CN117834780B - Intelligent outbound customer intention prediction analysis system - Google Patents
Intelligent outbound customer intention prediction analysis system Download PDFInfo
- Publication number
- CN117834780B CN117834780B CN202410248851.1A CN202410248851A CN117834780B CN 117834780 B CN117834780 B CN 117834780B CN 202410248851 A CN202410248851 A CN 202410248851A CN 117834780 B CN117834780 B CN 117834780B
- Authority
- CN
- China
- Prior art keywords
- outbound
- module
- customer
- client
- analysis
- Prior art date
- Legal status (The legal status 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 status listed.)
- Active
Links
- 238000004458 analytical method Methods 0.000 title claims abstract description 88
- 230000008451 emotion Effects 0.000 claims abstract description 72
- 230000006399 behavior Effects 0.000 claims abstract description 57
- 230000001149 cognitive effect Effects 0.000 claims abstract description 37
- 230000006978 adaptation Effects 0.000 claims abstract description 23
- 238000005516 engineering process Methods 0.000 claims abstract description 13
- 238000006243 chemical reaction Methods 0.000 claims abstract description 10
- 239000013598 vector Substances 0.000 claims description 44
- 238000000034 method Methods 0.000 claims description 31
- 230000003993 interaction Effects 0.000 claims description 30
- 230000007704 transition Effects 0.000 claims description 28
- 230000008569 process Effects 0.000 claims description 26
- 238000003066 decision tree Methods 0.000 claims description 21
- 238000010586 diagram Methods 0.000 claims description 18
- 238000012549 training Methods 0.000 claims description 15
- 230000004044 response Effects 0.000 claims description 12
- 230000003542 behavioural effect Effects 0.000 claims description 11
- 230000007246 mechanism Effects 0.000 claims description 11
- 230000009471 action Effects 0.000 claims description 7
- 238000013515 script Methods 0.000 claims description 7
- 238000004422 calculation algorithm Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000007405 data analysis Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000004140 cleaning Methods 0.000 claims description 3
- 238000013480 data collection Methods 0.000 claims description 3
- 238000013500 data storage Methods 0.000 claims description 3
- 230000008034 disappearance Effects 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000004880 explosion Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 230000014509 gene expression Effects 0.000 claims description 3
- 238000010801 machine learning Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000035484 reaction time Effects 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000004891 communication Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000002996 emotional effect Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- IZMVGEWQSBMVIU-UHFFFAOYSA-N 2-(2-azaniumyl-1-hydroxycyclobutyl)acetate Chemical compound NC1CCC1(O)CC(O)=O IZMVGEWQSBMVIU-UHFFFAOYSA-N 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04M—TELEPHONIC COMMUNICATION
- H04M3/00—Automatic or semi-automatic exchanges
- H04M3/42—Systems providing special services or facilities to subscribers
- H04M3/487—Arrangements for providing information services, e.g. recorded voice services or time announcements
- H04M3/493—Interactive information services, e.g. directory enquiries ; Arrangements therefor, e.g. interactive voice response [IVR] systems or voice portals
- H04M3/4936—Speech interaction details
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/10—Text processing
- G06F40/166—Editing, e.g. inserting or deleting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
- G06F40/35—Discourse or dialogue representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/18—Speech classification or search using natural language modelling
- G10L15/1822—Parsing for meaning understanding
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/26—Speech to text systems
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
- G10L25/51—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination
- G10L25/63—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination for estimating an emotional state
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Acoustics & Sound (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Signal Processing (AREA)
- Databases & Information Systems (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Child & Adolescent Psychology (AREA)
- Hospice & Palliative Care (AREA)
- Psychiatry (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to the technical field of intelligent outbound, in particular to an intelligent outbound customer intention prediction analysis system, which comprises a customer voice input module, a client voice input module and a client voice analysis module, wherein the customer voice input module is used for receiving a voice signal of a customer; the voice-to-text module converts the received voice signal into text information; the text emotion analysis module analyzes emotion tendencies of the converted text; the intention recognition module predicts the intention of the client according to the text information and the emotion analysis result and by combining a preset intention recognition model; the database module stores historical outbound data, customer feedback information and an intention recognition model; the outbound strategy generation module generates a targeted outbound strategy; the outbound execution module automatically executes outbound tasks; the cognitive behavior analysis and adaptation module uses a cognitive modeling technology to identify a client behavior mode and predict future behaviors and reactions.
Description
Technical Field
The invention relates to the technical field of intelligent outbound, in particular to an intelligent outbound customer intention prediction analysis system.
Background
With the rapid development of information technology, an intelligent outbound system has become one of key tools in the field of customer relationship management, and particularly plays an increasingly important role in the fields of customer service, market research, product popularization and the like. The traditional outbound system mainly relies on preset scripts and subjective experience of operators to conduct customer communication, the mode is worry about handling complex customer demands and intentions, the actual intentions of customers cannot be accurately understood, and the outbound efficiency is low and the customer satisfaction is low.
In addition, with the increase in market competition and the diversification of customer needs, it has been difficult for a single communication strategy to meet the needs of personalized services. The client interaction data contains rich emotion and behavior information, and how to accurately extract the intention of the client from the information and formulate a corresponding outbound strategy becomes a key for improving the outbound success rate and the client satisfaction.
At this time, a system capable of automatically and intelligently processing customer data, accurately predicting customer intention, and generating a personalized outbound policy based on the customer intention is urgently needed. The system can integrate various advanced technologies such as voice recognition, natural language processing, emotion analysis and the like to realize deep understanding and accurate prediction of customer intention, thereby remarkably improving the execution efficiency and success rate of outbound tasks and meeting the urgent demands of the modern customer service field for intelligent and personalized services.
Disclosure of Invention
Based on the above purpose, the invention provides an intelligent outbound customer intention prediction analysis system.
An intelligent outbound customer intention prediction analysis system comprises a customer voice input module, a voice-to-text module, a text emotion analysis module, an intention recognition module, a database module, an outbound strategy generation module, an outbound execution module and a cognitive behavior analysis and adaptation module, wherein the client voice input module is used for receiving voice from a customer;
the client voice input module is used for receiving a voice signal of a client;
The voice-to-text module converts the received voice signal into text information;
the text emotion analysis module analyzes emotion tendencies of the converted text and judges the emotion state of the client;
the intention recognition module predicts the intention of the client according to the text information and the emotion analysis result and by combining a preset intention recognition model;
The database module stores historical outbound data, customer feedback information and an intention recognition model;
the outbound strategy generation module generates a targeted outbound strategy according to the intention recognition result and historical data in the database;
The outbound execution module automatically executes outbound tasks according to the generated outbound strategy and interacts with clients;
the cognitive behavior analysis and adaptation module is responsible for collecting and analyzing behavior data of clients in past and current interactions, identifying client behavior patterns by using cognitive modeling technology, and predicting future behaviors and reactions.
Further, the text emotion analysis module specifically includes:
Cleaning and standardizing text data through a preprocessing step, including removing stop words, correcting spelling, part-of-speech tagging and stem extraction, and then converting the text into a vector form by utilizing an embedding technology for machine learning model processing;
and modeling and analyzing the emotion tendencies in the text data by adopting a transducer model.
Further, the transducer model specifically includes:
Self-attention mechanism: the purpose is to calculate for each token in the input sequence its attention weight for all tokens in the sequence, given a representation X 1,X2,...,Xn of one sequence, for each token in the sequence X i, the self-attention mechanism maps it to a query vector Q, a key vector K and a value vector V, obtained by linear transformation:
Q=xw Q,K＝XWK,V＝XWV, where W Q、WK、WV is a matrix of learnable parameters;
the attention weight is calculated and applied to the value vector, the calculation method is as follows:
Wherein d k is the dimension of the key vector, and division operation is used for scaling the dot product to prevent gradient disappearance or explosion;
Multi-head attention: the transducer improves the model performance through a multi-head attention mechanism, and the self-attention process is performed multiple times in parallel, and different parameter matrixes W Q、WK、WV are used each time:
MultiHead (Q, K, V) = Concat (head 1,...,headh)WO, where each head i is a self-attention layer and W O is another learnable parameter matrix for combining the outputs of different heads;
position feed forward network: each encoder and decoder layer in the transducer model also contains a position feed forward network, the same fully connected layer is applied independently to the tokens at each position:
FFN (x) =max (0, xw 1+b1)W2+b2, where W 1、W2、b1 and b 2 are learnable parameters, max (0, x) is a ReLU activation function.
Further, the intention recognition module adopts a polynomial naive bayes model, which specifically includes:
Receiving emotion analysis results from a text emotion analysis module, wherein the emotion analysis results comprise emotion tendencies (such as positive and negative), emotion intensity and emotion expression and text information directly converted from a client voice input module;
When training a polynomial naive Bayes model, ensuring that the model can process the expanded multidimensional feature vector, and training the polynomial naive Bayes model by using the expanded feature vector containing text information and emotion analysis results and corresponding intention labels;
performing intention recognition and classification, comprehensively considering word frequency information and emotion characteristics of a text, calculating posterior probability of each intention category under a given expansion feature vector, and selecting the category with the highest posterior probability as a predicted intention.
Further, the polynomial naive bayes model is specifically as follows:
based on the bayesian theorem, given a feature vector x= (X 1,x2,...,xn) and a target class C k, each feature is set to be independent of the other feature, the bayesian theorem is expressed as:
Where P (C k |x) is the posterior probability of class C k for a given feature vector X, P (x|c k) is the probability of feature vector X occurrence for class C k, decomposed into products of conditional probabilities of features under that class according to the na iotave bayes hypothesis:
p (X 1|Ck)×P(x2|Ck)×...×P(xn|Ck),P(Ck) is the prior probability of the class C k, i.e., the probability of the occurrence of the class C k in all data, P (X) is the probability of the occurrence of the feature vector X, and as a normalization factor, the sum of the posterior probabilities is ensured to be 1.
Further, the posterior probability is calculated as: in intent recognition, the posterior probability of each intent for a given feature vector (i.e., text feature and emotion analysis result) is calculated, and since P (X) is constant for all intent categories, the calculation of the simplified posterior probability is:
P(Ck|X)∝P(X|Ck)·P(Ck)；
Further simplified into ：P(Ck|X)∝P(x1|Ck)×P(x2|Ck)×...×P(xn|Ck)×P(Ck).
Further, the database module specifically includes:
Historical outbound data storage: the method comprises the steps of storing contact information of a client, outbound time, outbound duration, outbound script content and response of the client through a structured format, wherein each outbound attempt is recorded as one row in a table and contains all relevant detail fields;
Customer feedback information storage: in addition to basic outbound data, the database module also collects and stores direct feedback information of the customer, including evaluations, suggestions and complaints provided by the customer through different channels after the end of the outbound, the customer feedback information being stored in text form and associated with corresponding outbound records for subsequent analysis and training of the intent recognition model;
Intent recognition model data store: the database module also includes storing structures and parameters of the intent recognition model, including feature data, model parameters, and model performance metrics used for model training.
Further, the outbound policy generation template specifically includes:
Intent recognition result application: receiving a customer intent analysis result provided by an intent recognition module, including an intent category (request for help, complaint, product/service of interest, etc.) and an intent strength, as input for determining an outbound policy;
historical data analysis: the outbound strategy generation module analyzes the historical interaction mode, feedback trend and outbound result of the client or the client group by inquiring the historical outbound data and the client feedback information stored in the database module;
policy customization and optimization: based on the intention recognition result and the historical data analysis, a personalized outbound strategy is designed, including selecting the most appropriate outbound time, deciding the outbound script and the conversation strategy to use and setting up the response scheme of the customer questions and objections.
Further, the cognitive behavior analysis and adaptation module specifically includes:
And (3) data collection: the cognitive behavioral analysis and adaptation module is designed to capture and record behavioral data of the client in each interaction, including query content of the client, selected service options, response time, language used and emotion expressed;
Behavioral pattern analysis: analyzing the collected behavior data by adopting a cognitive modeling technology, wherein the cognitive modeling aims at understanding a decision process of a client by simulating a human thinking process and a behavior mode, and comprises the steps of constructing a decision tree and a state transition diagram to explain the behavior of the client;
Future behavior prediction: based on the results of the cognitive model analysis, the cognitive behavioral analysis and adaptation module predicts the behavior and response taken by the client under specific circumstances.
Further, the decision tree is based on a tree structure, wherein each internal node represents a test on an attribute, each branch represents a test result, a leaf node of the tree represents a category or a decision result, and in the cognitive behavior analysis and adaptation module, the decision tree is used for simulating a process of making a decision by a client, and specifically comprises:
collecting and sorting customer interaction data, including questions posed by the customer, selected service options, reaction time;
selecting features for predicting customer intent based on the collected data;
constructing a decision tree: the decision tree is constructed from training data using algorithm C4.5, the C4.5 algorithm uses the information gain ratio to select features, the information gain ratio is calculated as:
Wherein Gain (S, a) is the information Gain of feature a over dataset S, splitInfo (S, a) is the split information to split dataset S using feature a;
model application: predicting a possible result of the new customer interaction, namely the intention of the customer, by using the constructed decision tree model;
the state transition diagram is used for representing different states and transitions in the customer interaction process, and specifically comprises the following steps:
Defining a state: determining key states in the customer interaction process, including inquiring product information, requesting support and providing feedback;
defining a conversion: determining a transition condition between states based on the behavior of the client;
constructing a state transition diagram: drawing a state transition diagram which represents all states in the customer interaction process and transition paths among the states;
Model application: the state transition diagram is utilized to analyze the behavior patterns of the client, predict actions taken by the client in a particular state, and state transitions resulting from the actions.
The invention has the beneficial effects that:
According to the invention, the prediction accuracy of the intention of the client is remarkably improved by comprehensively applying the advanced technologies such as voice recognition, text emotion analysis, intention recognition, cognitive behavior analysis and the like, and the system can deeply understand the requirements and the emotion states of the client, so that the outbound task is more accurate and effective, and the satisfaction degree of the client is greatly improved.
The database module of the system can efficiently store and manage information such as historical outbound data, customer feedback, intention recognition models and the like, provides rich data support for the outbound strategy generation module, and the data driving method enables the outbound strategy to be more scientific and personalized and is beneficial to realizing optimal configuration of resources and maximization of outbound efficiency.
According to the invention, by introducing the cognitive behavior analysis and adaptation module (CBAA module), the system can analyze the client behavior mode based on the past and current interaction data, can predict future behaviors and reactions, and provides possibility for formulating a dynamically adapted outbound strategy. This dynamic adaptation capability ensures that the system can be continuously optimized as the customer's behavior changes, maintaining long-term service effectiveness and customer satisfaction.
Drawings
In order to more clearly illustrate the invention or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the description below are only of the invention and that other drawings can be obtained from them without inventive effort for a person skilled in the art.
Fig. 1 is a schematic diagram of functional modules of an analysis system according to an embodiment of the present invention.
Detailed Description
The present invention will be further described in detail with reference to specific embodiments in order to make the objects, technical solutions and advantages of the present invention more apparent.
It is to be noted that unless otherwise defined, technical or scientific terms used herein should be taken in a general sense as understood by one of ordinary skill in the art to which the present invention belongs. The terms "first," "second," and the like, as used herein, do not denote any order, quantity, or importance, but rather are used to distinguish one element from another. The word "comprising" or "comprises", and the like, means that elements or items preceding the word are included in the element or item listed after the word and equivalents thereof, but does not exclude other elements or items. The terms "connected" or "connected," and the like, are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "upper", "lower", "left", "right", etc. are used merely to indicate relative positional relationships, which may also be changed when the absolute position of the object to be described is changed.
As shown in fig. 1, an intelligent outbound customer intention prediction analysis system comprises a customer voice input module, a voice-to-text module, a text emotion analysis module, an intention recognition module, a database module, an outbound strategy generation module, an outbound execution module and a cognitive behavior analysis and adaptation module, wherein the client voice input module is used for receiving a voice message from a customer;
the client voice input module is used for receiving a voice signal of a client;
The voice-to-text module converts the received voice signal into text information;
the text emotion analysis module analyzes emotion tendencies of the converted text and judges the emotion state of the client;
the intention recognition module predicts the intention of the client according to the text information and the emotion analysis result and by combining a preset intention recognition model;
The database module stores historical outbound data, customer feedback information and an intention recognition model;
the outbound strategy generation module generates a targeted outbound strategy according to the intention recognition result and historical data in the database;
The outbound execution module automatically executes outbound tasks according to the generated outbound strategy and interacts with clients;
The cognitive behavior analysis and adaptation module is responsible for collecting and analyzing behavior data of clients in past and current interactions, recognizing the client behavior mode by using a cognitive modeling technology, predicting future behaviors and reactions, and can adjust outbound strategies according to the client cognitive mode and behavior history to realize higher-level personalized interactions.
The text emotion analysis module specifically comprises:
Cleaning and standardizing text data through a preprocessing step, including removing stop words, correcting spelling, part-of-speech tagging and stem extraction, and then converting the text into a vector form by utilizing an embedding technology for machine learning model processing;
and modeling and analyzing the emotion tendencies in the text data by adopting a transducer model.
The transducer model specifically comprises:
Self-attention mechanism: the purpose is to calculate for each token in the input sequence its attention weight for all tokens in the sequence, given a representation X 1,X2,...,Xn of one sequence, for each token in the sequence X i, the self-attention mechanism maps it to a query vector Q, a key vector K and a value vector V, obtained by linear transformation:
Q=xw Q,K＝XWK,V＝XWV, where W Q、WK、WV is a matrix of learnable parameters;
the attention weight is calculated and applied to the value vector, the calculation method is as follows:
Wherein d k is the dimension of the key vector, and division operation is used for scaling the dot product to prevent gradient disappearance or explosion;
Multi-head attention: the transducer improves the model performance through a multi-head attention mechanism, and the self-attention process is performed multiple times in parallel, and different parameter matrixes W Q、WK、WV are used each time:
MultiHead (Q, K, V) = Concat (head 1,...,headh)WO, where each head i is a self-attention layer and W O is another learnable parameter matrix for combining the outputs of different heads;
position feed forward network: each encoder and decoder layer in the transducer model also contains a position feed forward network, the same fully connected layer is applied independently to the tokens at each position:
FFN (x) =max (0, xw 1+b1)W2+b2, where W 1、W2、b1 and b 2 are learnable parameters, max (0, x) is a ReLU activation function.
In the text emotion analysis module, the components of the transform model are applied to the text emotion analysis module, the model firstly determines the importance of each word in the text through a self-attention mechanism, then analyzes the emotion of the text from different angles through a multi-head attention mechanism, and finally further refines the emotion representation of each word through a position feed-forward network. This series of processes enables the model to accurately identify the emotional state of the customer, whether expressed as a pronounced emotion or subtle emotional tendency implicit in a complex context.
Through the mode, the transducer model can accurately judge the emotion state of the client, and powerful support is provided for the intention recognition module and the cognitive behavior analysis and adaptation module, so that the intelligent outbound system can more accurately predict the intention of the client and formulate a personalized outbound strategy, and the satisfaction degree and outbound effect of the client are improved.
The intention recognition module adopts a polynomial naive Bayesian model, and specifically comprises the following steps:
Receiving emotion analysis results from a text emotion analysis module, wherein the emotion analysis results comprise emotion tendencies (such as positive and negative), emotion intensity and emotion expression and text information directly converted from a client voice input module; two additional features may be added to each text: one representing emotion polarity (e.g., positive = 1, negative = -1, neutral = 0) and the other representing emotion intensity (continuous value, ranging from 0 to 1), feature vectors are extended by stitching emotion features on the basis of a bag of words model. Thus, the feature vector of each text not only contains word frequency information, but also contains emotion polarity and emotion intensity information;
When training a polynomial naive Bayes model, ensuring that the model can process the expanded multidimensional feature vector, and training the polynomial naive Bayes model by using the expanded feature vector containing text information and emotion analysis results and corresponding intention labels;
performing intention recognition and classification, comprehensively considering word frequency information and emotion characteristics of a text, calculating posterior probability of each intention category under a given expansion feature vector, and selecting the category with the highest posterior probability as a predicted intention.
The polynomial na iotave bayes model is specifically as follows:
based on the bayesian theorem, given a feature vector x= (X 1,x2,...,xn) and a target class C k, each feature is set to be independent of the other feature, the bayesian theorem is expressed as:
Where P (C k |x) is the posterior probability of class C k for a given feature vector X, P (x|c k) is the probability of feature vector X occurrence for class C k, decomposed into products of conditional probabilities of features under that class according to the na iotave bayes hypothesis:
p (X 1|Ck)×P(x2|Ck)×...×P(xn|Ck),P(Ck) is the prior probability of the class C k, i.e., the probability of the occurrence of the class C k in all data, P (X) is the probability of the occurrence of the feature vector X, and as a normalization factor, the sum of the posterior probabilities is ensured to be 1.
The posterior probability is calculated as: in intent recognition, the posterior probability of each intent for a given feature vector (i.e., text feature and emotion analysis result) is calculated, and since P (X) is constant for all intent categories, the calculation of the simplified posterior probability is:
P(Ck|X)∝P(X|Ck)·P(Ck)；
Further simplified into ：P(Ck|X)∝P(x1|Ck)×P(x2|Ck)×...×P(xn|Ck)×P(Ck);
In the intelligent outbound system, X contains text features (such as keyword occurrence frequency) and emotion analysis results (such as emotion polarity and intensity) extracted from a customer dialog, each X i represents a feature, which may be TFIDF value, emotion polarity or emotion intensity of a word, etc., for each possible customer intention C k (such as request for help, complaint, etc.), a posterior probability P (C k |x) is calculated, and then the intention with the highest posterior probability is selected as the prediction result.
The database module specifically comprises:
Historical outbound data storage: the method comprises the steps of storing contact information of a client, outbound time, outbound duration, outbound script content and response of the client through a structured format, wherein each outbound attempt is recorded as one row in a table and contains all relevant detail fields;
Customer feedback information storage: in addition to basic outbound data, the database module also collects and stores direct feedback information of the customer, including evaluations, suggestions and complaints provided by the customer through different channels after the end of the outbound, the customer feedback information being stored in text form and associated with corresponding outbound records for subsequent analysis and training of the intent recognition model;
Intent recognition model data store: the database module also includes storing structures and parameters of the intent recognition model, including feature data, model parameters, and model performance metrics used for model training.
The outbound strategy generation template specifically comprises:
Intent recognition result application: receiving a customer intent analysis result provided by an intent recognition module, including an intent category (request for help, complaint, product/service of interest, etc.) and an intent strength, as input for determining an outbound policy;
historical data analysis: the outbound strategy generation module analyzes the historical interaction mode, feedback trend and outbound result of the client or the client group by inquiring the historical outbound data and the client feedback information stored in the database module;
policy customization and optimization: based on the intention recognition result and the historical data analysis, a personalized outbound strategy is designed, including selecting the most appropriate outbound time, deciding the outbound script and the conversation strategy to use and setting up the response scheme of the customer questions and objections.
The cognitive behavior analysis and adaptation module specifically comprises:
And (3) data collection: the cognitive behavior analysis and adaptation module is designed to capture and record behavior data of the client in each interaction, including query content of the client, selected service options, response time, used language and expressed emotion, and the data are derived from the current interaction instance and also include past interaction history of the client, so that rich context information is provided for deep analysis;
Behavioral pattern analysis: analyzing the collected behavior data by adopting a cognitive modeling technology, wherein the cognitive modeling aims at understanding a decision process of a client by simulating a human thinking process and a behavior mode, and comprises the steps of constructing a decision tree and a state transition diagram to explain the behavior of the client;
future behavior prediction: based on the result of the cognitive model analysis, the cognitive behavior analysis and adaptation module predicts the behavior and response taken by the client under a specific situation, and the prediction helps the outbound strategy generation module to formulate a more personalized and effective communication strategy so as to improve the satisfaction degree and the outbound success rate of the client.
The decision tree is based on a tree structure, wherein each internal node represents a test on an attribute, each branch represents a test result, a leaf node of the tree represents a category or a decision result, and in the cognitive behavior analysis and adaptation module, the decision tree is used for simulating a process of making a decision by a client, and the decision tree specifically comprises the following steps:
collecting and sorting customer interaction data, including questions posed by the customer, selected service options, reaction time;
selecting features for predicting customer intent based on the collected data;
constructing a decision tree: the decision tree is constructed from training data using algorithm C4.5, the C4.5 algorithm uses the information gain ratio to select features, the information gain ratio is calculated as:
Wherein Gain (S, a) is the information Gain of feature a over dataset S, splitInfo (S, a) is the split information to split dataset S using feature a;
model application: predicting a possible result of the new customer interaction, namely the intention of the customer, by using the constructed decision tree model;
The state transition diagram is used for representing different states and transitions in the customer interaction process, and specifically comprises the following steps:
Defining a state: determining key states in the customer interaction process, including inquiring product information, requesting support and providing feedback;
Defining a conversion: determining conversion conditions between states based on the behavior of the client, wherein the conversion from querying product information to requesting support may be based on the client making a specific support request;
constructing a state transition diagram: drawing a state transition diagram which represents all states in the customer interaction process and transition paths among the states;
Model application: analyzing the behavior mode of the client by using the state transition diagram, predicting actions taken by the client in a specific state, and state transitions caused by the actions;
In the intelligent outbound customer intention prediction analysis system, the cognitive behavior analysis and adaptation module can deeply understand the probability that a customer makes a specific decision under a specific situation through a decision tree, so that the intention of the customer is predicted, and the state transition diagram further enhances the understanding of the system to the customer interaction process, so that the system can identify the pattern of the customer behavior and predict the next action of the customer.
The method combining the decision tree and the state transition diagram not only improves the accuracy of intention recognition, but also enables the outbound strategy generation module to formulate more personalized and effective communication strategies according to the specific situation and the behavior mode of the client, and finally improves the satisfaction degree and the outbound success rate of the client.
Those of ordinary skill in the art will appreciate that: the discussion of any of the embodiments above is merely exemplary and is not intended to suggest that the scope of the invention is limited to these examples; the technical features of the above embodiments or in the different embodiments may also be combined within the idea of the invention, the steps may be implemented in any order and there are many other variations of the different aspects of the invention as described above, which are not provided in detail for the sake of brevity.
The present invention is intended to embrace all such alternatives, modifications and variances which fall within the broad scope of the appended claims. Therefore, any omission, modification, equivalent replacement, improvement, etc. of the present invention should be included in the scope of the present invention.
Claims (7)
1. The intelligent outbound customer intention prediction analysis system is characterized by comprising a customer voice input module, a voice-to-text module, a text emotion analysis module, an intention recognition module, a database module, an outbound strategy generation module, an outbound execution module and a cognitive behavior analysis and adaptation module, wherein the client voice input module is used for receiving voice from a customer;
the client voice input module is used for receiving a voice signal of a client;
The voice-to-text module converts the received voice signal into text information;
the text emotion analysis module analyzes emotion tendencies of the converted text and judges the emotion state of the client;
The intention recognition module predicts the intention of the client according to the text information and the emotion analysis result by combining a preset intention recognition model, and the intention recognition module adopts a polynomial naive Bayesian model and specifically comprises the following steps:
Receiving emotion analysis results from a text emotion analysis module, wherein the emotion analysis results comprise emotion tendencies, emotion intensities and emotion expressions, and text information is directly converted from a client voice input module;
When training a polynomial naive Bayes model, ensuring that the model can process the expanded multidimensional feature vector, and training the polynomial naive Bayes model by using the expanded feature vector containing text information and emotion analysis results and corresponding intention labels;
Performing intention recognition and classification, comprehensively considering word frequency information and emotion characteristics of a text, calculating posterior probability of each intention category under a given expansion feature vector, and selecting a category with the highest posterior probability as a prediction intention;
the polynomial naive bayes model is specifically as follows:
Based on the bayesian theorem, given a feature vector x= (X 1x2…xn) and a target class C k, each feature is set to be independent of the other feature, the bayesian theorem is expressed as:
Where P (C k |x) is the posterior probability of class C k for a given feature vector X, P (x|c k) is the probability of feature vector X occurrence for class C k, decomposed into products of conditional probabilities of features under that class according to the na iotave bayes hypothesis:
P (X 1|Ck)×P(x2|Ck)×…×P(xn|Ck),P(Ck) is the prior probability of the class C k, that is, the probability of the occurrence of the class C k in all data, P (X) is the probability of the occurrence of the feature vector X, and as a normalization factor, the sum of the posterior probabilities is ensured to be 1;
the posterior probability is calculated as: in intent recognition, the posterior probability of each intent for a given feature vector is calculated, and since P (X) is constant for all intent categories, the calculation of the simplified posterior probability is:
P(Ck|X)∝P(X|Ck)·P(Ck)；
Further simplified into ：P(Ck|X)∝P(x1|Ck)×P(x2|Ck)×…×P(xn|Ck)×P(Ck);
The database module stores historical outbound data, customer feedback information and an intention recognition model;
the outbound strategy generation module generates a targeted outbound strategy according to the intention recognition result and historical data in the database;
The outbound execution module automatically executes outbound tasks according to the generated outbound strategy and interacts with clients;
the cognitive behavior analysis and adaptation module is responsible for collecting and analyzing behavior data of clients in past and current interactions, identifying client behavior patterns by using cognitive modeling technology, and predicting future behaviors and reactions.
2. The intelligent outbound customer intent prediction analysis system as claimed in claim 1, wherein the text emotion analysis module specifically comprises:
Cleaning and standardizing text data through a preprocessing step, including removing stop words, correcting spelling, part-of-speech tagging and stem extraction, and then converting the text into a vector form by utilizing an embedding technology for machine learning model processing;
and modeling and analyzing the emotion tendencies in the text data by adopting a transducer model.
3. The intelligent outbound customer intent prediction analysis system according to claim 2, wherein the transducer model specifically comprises:
Self-attention mechanism: the purpose is to calculate for each token in the input sequence its attention weight for all tokens in the sequence, given a representation X 1,X2,…,Xn of one sequence, for each token in the sequence X i, the self-attention mechanism maps it to a query vector Q, a key vector K and a value vector V, obtained by linear transformation:
q=xw Q,K＝XWK,V＝XWV, where W Q、WK、WV is a matrix of learnable parameters;
the attention weight is calculated and applied to the value vector, the calculation method is as follows:
Wherein d k is the dimension of the key vector, and division operation is used for scaling the dot product to prevent gradient disappearance or explosion;
Multi-head attention: the transducer improves the model performance through a multi-head attention mechanism, and the self-attention process is performed multiple times in parallel, and different parameter matrixes W Q、WK、WV are used each time:
MultiHead (Q, K, V) = Concat (head 1,...,headh)WO, where each head i is a self-attention layer and W O is another learnable parameter matrix for combining the outputs of different heads;
Position feed forward network: each encoder and decoder layer in the transducer model also contains a position feed forward network, the same fully connected layer is applied independently to the tokens at each position:
FFN (x) =max (0, xw 1+b1)W2+b2, where W 1、W2、b1 and b 2 are learnable parameters, max (0, x) is a ReLU activation function.
4. An intelligent outbound customer intent prediction analysis system as claimed in claim 3 wherein said database module comprises:
Historical outbound data storage: the method comprises the steps of storing contact information of a client, outbound time, outbound duration, outbound script content and response of the client through a structured format, wherein each outbound attempt is recorded as one row in a table and contains all relevant detail fields;
Customer feedback information storage: in addition to basic outbound data, the database module also collects and stores direct feedback information of the customer, including evaluations, suggestions and complaints provided by the customer through different channels after the end of the outbound, the customer feedback information being stored in text form and associated with corresponding outbound records for subsequent analysis and training of the intent recognition model;
Intent recognition model data store: the database module also includes storing structures and parameters of the intent recognition model, including feature data, model parameters, and model performance metrics used for model training.
5. The intelligent outbound customer intent prediction analysis system according to claim 4, wherein the outbound policy generation template specifically comprises:
intent recognition result application: receiving a customer intent analysis result provided by an intent recognition module, wherein the customer intent analysis result comprises an intent category and an intent strength, and the customer intent analysis result is used as an input for determining an outbound strategy;
historical data analysis: the outbound strategy generation module analyzes the historical interaction mode, feedback trend and outbound result of the client or the client group by inquiring the historical outbound data and the client feedback information stored in the database module;
policy customization and optimization: based on the intention recognition result and the historical data analysis, a personalized outbound strategy is designed, including selecting the most appropriate outbound time, deciding the outbound script and the conversation strategy to use and setting up the response scheme of the customer questions and objections.
6. The intelligent outbound customer intent prediction analysis system of claim 1 wherein the cognitive behavioral analysis and adaptation module specifically comprises:
And (3) data collection: the cognitive behavioral analysis and adaptation module is designed to capture and record behavioral data of the client in each interaction, including query content of the client, selected service options, response time, language used and emotion expressed;
Behavioral pattern analysis: analyzing the collected behavior data by adopting a cognitive modeling technology, wherein the cognitive modeling aims at understanding a decision process of a client by simulating a human thinking process and a behavior mode, and comprises the steps of constructing a decision tree and a state transition diagram to explain the behavior of the client;
Future behavior prediction: based on the results of the cognitive model analysis, the cognitive behavioral analysis and adaptation module predicts the behavior and response taken by the client under specific circumstances.
7. The intelligent outbound customer intent prediction analysis system according to claim 6 wherein the decision tree is based on a tree structure wherein each internal node represents a test on an attribute, each branch represents the result of the test, and the leaf nodes of the tree represent categories or decision results, and wherein in the cognitive behavioral analysis and adaptation module the decision tree is used to simulate the process of making decisions by the customer, comprising in particular:
collecting and sorting customer interaction data, including questions posed by the customer, selected service options, reaction time;
selecting features for predicting customer intent based on the collected data;
constructing a decision tree: the decision tree is constructed from training data using algorithm C4.5, the C4.5 algorithm uses the information gain ratio to select features, the information gain ratio is calculated as:
Wherein Gain (S, a) is the information Gain of feature a over dataset S, splitInfo (S, a) is the split information to split dataset S using feature a;
model application: predicting a possible result of the new customer interaction, namely the intention of the customer, by using the constructed decision tree model;
the state transition diagram is used for representing different states and transitions in the customer interaction process, and specifically comprises the following steps:
Defining a state: determining key states in the customer interaction process, including inquiring product information, requesting support and providing feedback;
defining a conversion: determining a transition condition between states based on the behavior of the client;
constructing a state transition diagram: drawing a state transition diagram which represents all states in the customer interaction process and transition paths among the states;
Model application: the state transition diagram is utilized to analyze the behavior patterns of the client, predict actions taken by the client in a particular state, and state transitions resulting from the actions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410248851.1A CN117834780B (en) | 2024-03-05 | 2024-03-05 | Intelligent outbound customer intention prediction analysis system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410248851.1A CN117834780B (en) | 2024-03-05 | 2024-03-05 | Intelligent outbound customer intention prediction analysis system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117834780A CN117834780A (en) | 2024-04-05 |
| CN117834780B true CN117834780B (en) | 2024-05-14 |
Family
ID=90515789
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410248851.1A Active CN117834780B (en) | 2024-03-05 | 2024-03-05 | Intelligent outbound customer intention prediction analysis system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117834780B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108197742A (en) * | 2017-12-29 | 2018-06-22 | 平安健康保险股份有限公司 | Continuation of insurance behavior prediction method, system and the computer readable storage medium of user |
| KR101932707B1 (en) * | 2017-09-08 | 2018-12-26 | 중앙대학교 산학협력단 | Apparatus and Method for Sunfull Bot with Sunfull Automation Function |
| CN110909529A (en) * | 2019-11-27 | 2020-03-24 | 国网能源研究院有限公司 | User emotion analysis and prejudgment system of company image promotion system |
| CN113656551A (en) * | 2021-08-19 | 2021-11-16 | 中国银行股份有限公司 | Intelligent outbound interruption method and device, storage medium and electronic equipment |
| CN115858744A (en) * | 2022-11-09 | 2023-03-28 | 深圳市思为软件技术有限公司 | Outbound method, device and storage medium based on AI |
| CN116246632A (en) * | 2023-01-31 | 2023-06-09 | 中国工商银行股份有限公司 | Method and device for guiding external call operation |
| CN116542256A (en) * | 2023-07-05 | 2023-08-04 | 广东数业智能科技有限公司 | Natural language understanding method and device integrating dialogue context information |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104063427A (en) * | 2014-06-06 | 2014-09-24 | 北京搜狗科技发展有限公司 | Expression input method and device based on semantic understanding |
-
2024
- 2024-03-05 CN CN202410248851.1A patent/CN117834780B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101932707B1 (en) * | 2017-09-08 | 2018-12-26 | 중앙대학교 산학협력단 | Apparatus and Method for Sunfull Bot with Sunfull Automation Function |
| CN108197742A (en) * | 2017-12-29 | 2018-06-22 | 平安健康保险股份有限公司 | Continuation of insurance behavior prediction method, system and the computer readable storage medium of user |
| CN110909529A (en) * | 2019-11-27 | 2020-03-24 | 国网能源研究院有限公司 | User emotion analysis and prejudgment system of company image promotion system |
| CN113656551A (en) * | 2021-08-19 | 2021-11-16 | 中国银行股份有限公司 | Intelligent outbound interruption method and device, storage medium and electronic equipment |
| CN115858744A (en) * | 2022-11-09 | 2023-03-28 | 深圳市思为软件技术有限公司 | Outbound method, device and storage medium based on AI |
| CN116246632A (en) * | 2023-01-31 | 2023-06-09 | 中国工商银行股份有限公司 | Method and device for guiding external call operation |
| CN116542256A (en) * | 2023-07-05 | 2023-08-04 | 广东数业智能科技有限公司 | Natural language understanding method and device integrating dialogue context information |
Non-Patent Citations (3)
| Title |
|---|
| Implementation of The Naïve Bayes Algorithm with Feature Selection using Genetic Algorithm for Sentiment Review Analysis of Fashion Online Companies;Siti Ernawati，Eka Rini Yulia，etc.;《2018 6th International Conference on Cyber and IT Service Management (CITSM)》;20190328;全文 * |
| 基于深度学习的问答***设计;潘明;《CNKI-硕士论文》;20210615;正文第27，33，40页 * |
| 情绪识别技术在电力智能客服***中的应用研究;张晓慧;孙德艳;马永波;王明珠;曹璐;李承桓;;电子器件;20201020(05);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117834780A (en) | 2024-04-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20230377312A1 (en) | System and method for neural network orchestration | |
| US5671333A (en) | Training apparatus and method | |
| CN110032630A (en) | Talk about art recommendation apparatus, method and model training equipment | |
| Audhkhasi et al. | A globally-variant locally-constant model for fusion of labels from multiple diverse experts without using reference labels | |
| Ashraf et al. | A globally regularized joint neural architecture for music classification | |
| KR100436678B1 (en) | Method and system for recognizing end-user transactions | |
| CN117435716B (en) | Data processing method and system of power grid man-machine interaction terminal | |
| US20230237276A1 (en) | System and Method for Incremental Estimation of Interlocutor Intents and Goals in Turn-Based Electronic Conversational Flow | |
| CN111930912A (en) | Dialogue management method, system, device and storage medium | |
| Sridhar et al. | Unsupervised personalization of an emotion recognition system: The unique properties of the externalization of valence in speech | |
| CN117912710A (en) | Teenager mental health data analysis and early warning system | |
| CN117608650B (en) | Business flow chart generation method, processing device and storage medium | |
| JP6924975B2 (en) | Sound analyzer and its processing method, program | |
| CN116820711B (en) | Task driven autonomous agent method | |
| CN117834780B (en) | Intelligent outbound customer intention prediction analysis system | |
| CN117171403A (en) | Data processing method, device, computer equipment and storage medium | |
| Prudêncio et al. | Selective generation of training examples in active meta-learning | |
| Ali et al. | Meta-Analysis of Deep Learning Approaches for Machine Learning Chatbots | |
| Manousaki et al. | Vlmah: Visual-linguistic modeling of action history for effective action anticipation | |
| CN114490974A (en) | Automatic information reply method, device, system, electronic equipment and readable medium | |
| Tariq et al. | Time efficient end-state prediction through hybrid trace decomposition using process mining | |
| CN115640323B (en) | Emotion prediction method based on transition probability | |
| CN117056519B (en) | Cross-domain-oriented automatic generation method for comprehensive report of legal opinions | |
| US20240256636A1 (en) | Artificial intelligence system for media item classification using transfer learning and active learning | |
| EP4303716A1 (en) | Method for generating data input, data input system and computer program |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |