature vector is the same as the process of calculating the corpus feature vector or corpus processing feature vector, and the output result is represented by an array (M, N), which is the same as the first feature vector.
  • the output result corresponding to the vector is the first array, the size of the first array is also M rows and N columns, the output result of the second feature vector is the second array, and the size of the second array is also (M, N), where the first Each value in the first array and the second array belongs to the value range of [0, 1].
  • S156 Calculate the keyword judgment result, the first array and the second array according to the scoring formula to obtain a corresponding scoring value.
  • the obtained keyword judgment result, the first array and the second array can be calculated by the scoring calculation formula to obtain a scoring value corresponding to the reply video.
  • the keyword judgment result is first converted into a corresponding one.
  • Coefficient value is expressed. For example, if the keyword judgment result is that the text information matches the conversion keyword, the corresponding coefficient value is 1; if the keyword judgment result is that the text information does not match the conversion keyword, the corresponding coefficient value is 0.3.
  • the first array is S a
  • the second array is S b
  • the loss value between the first array and the second array is calculated according to the above calculation formula.
  • the technical methods in this application can be applied to smart government affairs/smart city management/smart community/smart security/smart logistics/smart medical care/smart education/smart environmental protection/smart transportation and other application scenarios including intelligent online training for service personnel, So as to promote the construction of smart cities.
  • historical business processing records are split according to the customer classification information and business scenario process obtained by statistics to obtain classified sample data information, and the relevant data is obtained from the classified sample data information.
  • the training request information matches the target training information and generates a corresponding virtual question video, obtains the reply video information fed back by the user, scores the reply video information, and sends the reply score information to the user terminal.
  • the classified sample data information containing a large amount of question and answer information is obtained based on the historical business processing records containing multiple customers, and the reply scoring information of the reply video information is obtained according to the scoring model and the target training information matching the training request information, so as to improve the The coverage of online training and the accurate scoring of reply video information can greatly improve the training efficiency of online training.
  • the embodiments of the present application further provide an artificial intelligence-based training device, which is used to execute any one of the foregoing artificial intelligence-based training methods.
  • FIG. 9 is a schematic block diagram of an artificial intelligence-based training apparatus provided by an embodiment of the present application.
  • the artificial intelligence-based training device can be configured in the management server 10 .
  • the artificial intelligence-based training device 100 includes a customer classification information acquisition unit 110 , a classification sample data information acquisition unit 120 , a target training information acquisition unit 130 , a virtual question video transmission unit 140 , and a response score information transmission unit 150 .
  • the customer classification information acquisition unit 110 is configured to, if receiving the historical business handling records input by the administrator, classify and count the customers in the historical business handling records according to the preset customer information quantification model and the preset customer classification model to obtain statistics. Get customer classification information.
  • the customer classification information acquisition unit 110 includes subunits: a customer quantitative information acquisition unit, a customer category acquisition unit, and a customer statistics unit.
  • the customer quantitative information acquisition unit is used to quantify the personal information of the customers in the historical business processing records according to the customer information quantitative model to obtain customer quantitative information corresponding to each customer; the customer category acquisition unit is used to quantify the customer information.
  • the customer quantitative information is sequentially input into the customer classification model to obtain the customer category corresponding to each customer quantitative information; the customer statistics unit is configured to perform statistics on the customers included in each of the customer categories to obtain the customer classification information.
  • the customer classification information acquisition unit 110 further includes a subunit: a customer classification model training unit.
  • the customer classification model training unit is configured to iteratively train the customer classification model according to a preset gradient descent training model and the training data set to obtain a trained customer classification model if the input training data set is received.
  • the classified sample data information obtaining unit 120 is configured to split the historical business handling records according to the preset business scenario flow and the customer classification information to obtain classified sample data information matching the customer classification information.
  • the classification sample data information obtaining unit 120 includes subunits: a business handling classification information obtaining unit and a question-and-answer information matching unit.
  • a business handling classification information acquisition unit used for classifying the business handling information in the historical business handling records according to the customer classification information, to obtain business handling classification information
  • a question-and-answer information matching unit used for classifying the business handling information according to the business scenario flow
  • the process node included in the process node obtains the question and answer information in which the business handling information of each customer category in the business handling classification information is matched with each of the process nodes respectively, as the classification sample data information.
  • the classified sample data information obtaining unit 120 further includes a subunit: a deduplication processing unit.
  • the deduplication processing unit is used to deduplicate the question and answer information of the same customer category in each process node according to the preset deduplication rule, and use the deduplicated question and answer information of each process node as the classification sample data information.
  • the target training information obtaining unit 130 is configured to obtain target training information matching the training request information in the classified sample data information if the training request information from the user terminal is received.
  • the virtual question video sending unit 140 is configured to generate a virtual question video corresponding to the question information in the target training information according to the virtual video generation model and send it to the user terminal to obtain the reply video information fed back by the user terminal, wherein , the reply video information includes at least one reply video.
  • the virtual question video sending unit 140 includes subunits: a category template matching unit and a question information conversion unit.
  • a category template matching unit for acquiring a category template in the virtual video generation model that matches the customer category of the target training information; a question information conversion unit for matching the The question information contained in the target training information is converted to obtain corresponding voice question information.
  • a virtual questioning video generating unit configured to combine the voice questioning information with the portrait model in the category template to generate a virtual questioning video.
  • the artificial intelligence-based training apparatus 100 further includes: a neural network set acquisition unit.
  • the neural network set acquisition unit is used to respectively perform language environment training on the initialized neural network according to the business corpus databases of multiple business scenarios and the preset training rules to obtain the neural network set, and the initialized neural network is based on the BERT network build get.
  • the reply scoring information sending unit 150 is configured to acquire reply scoring information corresponding to the reply video information according to a preset scoring model and the target training information, and send it to the user terminal.
  • the reply scoring information sending unit 150 includes subunits: a text information obtaining unit, a keyword matching unit, a feature vector obtaining unit, a target neural network obtaining unit, an array obtaining unit, and a scoring calculating unit.
  • a text information acquisition unit used for converting a piece of reply video in the reply video information according to the voice conversion rule to obtain corresponding text information
  • a keyword matching unit used for judging whether the text information is a key to node conversion The conversion keywords of the corresponding process nodes in the word are matched to obtain the keyword judgment result
  • the feature vector acquisition unit is used to convert the text information and the reply information of the process nodes according to the feature dictionary to obtain the first feature.
  • a target neural network acquisition unit for determining a target neural network matching the business scene information in the neural network set according to the business scene information
  • an array acquisition unit for The first feature vector and the second feature vector are input into the target neural network for calculation to obtain a first array and a second array respectively
  • a scoring calculation unit is used to determine the keyword judgment results, The first array and the second array are calculated to obtain a corresponding score value.
  • the artificial intelligence-based training device provided in the embodiment of the present application applies the above-mentioned artificial intelligence-based training method, and splits the historical business processing records according to the customer classification information obtained by statistics and the business scenario process to obtain the classified sample data information.
  • the target training information matching the training request information is obtained from the classified sample data information and a corresponding virtual question video is generated, and the reply video information fed back by the user is obtained for scoring, and the reply scoring information is obtained and sent to the user terminal.
  • the classified sample data information containing a large amount of question and answer information is obtained based on the historical business processing records containing multiple customers, and the reply scoring information of the reply video information is obtained according to the scoring model and the target training information matching the training request information, so as to improve the The coverage of online training and the accurate scoring of reply video information can greatly improve the training efficiency of online training.
  • the above-mentioned artificial intelligence-based training apparatus can be implemented in the form of a computer program, and the computer program can be executed on a computer device as shown in FIG. 10 .
  • FIG. 10 is a schematic block diagram of a computer device provided by an embodiment of the present application.
  • the computer device may be a management server 10 for executing an artificial intelligence-based training method for intelligent online training of service personnel.
  • the computer device 500 includes a processor 502 , a memory and a network interface 505 connected by a system bus 501 , wherein the memory may include a non-volatile storage medium 503 and an internal memory 504 .
  • the nonvolatile storage medium 503 can store an operating system 5031 and a computer program 5032 .
  • the computer program 5032 when executed, can cause the processor 502 to perform artificial intelligence-based training methods.
  • the processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500 .
  • the internal memory 504 provides an environment for the execution of the computer program 5032 in the non-volatile storage medium 503, and when the computer program 5032 is executed by the processor 502, can cause the processor 502 to execute the artificial intelligence-based training method.
  • the network interface 505 is used for network communication, such as providing transmission of data information.
  • the network interface 505 is used for network communication, such as providing transmission of data information.
  • FIG. 10 is only a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied.
  • the specific computer device 500 may include more or fewer components than shown, or combine certain components, or have a different arrangement of components.
  • the processor 502 is configured to run the computer program 5032 stored in the memory, so as to realize the corresponding functions in the above-mentioned artificial intelligence-based training method.
  • the embodiment of the computer device shown in FIG. 10 does not constitute a limitation on the specific structure of the computer device. Either some components are combined, or different component arrangements.
  • the computer device may only include a memory and a processor.
  • the structures and functions of the memory and the processor are the same as those of the embodiment shown in FIG. 10 , which will not be repeated here.
  • the processor 502 may be a central processing unit (Central Processing Unit, CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor can be a microprocessor or the processor can also be any conventional processor or the like.
  • a computer-readable storage medium may be a non-volatile computer-readable storage medium.
  • the computer-readable storage medium stores a computer program, wherein the computer program implements the steps included in the above-mentioned artificial intelligence-based training method when executed by the processor.
  • the disclosed apparatus, apparatus and method may be implemented in other manners.
  • the apparatus embodiments described above are only illustrative.
  • the division of the units is only logical function division.
  • there may be other division methods, or units with the same function may be grouped into one Units, such as multiple units or components, may be combined or may be integrated into another system, or some features may be omitted, or not implemented.
  • the shown or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may also be electrical, mechanical or other forms of connection.
  • the units described as separate components may or may not be physically separated, and components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solutions of the embodiments of the present application.
  • each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.
  • the above-mentioned integrated units may be implemented in the form of hardware, or may be implemented in the form of software functional units.
  • the integrated unit if implemented in the form of a software functional unit and sold or used as an independent product, may be stored in a computer-readable storage medium.
  • the read storage medium includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application.
  • the aforementioned computer-readable storage medium includes: U disk, mobile hard disk, Read-Only Memory (ROM, Read-Only Memory), magnetic disk or optical disk and other media that can store program codes.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Business, Economics & Management (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Strategic Management (AREA)
  • Evolutionary Computation (AREA)
  • Tourism & Hospitality (AREA)
  • Educational Technology (AREA)
  • Educational Administration (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Primary Health Care (AREA)
  • General Business, Economics & Management (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种基于人工智能的培训方法、装置、计算机设备及存储介质，方法包括：根据统计得到的客户分类信息及业务场景流程对历史业务办理记录进行拆分以得到分类样本数据信息，从分类样本数据信息中获取与培训请求信息相匹配的目标培训信息并生成对应的虚拟提问视频，获取用户反馈的答复视频信息进行评分得到答复评分信息并发送至用户终端。该培训方法基于语音交互技术，属于人工智能领域，基于包含多个客户的历史业务办理记录获取包含大量问答信息的分类样本数据信息，根据评分模型及与培训请求信息相匹配的目标培训信息获取答复视频信息的答复评分信息，以提高线上培训的覆盖面并实现对答复视频信息进行准确评分，从而大幅提升线上培训效率。

Description

基于人工智能的培训方法、装置、计算机设备及存储介质
本申请要求于2020年11月3日提交中国专利局、申请号为202011209903.2，发明名称为“基于人工智能的培训方法、装置、计算机设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及人工智能技术领域，属于智慧城市中对服务人员进行智能线上培训的应用场景，尤其涉及一种基于人工智能的培训方法、装置及计算机设备。
背景技术
在进行业务办理过程中，企业通常会选择一位服务人员作为客户的对接人并为客户提供服务，在线上进行远程视频沟通作为一种为客户提供服务的新形势越来越受到客户青睐，然而企业的新服务人员在通过进行远程视频沟开展业务过程中经常会因经验不足，而无法基于当前所处的业务办理阶段向客户准确推送相关信息。为提高服务人员的业务能力以更好地为客户提供服务，可获取客户进行业务办理的历史业务办理记录作为参考案例对服务人员进行培训，然而历史业务办理记录中客户的办理流程仅是针对单一客户在办理业务过程中遇到的实际问题，因此这一培训过程无法涉及业务办理过程中可能出现的各种问题，若解决这一问题可从历史业务办理记录中获取大量客户的办理流程作为大量参考案例以供服务人员进行学习，然而发明人发现这一培训方式需耗费大量时间，培训效率较低；并且服务人员采用这一培训方式难以对培训效果进行评价，导致对服务人员进行培训时的培训效果不理想。因此现有技术方法在对服务人员进行培训时存在培训效率较低的问题。
发明内容
本申请实施例提供了一种基于人工智能的培训方法、装置、计算机设备及存储介质，旨在解决现有技术方法中对服务人员进行培训时所存在的培训效率较低的问题。
第一方面，本申请实施例提供了一种基于人工智能的培训方法，其包括：
若接收到管理员输入的历史业务办理记录，根据预置的客户信息量化模型及预置的客户分类模型对所述历史业务办理记录中的客户进行分类统计以获取客户分类信息；
根据预置的业务场景流程及所述客户分类信息对所述历史业务办理记录进行拆分以获取与所述客户分类信息相匹配的分类样本数据信息；
若接收到来自所述用户终端的培训请求信息，获取所述分类样本数据信息中与所述培训请求信息相匹配的目标培训信息；
根据虚拟视频生成模型生成与所述目标培训信息中提问信息对应的虚拟提问视频发送至所述用户终端，以获取所述用户终端所反馈的答复视频信息，其中，所述答复视频信息包含至少一段答复视频；
根据预置的评分模型及所述目标培训信息获取与所述答复视频信息对应的答复评分信息并发送至所述用户终端。
第二方面，本申请实施例提供了一种基于人工智能的培训装置，其包括：
客户分类信息获取单元，用于若接收到管理员输入的历史业务办理记录，根据预置的客户信息量化模型及预置的客户分类模型对所述历史业务办理记录中的客户进行分类统计以获取客户分类信息；
分类样本数据信息获取单元，用于根据预置的业务场景流程及所述客户分类信息对所述历史业务办理记录进行拆分以获取与所述客户分类信息相匹配的分类样本数据信息；
目标培训信息获取单元，用于若接收到来自所述用户终端的培训请求信息，获取所述分类样本数据信息中与所述培训请求信息相匹配的目标培训信息；
虚拟提问视频发送单元，用于根据虚拟视频生成模型生成与所述目标培训信息中提问信息对应的虚拟提问视频发送至所述用户终端，以获取所述用户终端所反馈的答复视频信息，其中，所述答复视频信息包含至少一段答复视频；
答复评分信息发送单元，用于根据预置的评分模型及所述目标培训信息获取与所述答复视频信息对应的答复评分信息并发送至所述用户终端。
第三方面，本申请实施例又提供了一种计算机设备，其包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现上述第一方面所述的基于人工智能的培训方法。
第四方面，本申请实施例还提供了一种计算机可读存储介质，其中所述计算机可读存储介质存储有计算机程序，所述计算机程序当被处理器执行时使所述处理器执行上述第一方面所述的基于人工智能的培训方法。
本申请实施例提供了一种基于人工智能的培训方法、装置、计算机设备及存储介质。根据统计得到的客户分类信息及业务场景流程对历史业务办理记录进行拆分以得到分类样本数据信息，从分类样本数据信息中获取与培训请求信息相匹配的目标培训信息并生成对应的虚拟提问视频，获取用户反馈的答复视频信息进行评分得到答复评分信息并发送至用户终端。通过上述方法，基于包含多个客户的历史业务办理记录获取包含大量问答信息的分类样本数据信息，根据评分模型及与培训请求信息相匹配的目标培训信息获取答复视频信息的答复评分信息，以提高线上培训的覆盖面并实现对答复视频信息进行准确评分，从而大幅提升进行线上培训的培训效率。
附图说明
为了更清楚地说明本申请实施例技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本申请实施例提供的基于人工智能的培训方法的流程示意图；
图2为本申请实施例提供的基于人工智能的培训方法的应用场景示意图；
图3为本申请实施例提供的基于人工智能的培训方法的子流程示意图；
图4为本申请实施例提供的基于人工智能的培训方法的另一子流程示意图；
图5为本申请实施例提供的基于人工智能的培训方法的另一子流程示意图；
图6为本申请实施例提供的基于人工智能的培训方法的另一子流程示意图；
图7为本申请实施例提供的基于人工智能的培训方法的另一流程示意图；
图8为本申请实施例提供的基于人工智能的培训方法的另一子流程示意图；
图9为本申请实施例提供的基于人工智能的培训装置的示意性框图；
图10为本申请实施例提供的计算机设备的示意性框图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
应当理解，当在本说明书和所附权利要求书中使用时，术语“包括”和“包含”指示所描述特征、整体、步骤、操作、元素和/或组件的存在，但并不排除一个或多个其它特征、整体、步骤、操作、元素、组件和/或其集合的存在或添加。
还应当理解，在此本申请说明书中所使用的术语仅仅是出于描述特定实施例的目的而并不意在限制本申请。如在本申请说明书和所附权利要求书中所使用的那样，除非上下文清楚地指明其它情况，否则单数形式的“一”、“一个”及“该”意在包括复数形式。
还应当进一步理解，在本申请说明书和所附权利要求书中使用的术语“和/或”是指相关联列出的项中的一个或多个的任何组合以及所有可能组合，并且包括这些组合。
请参阅图1及图2，图1是本申请实施例提供的基于人工智能的培训方法的流程示意图，图2为本申请实施例提供的基于人工智能的培训方法的应用场景示意图，该基于人工智能的培训方法应用于管理服务器10中，该方法通过安装于管理服务器10中的应用软件进行执行，管理服务器10与至少一台用户终端20进行通信，用户终端20可发送培训请求信息至管理服务器10以获取虚拟提问视频，用户根据虚拟提问视频反馈答复视频信息至管理服务器10，管理服务器10对答复视频信息进行评分得到答复评分信息并反馈至用户终端20，管理员即为管理服务器10的使用者，用户终端20的使用者可以是企业的服务人员。管理服务器10即是用于执行基于人工智能的培训方法以对服务人员进行智能线上培训的服务器端，管理服务器10可以是企业所设置的服务器端，用户终端20即是可用于与管理服务器10进行通信的终端设备，例如笔记本电脑、平板电脑或手机等。图2中仅仅示意出一台用户终端20与管理服务器10进行信息传输，在实际应用中，该管理服务器10也可与多台用户终端20同时进行信息传输。如图1所示，该方法包括步骤S110～S150。
S110、若接收到管理员输入的历史业务办理记录，根据预置的客户信息量化模型及预置的客户分类模型对所述历史业务办理记录中的客户进行分类统计以获取客户分类信息。
若接收到输入的历史业务办理记录，根据预置的客户信息量化模型及预置的客户分类模型对所述历史业务办理记录中的客户进行分类以获取客户分类信息。其中，管理员即为管理服务器的使用者，历史业务办理记录包含已完成业务办理的多个客户的业务办理信息，以及相应客户的个人信息。具体的，客户信息量化模型即为对客户的个人信息进行量化的模型，可将每一客户的个人信息转换为客户量化信息进行量化表示，客户分类模型即为基于客户量化信息对客户进行分类的神经网络模型，可根据客户分类模型对客户量化信息进行分类处理得到每一客户所属的客户类别，得到客户分类信息，并对每一客户类别包含的客户进行统计得到客户分类信息，则客户分类信息中包含与每一客户类别相匹配的多个客户。其中，客户的个人信息中包括客户姓名、性别、年龄、职业、兴趣爱好、月收入、婚姻状态、生育状态等信息，业务办理信息即为客户在业务办理阶段与服务人员进行沟通的文字信息。
在一实施例中，如图3所示，步骤S110包括子步骤S111、S112和S113。
S111、根据所述客户信息量化模型对所述历史业务办理记录中客户的个人信息进行量化得到与每一客户对应的客户量化信息。
根据所述客户信息量化模型对所述历史业务办理记录中客户的个人信息进行量化得到与每一客户对应的客户量化信息。具体的，客户信息量化模型中包含多个量化项目，量化项目的数量可以等于或少于客户的个人信息中所包含的信息项数，每一量化项目均可将个人信息中对应的一项信息转换为一个向量值进行表示，每一客户的个人信息所对应的多个向量值即组合为该客户的客户量化信息，客户量化信息可表示为一个多维的特征向量，对每一量化项 目的一项信息进行量化所得量化值的范围均为[0，1]。
具体的，对于与量化项目对应的信息以非数值方式进行表示的情况，则直接获取量化项目中与该非数值相匹配的关键字所对应的数据，作为与该非数值对应的量化值。例如，客户信息量化模型与婚姻状态这一量化项目中包含“已婚”及“未婚”两个关键字，与“已婚”对应的数据为“1”、与“未婚”对应的数据为“0”，某客户的个人信息中婚姻状态为已婚，则对应的量化值为“1”。
对于与量化项目对应的信息以数值方式表示的情况，客户信息量化模型中对应的量化规则为一个激活函数及一个中间值，根据激活函数对中间值及量化项目的一项信息进行计算，即可得到对应的量化值。
例如，激活函数可表示为
Figure PCTCN2021091287-appb-000001
其中，x为与量化项目对应的一项信息，v为与该量化项目对应的中间值。与月收入这一量化项目对应的中间值为v＝8000，某客户对应的客户信息中月收入为x＝10000，则根据上述激活函数计算得到对应的量化值为0.4378。
S112、将所述客户量化信息依次输入所述客户分类模型以获取与每一客户量化信息对应的客户类别；S113、对每一所述客户类别包含的客户进行统计得到客户分类信息。
客户分类模型是基于神经网络所构建的，客户分类模型中由多个输入节点、多个输出节点及全连接层组成，每一输入节点均对应客户特征向量中一个维度的向量值，每一输出节点均对应一个客户类别，根据分类目的可在客户分类模型中配置与多个客户类别对应的多个输出节点。输入节点与输出节点之间包含全连接层，全连接层中包含多个特征单元，输入节点与全连接层之间设置有第一公式组，输出节点与全连接层之间设置有第二公式组。其中，第一公式组包含所有输入节点至所有特征单元的公式，第一公式组中的公式均以输入节点值作为输入值、特征单元值作为输出值，第二公式组包含所有输出节点至所有特征单元的公式，第二公式组中的公式均以特征单元值作为输入值、输出节点值作为输出值，客户分类模型中所包含的每一公式中均拥有对应的参数值。输出节点值也即是客户量化信息与该输出节点对应的客户类别之间的匹配概率，根据计算得到某一客户的客户量化信息与每一输出节点的匹配概率，选择匹配概率最高的一个输出节点所对应的客户类别作为与该客户量化信息对应的分类结果，其中，客户类别可以包括都市白领、青年学生、家庭主妇、孕期妈妈等。根据所得到的分类结果，对每一客户类别包含的客户进行统计得到客户分类信息。
在一实施例中，步骤S112之前还包括：若接收到所输入的训练数据集，根据预置的梯度下降训练模型及所述训练数据集对所述客户分类模型进行迭代训练以得到训练后的客户分类模型。
若接收到所输入的训练数据集，根据预置的梯度下降训练模型及所述训练数据集对所述客户分类模型进行迭代训练以得到训练后的客户分类模型。其中输入训练数据集的可以是管理服务器的管理员，为使客户分类模型在对客户量化信息进行分类时可以有更高的准确率，可在使用客户分类模型之前对客户分类模型进行迭代训练，也即是对客户分类模型的第一公式组及第二公式组中的参数值进行调整，训练后所得到的客户分类模型可以对客户量化信息进行更精准的分类。梯度下降训练模型即为对客户分类模型进行训练的模型，梯度下降训练模型中包括损失值计算公式及梯度计算公式，训练数据集中包含多条训练数据，每一条训练数据对应一个客户，每一训练数据中均包含一个客户的客户量化信息以及该客户的分类标签；将某一客户的客户量化信息输入客户分类模型得到多个输出节点对应的匹配概率，根据损失值计算公式及分类标签对多个输出节点对应的匹配概率进行计算即可得到对应的损失值，根 据损失值及梯度计算公式即可计算得到第一公式组及第二公式组中每一参数对应的更新值，通过更新值即可对每一参数对应的参数值进行更新，这一对参数值进行更新的过程即为对客户分类模型进行训练的具体过程。
具体的，损失值计算公式可表示为
Figure PCTCN2021091287-appb-000002
其中，f s为与分类标签对应的一个输出节点的匹配概率，f n为第n个输出节点的匹配概率，f s及f n的取值范围均为[0，1]。
根据所述梯度计算公式、所述损失值及所述客户分类模型的计算值计算得到所述客户分类模型中每一参数的更新值。具体的，将客户分类模型中一个参数对客户量化信息进行计算所得到的计算值输入梯度计算公式，并结合上述损失值，即可计算得到与该参数对应的更新值，这一计算过程也即为梯度下降计算。
具体的，梯度计算公式可表示为：
Figure PCTCN2021091287-appb-000003
其中，
Figure PCTCN2021091287-appb-000004
为计算得到的参数x的更新值，ω x为参数x的原始参数值，η为梯度计算公式中预置的学习率，
Figure PCTCN2021091287-appb-000005
为基于损失值及参数x对应的计算值对该参数x的偏导值(这一计算过程中需使用参数对应的计算值)。
根据每一所述参数的更新值对所述客户分类模型中对应参数的参数值进行更新，以对所述客户分类模型进行训练。基于所计算得到更新值对客户分类模型中每一参数的参数值对应更新，即完成对客户分类模型的一次训练过程。基于一次训练后所得到的客户分类模型对训练数据集中另一条训练数据再次进行计算处理，并重复上述训练过程，即可实现对客户分类模型进行迭代训练；当所计算得到的损失值小于预设的损失阈值或训练数据集中条训练数据均被用于训练后，即终止训练过程得到训练后的客户分类模型。
S120、根据预置的业务场景流程及所述客户分类信息对所述历史业务办理记录进行拆分以获取与所述客户分类信息相匹配的分类样本数据信息。
根据预置的业务场景流程及所述客户分类信息对所述历史业务办理记录进行拆分以获取与所述客户分类信息相匹配的分类样本数据信息。业务场景流程中还包括每一客户的业务办理信息，每一条业务办理信息中均包含至少一个问答信息，一条问答信息即为客户在进行某一项业务办理阶段与服务人员进行一轮问答沟通的信息，每一条问答信息中均包含客户所输入的提问信息及服务人员进行解释、引导的答复信息。业务场景流程中包含多个业务场景，每一业务场景涉及对一种业务进行办理的具体场景，每一业务场景包含对应的多个流程节点，可先根据客户分类信息对历史业务办理记录中每一客户的业务办理信息进行分类，再根据多个业务场景分别对应的流程节点对每一客户类别的业务办理信息进行拆分得到分类样本数据信息，则分类样本数据信息中涉及多个业务场景，每一业务场景均对应每一客户类别的多条问答信息，每一条问答信息均与一个流程节点相对应。
在一实施例中，如图4所示，步骤S120包括子步骤S121和S122。
S121、根据所述客户分类信息对所述历史业务办理记录中的业务办理信息进行分类，以得到业务办理分类信息。
根据所述客户分类信息对所述历史业务办理记录的业务办理信息进行分类，以得到业务办理分类信息。具体的，根据客户分类信息中每一客户类别包含的客户获取历史业务办理记录中相应客户的业务办理信息，得到业务办理分类信息，则业务办理分类信息中包含每一客户类别对应的业务办理信息。
S122、根据所述业务场景流程中所包含的流程节点，获取所述业务办理分类信息中每一客户类别的业务办理信息分别与每一所述流程节点相匹配的问答信息，作为所述分类样本数据信息。
根据业务场景流程中每一业务场景包含的流程节点，获取每一客户类别的业务办理信息分别与每一流程节点相匹配的问答信息，得到分类样本数据信息，则分类样本数据信息中的问答信息被分类至多个业务场景对应的流程节点，且每一流程节点包含多个客户类别对应的问答信息。
在一实施例中，如图5所示，步骤S122之后还包括步骤S123。
S123、根据预置的去重规则对每一流程节点中同一客户类别的问答信息进行去重处理，将每一流程节点的去重处理后的问答信息作为所述分类样本数据信息。
根据预置的去重规则对每一流程节点中同一客户类别的问答信息进行去重处理，得到每一流程节点的去重处理后的问答信息。一个流程节点中同一客户类别的多个问答信息可能存在重复，则需要通过去重规则对每一流程节点中同一客户类别的问答信息进行去重处理，去重规则即为可对问答信息进行去重处理的具体规则。具体的，可根据去重规则判断同一客户类别的问答信息中的两个提问信息或两个答复信息是否重复，若重复则选择其中一个进行保留；若不重复则两个均进行保留。
例如，获取两个提问信息或两个答复信息对应的两段文字信息，可将去重规则设置为判断两段文字信息的字符重复率是否不小于90％，若两段文字信息中字符的重复率不小于90％，则判断两段文字信息重复，否则判断两段文字信息不重复。
S130、若接收到来自所述用户终端的培训请求信息，获取所述分类样本数据信息中与所述培训请求信息相匹配的目标培训信息。
若接收到来自所述用户终端的培训请求信息，获取所述样本数据信息中与所述培训请求信息相匹配的目标培训信息。具体的，服务人员可通过用户终端向管理服务器发送培训请求信息，培训请求信息中包含业务场景信息，则服务人员可通过发送培训请求信息对某一种业务进行办理的具体场景所对应的问题进行针对性训练，可从样本数据信息中获取与培训请求信息相匹配的目标培训信息，目标培训信息中包含与业务场景信息中多个流程节点对应的多条问答信息，目标培训信息的多条问答信息均属于同一客户类别。
在一实施例中，步骤S130包括子步骤：随机选取所述分类样本数据信息中一个客户类别的问答信息作为目标类别问答信息；根据所述培训请求信息中的业务场景信息，获取所述目标类别问答信息中与所述业务场景信息的每一流程节点相匹配的一条问答信息进行组合得到与所述培训请求信息相匹配的目标培训信息。
分类样本数据信息中包含多个业务场景对应的流程节点，每一流程节点包含多个客户类别对应的问答信息，可选取一个客户类别作为目标客户类别，并获取每一流程节点中与该目标客户类别相匹配的问答信息作为目标类别问答信息。
培训请求信息中包括业务场景信息，根据培训请求信息中的业务场景信息，获取目标类别问答信息中与业务场景信息的每一流程节点相匹配的问答信息，并从与每一流程节点相匹配的多条问答信息中随机选择一条问答信息进行组合，即可得到与培训请求信息相匹配的目标培训信息。
S140、根据虚拟视频生成模型生成与所述目标培训信息中提问信息对应的虚拟提问视频发送至所述用户终端，以获取所述用户终端所反馈的答复视频信息。
根据虚拟视频生成模型生成与所述目标培训信息中提问信息对应的虚拟提问视频发送至所述用户终端，以获取所述用户终端所反馈的答复视频信息，其中，所述答复视频信息包含至少一段答复视频。为提高对服务人员进行培训的效果，可通过虚拟视频生成模型生成相应的虚拟提问视频，虚拟提问视频中包含与每一流程节点对应的一段虚拟视频，每一段虚拟视频均包含与一个流程节点对应的一条提问信息中的内容，服务人员通过用户终端接收虚拟提问视频，并针对虚拟提问视频中的每一段虚拟视频分别进行视频答复得到相应答复视频信息并反馈至管理服务器。
在一实施例中，如图6所示，步骤S140包括子步骤S141、S142和S143。
S141、获取所述虚拟视频生成模型中与所述目标培训信息的客户类别相匹配的一个类别模板。
虚拟视频生成模型中包含多个类别模板，每一类别模板与一个客户类别相匹配，则可获取虚拟视频生成模型中与目标培训信息的客户类别相匹配的一个类别模板。
S142、根据所述类别模板中的语音词典对所述目标培训信息所包含的提问信息进行转换以得到对应的语音提问信息。
类别模板中包含语音词典，可通过语音词典将目标培训信息中以文字形式记载的提问信息转换为语音提问信息，语音提问信息中包含与每一提问信息对应的语音信息，也即是一个提问信息对应一段语音信息。具体的，语音词典中包含多个词组及多个单词，每一词组对应一个词组发音，每一单词对应一个单词发音，获取目标培训信息中与一个流程节点对应的提问信息，对该提问信息中的字符与语音词典中的词组进行匹配，并获取相应词组的词组发音，对该提问信息中所剩余的不与词组相匹配的字符，则与语音词典中的单词进行匹配，并获取相应单词的单词发音，根据该提问信息中字符的顺序对词组发音与单词发音进行组合即可得到与该提问信息对应的一段语音信息。
例如，某一提问信息为“我想要了解这个重大疾病险”，语音词典中与“想要”对应的词组发音为“xiǎng,yào”，与“了解”对应的发音为“liǎo,jiě”，与“这个”对应的发音为“zhè,gè”，与“重大疾病”对应的发音为“zhòng,dà,jí,bìng”，与“我”对应的单词发音为“wǒ”，与“险”对应的单词发音为“xiǎn”，对上述发音进行组合即可得到对应的一段语音信息。
S143、将所述语音提问信息与所述类别模板中的人像模型进行组合以生成虚拟提问视频。
获取语音提问信息中每一段语音信息的语音长度，根据人像模型生成与语音长度对应的视频信息，人像模型即为与一个客户类别相匹配的虚拟人像，将语音提问信息与视频信息进行组合即可得到与一个提问信息对应的一段虚拟视频，获取与每一段提问信息对应的虚拟视频作为虚拟提问视频。
在一实施例中，如图7所示，步骤S150之前包括步骤S1510。
S1510、根据多个业务场景的业务语料数据库及预置的训练规则分别对初始化的神经网络进行语言环境训练以得到所述神经网络集合，所述初始化的神经网络基于BERT网络构建得到。
具体的，一个业务场景的业务语料数据库可对初始化的神经网络进行训练得到一个与该业务场景相匹配的神经网络，多个业务语料数据库分别对初始化的神经网络进行训练即可得到包含多个神经网络的神经网络集合，神经网络集合中每一神经网络均与一个业务场景相匹配。训练规则包括比例值、损失函数计算公式及梯度计算公式，训练规则即为对初始化的神经网络进行训练的规则信息，基于一个业务场景的业务语料数据库对初始化的神经网络进行 训练的过程包括步骤(1)至(6)：
(1)从一个所述业务语料数据库中随机选择与所述比例值对应的部分语料数据作为目标语料数据。
业务语料数据库中的每一条语料数据均是一个完整的句子，则每一语料数据均由多个字符组成，训练规则中还设置有比例值，可根据比例值从一个业务语料数据库中随机选择得到相应数量的语料数据作为目标语料数据，例如，比例值可设置为10-90％。
(2)对所述目标语料数据进行随机遮盖处理，得到语料处理数据。
每一语料均由多个字符组成，可对每一语料中的任意一个字符进行遮盖处理，得到包含遮盖字符的语料处理数据。
例如，某一目标语料数据为“想了解一下财产保险”，进行随机遮盖处理后得到的语料处理数据为“想X解一下财产保险”，其中“X”即表示被遮盖的字符。
具体的，步骤(2)之中还可以包括：对所述语料处理数据中与所述比例值对应的部分语料处理数据中被遮盖的字符进行随机替换。为增强预训练效果，还可根据上述比例值将部分语料处理数据中被遮盖的字符随机替换为其他字符。
例如，某一语料处理数据为“想X解一下财产保险”，进行随机替换后得到的语料处理数据为“想理解一下财产保险”。
(3)根据所述转换词典分别对所述目标语料数据及所述语料处理数据进行转换得到语料特征向量及语料处理特征向量。
每一字符均可在转换词典中匹配到对应的一个特征码，则可根据转换词典将目标语料数据中所包含的字符进行转换，得到大小为(1，M)的语料特征向量，其表示语料特征向量为1行M列，目标语料数据的特征码作为数值填充语料特征向量，其中未被填充的数值记为“0”。采用同样转换方式对与目标语料数据相对应的一条语料处理数据进行转换，得到语料处理特征向量。
例如，“想”在转换词典中对应的特征码为“2318”；“理”的特征码为“6522”，“解”的特征码为“7351”，“一”的特征码为“0100”，“下”的特征码为“8631”，“财”的特征码为“3621”、“产”的特征码为“1531”、“保”的特征码为“4280”、“险”的特征码为“6634”。“101”代表句子的开始特征码，“102”代表句子的结束特征码。则“想理解一下财产保险”的对应组合得到语料处理特征向量可表示为[101,2318,6522,7351,0100,8631,3621,1531,4280,6634,102,……，0]。
(4)将一个所述语料特征向量与对应的一个所述语料处理特征向量输入所述初始化的神经网络进行计算分别得到语料数组及语料处理数组。
初始化的神经网络由一个输入层、多个中间层及一个输出层组成，输入层与中间层之间、中间层与其他中间层之间、中间层与输出层之间均通过关联公式进行关联，例如某一关联公式可表示为y＝r×x+t，r和t即为该关联公式中的参数值。输入层中包含的输入节点的数量与第一特征向量的长度相对应，则第一特征向量中每一向量值与一个输入节点相对应，将第一特征向量输入初始化的神经网络进行计算，即可从其输出层获取输出结果，输出结果采用一个数组(M,N)进行表示，与语料特征向量对应的输出结果为语料数组，则语料数组的大小为M行N列。采用同样方式将语料处理特征向量输入识别模板进行计算，得到语料处理数组，其大小也为(M,N)，其中语料数组及语料处理数组中每一数值均属于[0，1]这一取值范围。
(5)根据所述训练规则中的损失函数计算公式计算所述语料数组与所述语料处理数组之 间的损失值。
损失值可用于对语料数组与语料处理数组之间的差别进行量化表示，具体的，计算语料数组S 1与语料处理数组S 2之间的损失值可通过损失函数计算公式计算得到
Figure PCTCN2021091287-appb-000006
其中，Ls为计算得到的损失值，a xy为语料数组S 1中第x行第y列的数值，b xy为语料处理数组S 2中第x行第y列的数值，M为语料数组S 1的总行数，N为语料数组S 1的总列数。
(6)根据所述训练规则中的梯度计算公式、所述损失值及所述识别模板的计算值计算得到所述初始化的神经网络中相应参数的更新值以更新所述参数的参数值。
根据所述训练规则中的梯度计算公式、所述损失值及所述初始化的神经网络的计算值计算得到所述识别模板的初始化的神经网络中每一参数的更新值以更新所述参数的参数值。具体的，将初始化的神经网络中一个参数对语料特征向量进行计算所得到的计算值输入梯度计算公式，并结合上述损失值，即可计算得到与该参数对应的更新值，这一计算过程也即为梯度下降计算，对参数值进行更新的过程与上述过程相同，在此不作赘述。
S150、根据预置的评分模型及所述目标培训信息获取与所述答复视频信息对应的答复评分信息并发送至所述用户终端。
根据预置的评分模型及所述目标培训信息获取与所述答复视频信息对应的答复评分信息并发送至所述用户终端，其中，所述评分模型包括语音转换规则、节点转化关键字、特征词典、神经网络集合及评分计算公式，所述神经网络集合中的神经网络均是基于BERT(Bidirectional Encoder Representations from Transformers)网络构建得到。可根据评分模型对答复视频信息进行分析，以得到答复评分信息并发送至用户终端，服务人员可通过用户终端获取相应的答复评分以获取培训的结果。答复视频信息中包含多段答复视频，每一段答复视频均与业务场景信息中的一个流程节点相匹配，可分别获取答复视频信息中多段答复视频的评分值作为答复评分信息。
在一实施例中，如图8所示，步骤S150包括子步骤S151、S152、S153、S154、S155和S156。
S151、根据所述语音转换规则对所述答复视频信息中的一段答复视频进行转换以得到对应的文字信息。
具体的，语音转换规则中包含声学模型、拼音词典及语义解析模型。一段答复视频中所包含的语音信息由多个字符发音的音素而组成，一个字符的音素包括该字符发音的频率和音色。声学模型中包含所有字符发音的音素，通过将语音信息与声学模型中所有的音素进行匹配，即可对语音信息中单个字符的音素进行切分，通过切分最终得到答复视频中所包含的多个音素。拼音词典中包含所有字符拼音对应的音素信息，通过将所得到的音素与字符拼音对应的音素信息进行匹配，即可将单个字符的音素转换为拼音词典中与该音素相匹配的字符拼音，以实现将答复视频的语音信息中所包含的所有音素转换为拼音信息。语义解析模型中包含拼音信息与文字信息之间所对应的映射关系，通过语义解析模型中所包含的映射关系即可对所得到的拼音信息进行语义解析以将拼音信息转换为对应的文字信息，所得到的文字信息即为与答复视频对应的一段文字信息。
例如，拼音“bàn，lǐ”在语义解析模型中所对应的文字信息为“办理”。
S152、判断所述文字信息是否与节点转化关键字中相应流程节点的转化关键字相匹配以 得到关键字判断结果。
节点转化关键字中包含与每一流程节点对应的转换关键字，可获取节点转换关键字中与文字信息对应的流程节点的转换关键字，并判断文字信息是否与转化关键字相匹配，可判断文字信息中是否包含该转化关键字以得到文字信息是否与转化关键字相匹配的关键字判断结果。若文字信息与转化关键字相匹配，则表明该文字信息对应的答复视频符合相应客户类别的核心诉求，可触发业务场景流程中相应的转化节点并推进业务办理过程，触发相应转化节点后业务办理可从当前流程节点顺利流转至下一流程节点；若文字信息不与转化关键字相匹配，则表明该文字信息对应的答复视频不符合相应客户类别的核心诉求。
S153、根据所述特征词典分别对所述文字信息及所述流程节点的答复信息进行转换得到第一特征向量及第二特征向量。
转换词典即为对字符进行转换的词典，每一字符均可在转换词典中匹配到对应的一个特征码，则可根据转换词典将文字信息中所包含的字符进行转换，将每一字符对应的特征码进行组合得到第一特征向量，所得到的第一特征向量即为将文字信息的特征采用向量方式进行表示，第一特征向量的大小为(1，M)，其表示第一特征向量为1行M列，第一特征向量的长度M可由用户预先设定，如可设定第一特征向量及第二特征向量中数值的数量为30(M＝30)，文字信息的特征码作为数值填充第一特征向量，第一特征向量中未被填充的数值记为“0”。采用同样转换方式对与文字信息相匹配的流程节点的答复信息进行转换，得到第二特征向量。
S154、根据所述业务场景信息确定所述神经网络集合中与所述业务场景信息相匹配的一个目标神经网络。
根据所述业务场景信息确定所述神经网络集合中与所述业务场景信息相匹配的一个目标神经网络。神经网络集合中包含多个神经网络，其中的神经网络均是基于BERT网络构建得到，每一神经网络均与一个业务场景相匹配，也即是一个神经网络适用于与之相匹配的一个业务场景的特定语言环境。采用一个业务场景的语料数据对初始化的神经网络进行预训练，以得到一个适用于该业务场景的语言环境的神经网络，多个业务场景对应的神经网络组合成为所述神经网络集合。通过获取与业务场景信息相匹配的目标神经网络，可基于目标神经网络的语言环境对第一特征向量及第二特征向量进行评分，得到更准确的评分结果。
S155、将所述第一特征向量与所述第二特征向量输入所述目标神经网络进行计算分别得到第一数组及第二数组。
具体的，对第一特征向量及第二特征向量进行计算的过程与对语料特征向量或语料处理特征向量进行计算的过程相同，输出结果采用一个数组(M,N)进行表示，与第一特征向量对应的输出结果为第一数组，则第一数组的大小也为M行N列，第二特征向量的输出结果为第二数组，第二数组的大小也为(M,N)，其中第一数组及第二数组中每一数值均属于[0，1]这一取值范围。
S156、根据所述评分计算公式对所述关键字判断结果、所述第一数组及所述第二数组进行计算得到对应的一个评分值。
可通过评分计算公式对所得到的关键字判断结果、第一数组及第二数组进行计算，以获取与该答复视频对应的一个评分值，具体的，先将关键字判断结果转换为对应的一个系数值进行表示。例如，若关键字判断结果为文字信息与转化关键字相匹配，对应的系数值为1；若关键字判断结果为文字信息不与转化关键字相匹配，对应的系数值为0.3。具体的，第一数组为S a，第二数组为S b，根据上述计算公式计算得到第一数组与第二数组之间的损失值， 根据评分计算公式可表示为：P 0＝C 0×(1-L 0)；其中，P 0为计算得到的评分值，C 0为所得到的系数值，L 0为第一数组S a与第二数组S b之间的损失值。
本申请中的技术方法可应用于智慧政务/智慧城管/智慧社区/智慧安防/智慧物流/智慧医疗/智慧教育/智慧环保/智慧交通等包含对服务人员进行智能线上培训的应用场景中，从而推动智慧城市的建设。
在本申请实施例所提供的基于人工智能的培训方法中，根据统计得到的客户分类信息及业务场景流程对历史业务办理记录进行拆分以得到分类样本数据信息，从分类样本数据信息中获取与培训请求信息相匹配的目标培训信息并生成对应的虚拟提问视频，获取用户反馈的答复视频信息进行评分得到答复评分信息并发送至用户终端。通过上述方法，基于包含多个客户的历史业务办理记录获取包含大量问答信息的分类样本数据信息，根据评分模型及与培训请求信息相匹配的目标培训信息获取答复视频信息的答复评分信息，以提高线上培训的覆盖面并实现对答复视频信息进行准确评分，从而大幅提升进行线上培训的培训效率。
本申请实施例还提供一种基于人工智能的培训装置，该基于人工智能的培训装置用于执行前述基于人工智能的培训方法的任一实施例。具体地，请参阅图9，图9是本申请实施例提供的基于人工智能的培训装置的示意性框图。该基于人工智能的培训装置可配置于管理服务器10中。
如图9所示，基于人工智能的培训装置100包括客户分类信息获取单元110、分类样本数据信息获取单元120、目标培训信息获取单元130虚拟提问视频发送单元140和答复评分信息发送单元150。
客户分类信息获取单元110，用于若接收到管理员输入的历史业务办理记录，根据预置的客户信息量化模型及预置的客户分类模型对所述历史业务办理记录中的客户进行分类统计以获取客户分类信息。
在一实施例中，所述客户分类信息获取单元110包括子单元：客户量化信息获取单元、客户类别获取单元和客户统计单元。
客户量化信息获取单元，用于根据所述客户信息量化模型对所述历史业务办理记录中客户的个人信息进行量化得到与每一客户对应的客户量化信息；客户类别获取单元，用于将所述客户量化信息依次输入所述客户分类模型以获取与每一客户量化信息对应的客户类别；客户统计单元，用于对每一所述客户类别包含的客户进行统计得到客户分类信息。
在一实施例中，所述客户分类信息获取单元110还包括子单元：客户分类模型训练单元。
客户分类模型训练单元，用于若接收到所输入的训练数据集，根据预置的梯度下降训练模型及所述训练数据集对所述客户分类模型进行迭代训练以得到训练后的客户分类模型。
分类样本数据信息获取单元120，用于根据预置的业务场景流程及所述客户分类信息对所述历史业务办理记录进行拆分以获取与所述客户分类信息相匹配的分类样本数据信息。
在一实施例中，所述分类样本数据信息获取单元120包括子单元：业务办理分类信息获取单元和问答信息匹配单元。
业务办理分类信息获取单元，用于根据所述客户分类信息对所述历史业务办理记录中的业务办理信息进行分类，以得到业务办理分类信息；问答信息匹配单元，用于根据所述业务场景流程中所包含的流程节点，获取所述业务办理分类信息中每一客户类别的业务办理信息分别与每一所述流程节点相匹配的问答信息，作为所述分类样本数据信息。
在一实施例中，所述分类样本数据信息获取单元120还包括子单元：去重处理单元。
去重处理单元，用于根据预置的去重规则对每一流程节点中同一客户类别的问答信息进行去重处理，将每一流程节点的去重处理后的问答信息作为所述分类样本数据信息。
目标培训信息获取单元130，用于若接收到来自所述用户终端的培训请求信息，获取所述分类样本数据信息中与所述培训请求信息相匹配的目标培训信息。
虚拟提问视频发送单元140，用于根据虚拟视频生成模型生成与所述目标培训信息中提问信息对应的虚拟提问视频发送至所述用户终端，以获取所述用户终端所反馈的答复视频信息，其中，所述答复视频信息包含至少一段答复视频。
在一实施例中，所述虚拟提问视频发送单元140包括子单元：类别模板匹配单元和提问信息转换单元。
类别模板匹配单元，用于获取所述虚拟视频生成模型中与所述目标培训信息的客户类别相匹配的一个类别模板；提问信息转换单元，用于根据所述类别模板中的语音词典对所述目标培训信息所包含的提问信息进行转换以得到对应的语音提问信息。虚拟提问视频生成单元，用于将所述语音提问信息与所述类别模板中的人像模型进行组合以生成虚拟提问视频。
在一实施例中，所述基于人工智能的培训装置100还包括：神经网络集合获取单元。
神经网络集合获取单元，用于根据多个业务场景的业务语料数据库及预置的训练规则分别对初始化的神经网络进行语言环境训练以得到所述神经网络集合，所述初始化的神经网络基于BERT网络构建得到。
答复评分信息发送单元150，用于根据预置的评分模型及所述目标培训信息获取与所述答复视频信息对应的答复评分信息并发送至所述用户终端。
在一实施例中，所述答复评分信息发送单元150包括子单元：文字信息获取单元、关键字匹配单元、特征向量获取单元、目标神经网络获取单元、数组获取单元和评分计算单元。
文字信息获取单元，用于根据所述语音转换规则对所述答复视频信息中的一段答复视频进行转换以得到对应的文字信息；关键字匹配单元，用于判断所述文字信息是否与节点转化关键字中相应流程节点的转化关键字相匹配以得到关键字判断结果；特征向量获取单元，用于根据所述特征词典分别对所述文字信息及所述流程节点的答复信息进行转换得到第一特征向量及第二特征向量；目标神经网络获取单元，用于根据所述业务场景信息确定所述神经网络集合中与所述业务场景信息相匹配的一个目标神经网络；数组获取单元，用于将所述第一特征向量与所述第二特征向量输入所述目标神经网络进行计算分别得到第一数组及第二数组；评分计算单元，用于根据所述评分计算公式对所述关键字判断结果、所述第一数组及所述第二数组进行计算得到对应的一个评分值。
在本申请实施例所提供的基于人工智能的培训装置应用上述基于人工智能的培训方法，根据统计得到的客户分类信息及业务场景流程对历史业务办理记录进行拆分以得到分类样本数据信息，从分类样本数据信息中获取与培训请求信息相匹配的目标培训信息并生成对应的虚拟提问视频，获取用户反馈的答复视频信息进行评分得到答复评分信息并发送至用户终端。通过上述方法，基于包含多个客户的历史业务办理记录获取包含大量问答信息的分类样本数据信息，根据评分模型及与培训请求信息相匹配的目标培训信息获取答复视频信息的答复评分信息，以提高线上培训的覆盖面并实现对答复视频信息进行准确评分，从而大幅提升进行线上培训的培训效率。
上述基于人工智能的培训装置可以实现为计算机程序的形式，该计算机程序可以在如图10所示的计算机设备上运行。
请参阅图10，图10是本申请实施例提供的计算机设备的示意性框图。该计算机设备可以是用于执行基于人工智能的培训方法以对服务人员进行智能线上培训的管理服务器10。
参阅图10，该计算机设备500包括通过系统总线501连接的处理器502、存储器和网络接口505，其中，存储器可以包括非易失性存储介质503和内存储器504。
该非易失性存储介质503可存储操作系统5031和计算机程序5032。该计算机程序5032被执行时，可使得处理器502执行基于人工智能的培训方法。
该处理器502用于提供计算和控制能力，支撑整个计算机设备500的运行。
该内存储器504为非易失性存储介质503中的计算机程序5032的运行提供环境，该计算机程序5032被处理器502执行时，可使得处理器502执行基于人工智能的培训方法。
该网络接口505用于进行网络通信，如提供数据信息的传输等。本领域技术人员可以理解，图10中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备500的限定，具体的计算机设备500可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
其中，所述处理器502用于运行存储在存储器中的计算机程序5032，以实现上述的基于人工智能的培训方法中对应的功能。
本领域技术人员可以理解，图10中示出的计算机设备的实施例并不构成对计算机设备具体构成的限定，在其他实施例中，计算机设备可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。例如，在一些实施例中，计算机设备可以仅包括存储器及处理器，在这样的实施例中，存储器及处理器的结构及功能与图10所示实施例一致，在此不再赘述。
应当理解，在本申请实施例中，处理器502可以是中央处理单元(Central Processing Unit，CPU)，该处理器502还可以是其他通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现成可编程门阵列(Field-Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。其中，通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
在本申请的另一实施例中提供计算机可读存储介质。该计算机可读存储介质可以为非易失性的计算机可读存储介质。该计算机可读存储介质存储有计算机程序，其中计算机程序被处理器执行时实现上述的基于人工智能的培训方法中所包含的步骤。
所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，上述描述的设备、装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。本领域普通技术人员可以意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、计算机软件或者二者的结合来实现，为了清楚地说明硬件和软件的可互换性，在上述说明中已经按照功能一般性地描述了各示例的组成及步骤。这些功能究竟以硬件还是软件方式来执行取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本申请的范围。
在本申请所提供的几个实施例中，应该理解到，所揭露的设备、装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为逻辑功能划分，实际实现时可以有另外的划分方式，也可以将具有相同功能的单元集合成一个单元，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以 忽略，或不执行。另外，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口、装置或单元的间接耦合或通信连接，也可以是电的，机械的或其它的形式连接。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本申请实施例方案的目的。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以是两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读存储介质中。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分，或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个计算机可读存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的计算机可读存储介质包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述，仅为本申请的具体实施方式，但本申请的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本申请揭露的技术范围内，可轻易想到各种等效的修改或替换，这些修改或替换都应涵盖在本申请的保护范围之内。因此，本申请的保护范围应以权利要求的保护范围为准。

Claims (20)

  1. 一种基于人工智能的培训方法，应用于管理服务器，所述管理服务器与至少一台用户终端通过网络连接实现数据信息的传输，其中，所述方法包括：
    若接收到管理员输入的历史业务办理记录，根据预置的客户信息量化模型及预置的客户分类模型对所述历史业务办理记录中的客户进行分类统计以获取客户分类信息；
    根据预置的业务场景流程及所述客户分类信息对所述历史业务办理记录进行拆分以获取与所述客户分类信息相匹配的分类样本数据信息；
    若接收到来自所述用户终端的培训请求信息，获取所述分类样本数据信息中与所述培训请求信息相匹配的目标培训信息；
    根据虚拟视频生成模型生成与所述目标培训信息中提问信息对应的虚拟提问视频发送至所述用户终端，以获取所述用户终端所反馈的答复视频信息，其中，所述答复视频信息包含至少一段答复视频；
    根据预置的评分模型及所述目标培训信息获取与所述答复视频信息对应的答复评分信息并发送至所述用户终端。
  2. 根据权利要求1所述的基于人工智能的培训方法，其中，所述根据预置的客户信息量化模型及预置的客户分类模型对所述历史业务办理记录中的客户进行分类统计以获取客户分类信息，包括：
    根据所述客户信息量化模型对所述历史业务办理记录中客户的个人信息进行量化得到与每一客户对应的客户量化信息；
    将所述客户量化信息依次输入所述客户分类模型以获取与每一客户量化信息对应的客户类别；
    对每一所述客户类别包含的客户进行统计得到客户分类信息。
  3. 根据权利要求1所述的基于人工智能的培训方法，其中，所述根据预置的业务场景流程及所述客户分类信息对所述历史业务办理记录进行拆分以获取与所述客户分类信息相匹配的分类样本数据信息，包括：
    根据所述客户分类信息对所述历史业务办理记录中的业务办理信息进行分类，以得到业务办理分类信息；
    根据所述业务场景流程中所包含的流程节点，获取所述业务办理分类信息中每一客户类别的业务办理信息分别与每一所述流程节点相匹配的问答信息，作为所述分类样本数据信息。
  4. 根据权利要求3所述的基于人工智能的培训方法，其中，所述获取所述业务办理分类信息中每一客户类别的业务办理信息分别与每一所述流程节点相匹配的问答信息之后，还包括：
    根据预置的去重规则对每一流程节点中同一客户类别的问答信息进行去重处理，将每一流程节点的去重处理后的问答信息作为所述分类样本数据信息。
  5. 根据权利要求1所述的基于人工智能的培训方法，其中，所述根据预置的虚拟视频生成模型生成与所述目标培训信息中提问信息对应的虚拟提问视频发送至所述用户终端，包括：
    获取所述虚拟视频生成模型中与所述目标培训信息的客户类别相匹配的一个类别模板；
    根据所述类别模板中的语音词典对所述目标培训信息所包含的提问信息进行转换以得到对应的语音提问信息；
    将所述语音提问信息与所述类别模板中的人像模型进行组合以生成虚拟提问视频。
  6. 根据权利要求1所述的基于人工智能的培训方法，其中，所述评分模型包括语音转换 规则、节点转化关键字、特征词典、神经网络集合及评分计算公式，所述答复评分信息包含每一段答复视频的评分值，所述根据预置的评分模型及所述目标培训信息获取与所述答复视频信息对应的答复评分信息并发送至所述用户终端，包括：
    根据所述语音转换规则对所述答复视频信息中的一段答复视频进行转换以得到对应的文字信息；
    判断所述文字信息是否与节点转化关键字中相应流程节点的转化关键字相匹配以得到关键字判断结果；
    根据所述特征词典分别对所述文字信息及所述流程节点的答复信息进行转换得到第一特征向量及第二特征向量；
    根据所述业务场景信息确定所述神经网络集合中与所述业务场景信息相匹配的一个目标神经网络；
    将所述第一特征向量与所述第二特征向量输入所述目标神经网络进行计算分别得到第一数组及第二数组；
    根据所述评分计算公式对所述关键字判断结果、所述第一数组及所述第二数组进行计算得到对应的一个评分值。
  7. 根据权利要求6所述的基于人工智能的培训方法，其中，所述根据预置的评分模型及所述目标培训信息获取与所述答复视频信息对应的答复评分信息并发送至所述用户终端之前，还包括：
    根据多个业务场景的业务语料数据库及预置的训练规则分别对初始化的神经网络进行语言环境训练以得到所述神经网络集合，所述初始化的神经网络基于BERT网络构建得到。
  8. 一种基于人工智能的培训装置，包括：
    客户分类信息获取单元，用于若接收到管理员输入的历史业务办理记录，根据预置的客户信息量化模型及预置的客户分类模型对所述历史业务办理记录中的客户进行分类统计以获取客户分类信息；
    分类样本数据信息获取单元，用于根据预置的业务场景流程及所述客户分类信息对所述历史业务办理记录进行拆分以获取与所述客户分类信息相匹配的分类样本数据信息；
    目标培训信息获取单元，用于若接收到来自所述用户终端的培训请求信息，获取所述分类样本数据信息中与所述培训请求信息相匹配的目标培训信息；
    虚拟提问视频发送单元，用于根据虚拟视频生成模型生成与所述目标培训信息中提问信息对应的虚拟提问视频发送至所述用户终端，以获取所述用户终端所反馈的答复视频信息，其中，所述答复视频信息包含至少一段答复视频；
    答复评分信息发送单元，用于根据预置的评分模型及所述目标培训信息获取与所述答复视频信息对应的答复评分信息并发送至所述用户终端。
  9. 一种计算机设备，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，其中，所述处理器执行所述计算机程序时实现以下步骤：
    若接收到管理员输入的历史业务办理记录，根据预置的客户信息量化模型及预置的客户分类模型对所述历史业务办理记录中的客户进行分类统计以获取客户分类信息；
    根据预置的业务场景流程及所述客户分类信息对所述历史业务办理记录进行拆分以获取与所述客户分类信息相匹配的分类样本数据信息；
    若接收到来自所述用户终端的培训请求信息，获取所述分类样本数据信息中与所述培训 请求信息相匹配的目标培训信息；
    根据虚拟视频生成模型生成与所述目标培训信息中提问信息对应的虚拟提问视频发送至所述用户终端，以获取所述用户终端所反馈的答复视频信息，其中，所述答复视频信息包含至少一段答复视频；
    根据预置的评分模型及所述目标培训信息获取与所述答复视频信息对应的答复评分信息并发送至所述用户终端。
  10. 根据权利要求9所述的计算机设备，其中，所述根据预置的客户信息量化模型及预置的客户分类模型对所述历史业务办理记录中的客户进行分类统计以获取客户分类信息，包括：
    根据所述客户信息量化模型对所述历史业务办理记录中客户的个人信息进行量化得到与每一客户对应的客户量化信息；
    将所述客户量化信息依次输入所述客户分类模型以获取与每一客户量化信息对应的客户类别；
    对每一所述客户类别包含的客户进行统计得到客户分类信息。
  11. 根据权利要求9所述的计算机设备，其中，所述根据预置的业务场景流程及所述客户分类信息对所述历史业务办理记录进行拆分以获取与所述客户分类信息相匹配的分类样本数据信息，包括：
    根据所述客户分类信息对所述历史业务办理记录中的业务办理信息进行分类，以得到业务办理分类信息；
    根据所述业务场景流程中所包含的流程节点，获取所述业务办理分类信息中每一客户类别的业务办理信息分别与每一所述流程节点相匹配的问答信息，作为所述分类样本数据信息。
  12. 根据权利要求11所述的计算机设备，其中，所述获取所述业务办理分类信息中每一客户类别的业务办理信息分别与每一所述流程节点相匹配的问答信息之后，还包括：
    根据预置的去重规则对每一流程节点中同一客户类别的问答信息进行去重处理，将每一流程节点的去重处理后的问答信息作为所述分类样本数据信息。
  13. 根据权利要求9所述的计算机设备，其中，所述根据预置的虚拟视频生成模型生成与所述目标培训信息中提问信息对应的虚拟提问视频发送至所述用户终端，包括：
    获取所述虚拟视频生成模型中与所述目标培训信息的客户类别相匹配的一个类别模板；
    根据所述类别模板中的语音词典对所述目标培训信息所包含的提问信息进行转换以得到对应的语音提问信息；
    将所述语音提问信息与所述类别模板中的人像模型进行组合以生成虚拟提问视频。
  14. 根据权利要求9所述的计算机设备，其中，所述评分模型包括语音转换规则、节点转化关键字、特征词典、神经网络集合及评分计算公式，所述答复评分信息包含每一段答复视频的评分值，所述根据预置的评分模型及所述目标培训信息获取与所述答复视频信息对应的答复评分信息并发送至所述用户终端，包括：
    根据所述语音转换规则对所述答复视频信息中的一段答复视频进行转换以得到对应的文字信息；
    判断所述文字信息是否与节点转化关键字中相应流程节点的转化关键字相匹配以得到关键字判断结果；
    根据所述特征词典分别对所述文字信息及所述流程节点的答复信息进行转换得到第一特 征向量及第二特征向量；
    根据所述业务场景信息确定所述神经网络集合中与所述业务场景信息相匹配的一个目标神经网络；
    将所述第一特征向量与所述第二特征向量输入所述目标神经网络进行计算分别得到第一数组及第二数组；
    根据所述评分计算公式对所述关键字判断结果、所述第一数组及所述第二数组进行计算得到对应的一个评分值。
  15. 根据权利要求14所述的计算机设备，其中，所述根据预置的评分模型及所述目标培训信息获取与所述答复视频信息对应的答复评分信息并发送至所述用户终端之前，还包括：
    根据多个业务场景的业务语料数据库及预置的训练规则分别对初始化的神经网络进行语言环境训练以得到所述神经网络集合，所述初始化的神经网络基于BERT网络构建得到。
  16. 一种计算机可读存储介质，其中，所述计算机可读存储介质存储有计算机程序，所述计算机程序当被处理器执行时使所述处理器执行以下操作：
    若接收到管理员输入的历史业务办理记录，根据预置的客户信息量化模型及预置的客户分类模型对所述历史业务办理记录中的客户进行分类统计以获取客户分类信息；
    根据预置的业务场景流程及所述客户分类信息对所述历史业务办理记录进行拆分以获取与所述客户分类信息相匹配的分类样本数据信息；
    若接收到来自所述用户终端的培训请求信息，获取所述分类样本数据信息中与所述培训请求信息相匹配的目标培训信息；
    根据虚拟视频生成模型生成与所述目标培训信息中提问信息对应的虚拟提问视频发送至所述用户终端，以获取所述用户终端所反馈的答复视频信息，其中，所述答复视频信息包含至少一段答复视频；
    根据预置的评分模型及所述目标培训信息获取与所述答复视频信息对应的答复评分信息并发送至所述用户终端。
  17. 根据权利要求16所述的计算机可读存储介质，其中，所述根据预置的客户信息量化模型及预置的客户分类模型对所述历史业务办理记录中的客户进行分类统计以获取客户分类信息，包括：
    根据所述客户信息量化模型对所述历史业务办理记录中客户的个人信息进行量化得到与每一客户对应的客户量化信息；
    将所述客户量化信息依次输入所述客户分类模型以获取与每一客户量化信息对应的客户类别；
    对每一所述客户类别包含的客户进行统计得到客户分类信息。
  18. 根据权利要求16所述的计算机可读存储介质，其中，所述根据预置的业务场景流程及所述客户分类信息对所述历史业务办理记录进行拆分以获取与所述客户分类信息相匹配的分类样本数据信息，包括：
    根据所述客户分类信息对所述历史业务办理记录中的业务办理信息进行分类，以得到业务办理分类信息；
    根据所述业务场景流程中所包含的流程节点，获取所述业务办理分类信息中每一客户类别的业务办理信息分别与每一所述流程节点相匹配的问答信息，作为所述分类样本数据信息。
  19. 根据权利要求18所述的计算机可读存储介质，其中，所述获取所述业务办理分类信 息中每一客户类别的业务办理信息分别与每一所述流程节点相匹配的问答信息之后，还包括：
    根据预置的去重规则对每一流程节点中同一客户类别的问答信息进行去重处理，将每一流程节点的去重处理后的问答信息作为所述分类样本数据信息。
  20. 根据权利要求16所述的计算机可读存储介质，其中，所述根据预置的虚拟视频生成模型生成与所述目标培训信息中提问信息对应的虚拟提问视频发送至所述用户终端，包括：
    获取所述虚拟视频生成模型中与所述目标培训信息的客户类别相匹配的一个类别模板；
    根据所述类别模板中的语音词典对所述目标培训信息所包含的提问信息进行转换以得到对应的语音提问信息；
    将所述语音提问信息与所述类别模板中的人像模型进行组合以生成虚拟提问视频。
PCT/CN2021/091287 2020-11-03 2021-04-30 基于人工智能的培训方法、装置、计算机设备及存储介质 WO2022095378A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011209903.2 2020-11-03
CN202011209903.2A CN112328742B (zh) 2020-11-03 2020-11-03 基于人工智能的培训方法、装置、计算机设备及存储介质

Publications (1)

Publication Number Publication Date
WO2022095378A1 true WO2022095378A1 (zh) 2022-05-12

Family

ID=74323110

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/091287 WO2022095378A1 (zh) 2020-11-03 2021-04-30 基于人工智能的培训方法、装置、计算机设备及存储介质

Country Status (2)

Country Link
CN (1) CN112328742B (zh)
WO (1) WO2022095378A1 (zh)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115223366A (zh) * 2022-07-20 2022-10-21 重庆城市综合交通枢纽（集团）有限公司 一种基于网络信息覆盖的诱导屏布点方法
CN115391414A (zh) * 2022-10-28 2022-11-25 北京双赢天下管理咨询有限公司 一种基于大数据的银行市场拓展系统及方法
CN115809837A (zh) * 2023-02-09 2023-03-17 恒丰银行股份有限公司 基于数字化模拟场景的金融企业管理方法、设备及介质
CN116303982A (zh) * 2023-05-24 2023-06-23 杭州实在智能科技有限公司 基于rpa和自学习机制的智能应答与业务处理方法及系统
CN116402399A (zh) * 2023-04-14 2023-07-07 北京智慧大王科技有限公司 基于人工智能与电子商城的业务数据处理方法及系统
CN116401355A (zh) * 2023-06-08 2023-07-07 浙江天正思维信息技术有限公司 一种基于数智交互的咨询业务决策管理方法及系统
CN116431246A (zh) * 2023-06-13 2023-07-14 建信金融科技有限责任公司 办公自动化软件配置方法及装置
CN117828320A (zh) * 2024-03-05 2024-04-05 元创者(厦门)数字科技有限公司 一种虚拟数字人构建方法及其系统

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112328742B (zh) * 2020-11-03 2023-08-18 平安科技（深圳）有限公司 基于人工智能的培训方法、装置、计算机设备及存储介质
CN113821619A (zh) * 2021-08-31 2021-12-21 前海人寿保险股份有限公司 培训方法、装置、系统与计算机可读存储介质
CN113505606B (zh) * 2021-09-10 2021-12-28 北京明略昭辉科技有限公司 一种培训信息获取方法、装置、电子设备及存储介质
CN114401431B (zh) * 2022-01-19 2024-04-09 中国平安人寿保险股份有限公司 一种虚拟人讲解视频生成方法及相关装置
CN114529267A (zh) * 2022-02-24 2022-05-24 宁波中联人力资源开发有限公司 基于虚拟现实的人才管理方法及装置
CN114997986A (zh) * 2022-05-27 2022-09-02 湖南三湘银行股份有限公司 一种基于大数据的对公生态系统业务优化方法及装置

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105184520A (zh) * 2015-10-22 2015-12-23 成都往来教育科技有限公司 教师职业能力测评方法及装置
US20170039502A1 (en) * 2013-06-28 2017-02-09 Healthtap, Inc. Systems and methods for evaluating and selecting a healthcare professional using a healthcare operating system
CN107909338A (zh) * 2017-10-30 2018-04-13 平安科技（深圳）有限公司 培训管理方法、 装置、 计算机设备及存储介质
CN110458732A (zh) * 2019-06-17 2019-11-15 深圳追一科技有限公司 培训方法、装置、计算机设备和存储介质
CN111340148A (zh) * 2020-05-22 2020-06-26 支付宝(杭州)信息技术有限公司 一种业务分类模型的训练方法、业务分类方法以及终端
CN111553555A (zh) * 2020-03-27 2020-08-18 深圳追一科技有限公司 培训方法、装置、计算机设备和存储介质
CN112328742A (zh) * 2020-11-03 2021-02-05 平安科技（深圳）有限公司 基于人工智能的培训方法、装置、计算机设备及存储介质

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2002196659A (ja) * 2001-10-01 2002-07-12 Fujitsu Ltd 接客教育システム
CN101517512A (zh) * 2006-07-24 2009-08-26 查查检索有限公司 用于信息搜索系统中的网络广播和视频培训的方法、系统和计算机可读存储器
DE102013007502A1 (de) * 2013-04-25 2014-10-30 Elektrobit Automotive Gmbh Computer-implementiertes Verfahren zum automatischen Trainieren eins Dialogsystems und Dialogsystem zur Erzeugung von semantischen Annotationen
CN108734379B (zh) * 2018-04-03 2022-02-11 四川新网银行股份有限公司 一种对客服人员实现差异化的线上培训方法
CN110580282B (zh) * 2018-05-22 2023-06-13 阿里巴巴集团控股有限公司 通过仿真用户与客服交互的方法及装置
CN110472982A (zh) * 2019-08-01 2019-11-19 深圳市遇住信息科技有限公司 一种客服智能管理系统及方法
CN111046297A (zh) * 2020-03-12 2020-04-21 深圳市成功快车科技有限公司 一种基于机器学习算法的服务智能匹配推荐方法、装置、设备及存储介质

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170039502A1 (en) * 2013-06-28 2017-02-09 Healthtap, Inc. Systems and methods for evaluating and selecting a healthcare professional using a healthcare operating system
CN105184520A (zh) * 2015-10-22 2015-12-23 成都往来教育科技有限公司 教师职业能力测评方法及装置
CN107909338A (zh) * 2017-10-30 2018-04-13 平安科技（深圳）有限公司 培训管理方法、 装置、 计算机设备及存储介质
CN110458732A (zh) * 2019-06-17 2019-11-15 深圳追一科技有限公司 培训方法、装置、计算机设备和存储介质
CN111553555A (zh) * 2020-03-27 2020-08-18 深圳追一科技有限公司 培训方法、装置、计算机设备和存储介质
CN111340148A (zh) * 2020-05-22 2020-06-26 支付宝(杭州)信息技术有限公司 一种业务分类模型的训练方法、业务分类方法以及终端
CN112328742A (zh) * 2020-11-03 2021-02-05 平安科技（深圳）有限公司 基于人工智能的培训方法、装置、计算机设备及存储介质

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115223366A (zh) * 2022-07-20 2022-10-21 重庆城市综合交通枢纽（集团）有限公司 一种基于网络信息覆盖的诱导屏布点方法
CN115223366B (zh) * 2022-07-20 2023-08-29 重庆城市综合交通枢纽（集团）有限公司 一种基于网络信息覆盖的诱导屏布点方法
CN115391414A (zh) * 2022-10-28 2022-11-25 北京双赢天下管理咨询有限公司 一种基于大数据的银行市场拓展系统及方法
CN115391414B (zh) * 2022-10-28 2023-01-13 北京双赢天下管理咨询有限公司 一种基于大数据的银行市场拓展系统及方法
CN115809837B (zh) * 2023-02-09 2023-05-30 恒丰银行股份有限公司 基于数字化模拟场景的金融企业管理方法、设备及介质
CN115809837A (zh) * 2023-02-09 2023-03-17 恒丰银行股份有限公司 基于数字化模拟场景的金融企业管理方法、设备及介质
CN116402399A (zh) * 2023-04-14 2023-07-07 北京智慧大王科技有限公司 基于人工智能与电子商城的业务数据处理方法及系统
CN116402399B (zh) * 2023-04-14 2023-12-29 上海锦咏数据科技有限公司 基于人工智能与电子商城的业务数据处理方法及系统
CN116303982A (zh) * 2023-05-24 2023-06-23 杭州实在智能科技有限公司 基于rpa和自学习机制的智能应答与业务处理方法及系统
CN116303982B (zh) * 2023-05-24 2023-08-22 杭州实在智能科技有限公司 基于rpa和自学习机制的智能应答与业务处理方法及系统
CN116401355A (zh) * 2023-06-08 2023-07-07 浙江天正思维信息技术有限公司 一种基于数智交互的咨询业务决策管理方法及系统
CN116401355B (zh) * 2023-06-08 2023-08-15 浙江天正思维信息技术有限公司 一种基于数智交互的咨询业务决策管理方法及系统
CN116431246A (zh) * 2023-06-13 2023-07-14 建信金融科技有限责任公司 办公自动化软件配置方法及装置
CN116431246B (zh) * 2023-06-13 2023-08-11 建信金融科技有限责任公司 办公自动化软件配置方法及装置
CN117828320A (zh) * 2024-03-05 2024-04-05 元创者(厦门)数字科技有限公司 一种虚拟数字人构建方法及其系统
CN117828320B (zh) * 2024-03-05 2024-05-07 元创者(厦门)数字科技有限公司 一种虚拟数字人构建方法及其系统

Also Published As

Publication number Publication date
CN112328742A (zh) 2021-02-05
CN112328742B (zh) 2023-08-18

Similar Documents

Publication Publication Date Title
WO2022095378A1 (zh) 基于人工智能的培训方法、装置、计算机设备及存储介质
US10824949B2 (en) Method and system for extracting information from graphs
WO2022134071A1 (zh) 文本抽取方法及装置、计算机可读存储介质以及电子设备
US11676067B2 (en) System and method for creating data to train a conversational bot
WO2022048173A1 (zh) 基于人工智能的客户意图识别方法、装置、设备及介质
CN107679234B (zh) 客服信息提供方法、装置、电子设备、存储介质
US10599686B1 (en) Method and system for extracting information from graphs
WO2019153737A1 (zh) 用于对评论进行评估的方法、装置、设备和存储介质
WO2020147428A1 (zh) 交互内容生成方法、装置、计算机设备及存储介质
WO2020244073A1 (zh) 基于语音的用户分类方法、装置、计算机设备及存储介质
WO2017092380A1 (zh) 用于人机对话的方法、神经网络系统和用户设备
WO2022252636A1 (zh) 基于人工智能的回答生成方法、装置、设备及存储介质
CN110727779A (zh) 基于多模型融合的问答方法及系统
JP6832501B2 (ja) 意味生成方法、意味生成装置及びプログラム
WO2021120779A1 (zh) 一种基于人机对话的用户画像构建方法、系统、终端及存储介质
WO2023040493A1 (zh) 事件检测
WO2022048194A1 (zh) 事件主体识别模型优化方法、装置、设备及可读存储介质
CN116304748B (zh) 一种文本相似度计算方法、系统、设备及介质
US20230359825A1 (en) Knowledge graph entities from text
CN116955561A (zh) 问答方法、装置、电子设备以及存储介质
KR20200041199A (ko) 챗봇 구동 방법, 장치 및 컴퓨터 판독가능 매체
CN117494815A (zh) 面向档案的可信大语言模型训练、推理方法和装置
CN117312641A (zh) 智能获取信息的方法、装置、设备及存储介质
CN118378148A (zh) 多标签分类模型的训练方法、多标签分类方法及相关装置
CN111680501B (zh) 基于深度学习的问询信息识别方法、装置及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21888082

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21888082

Country of ref document: EP

Kind code of ref document: A1