CN114416985A - Customer intention analysis method, system, device and storage medium - Google Patents
Customer intention analysis method, system, device and storage medium Download PDFInfo
- Publication number
- CN114416985A CN114416985A CN202210039128.3A CN202210039128A CN114416985A CN 114416985 A CN114416985 A CN 114416985A CN 202210039128 A CN202210039128 A CN 202210039128A CN 114416985 A CN114416985 A CN 114416985A
- Authority
- CN
- China
- Prior art keywords
- intention
- client
- customer
- model
- sequences
- 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.)
- Pending
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
- G06F16/35—Clustering; Classification
-
- 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
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- 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
- G06F16/3347—Query execution using vector based model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06N3/047—Probabilistic or stochastic networks
-
- 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
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Human Computer Interaction (AREA)
- Machine Translation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application provides a method, a system, equipment and a storage medium for analyzing a client intention, which are used for identifying the client intention according to the conversation interaction information of the client to obtain a plurality of intention sequences of a plurality of clients; the sequence of intentions includes a plurality of customer intentions arranged in time; embedding a plurality of intention sequence training words of a plurality of clients into a network to obtain a client intention model; inputting at least one time of interaction information of a customer to be predicted into a customer intention model to obtain a plurality of customer intention vectors; calculating cosine distances between the plurality of customer intention vectors and the success intention vectors, comparing the cosine distances, and selecting the customer intention vector corresponding to the smallest cosine distance; the corresponding obtained client intention is a client intention prediction result; the success intention vector is a pre-stored success intention vector of a service or a single client. According to the method and the device, the client intention modeling is carried out according to the client intention sequence, so that the accuracy of the client intention can be improved, and the business can be predicted in a single direction.
Description
Technical Field
The application belongs to the technical field of artificial intelligence, and particularly relates to a method, a system, equipment and a storage medium for analyzing a client intention.
Background
Recognition of intent plays an important role in modern AI dialog systems, and natural Language processing nlp (natural Language processing) is a subject of Language questions that studies human interaction with computers. The client intention generated by the NLP system plays a very important role in the process of representing user portraits and service recommendations for users.
However, the client intentions generated by the NLP system are usually analyzed independently as isolated components, and the relationship between the client intentions, such as time information, is ignored, so that it is difficult to capture the continuous intention change of the client during the client intention analysis, and thus accurate intention analysis and modeling cannot be performed. Even if the multiple rounds of intention recognition algorithms take the form of only roughly inputting previous independent customer information into the model, intention recognition results intervening toward the direction of business advantage cannot be obtained.
Disclosure of Invention
The invention provides a method, a system, equipment and a storage medium for analyzing a client intention, and aims to solve the problem that the client intention is inaccurately identified in the conventional AI dialogue system.
According to a first aspect of embodiments of the present application, there is provided a customer intention analysis method, including the steps of:
identifying the intention of the client according to the conversation interaction information of the client to obtain a plurality of intention sequences of a plurality of clients; the sequence of intentions includes a plurality of customer intentions arranged in time;
embedding a plurality of intention sequence training words of a plurality of clients into a network to obtain a client intention model;
inputting at least one time of interaction information of a customer to be predicted into a customer intention model to obtain a plurality of customer intention vectors;
calculating cosine distances between the plurality of customer intention vectors and the success intention vectors, comparing the cosine distances, and selecting the customer intention vector corresponding to the smallest cosine distance; the corresponding obtained client intention is a client intention prediction result; the success intention vector is a pre-stored success intention vector of a service or a single client.
In some embodiments of the present application, the obtaining a client intention model by embedding multiple intention sequence training words of multiple clients into a network specifically includes:
arranging a plurality of intention sequences to obtain an intention directed graph; the nodes of the intention directed graph are client intents, and the weight of the edges of the intention directed graph is the probability of transition between the nodes of the intention directed graph;
and traversing the intention directed graph to obtain a plurality of random intention sequences, and embedding training words into the network according to the plurality of random intention sequences to obtain a client intention model.
In some embodiments of the present application, the calculation formula of the probability of transition between nodes of the intention directed graph is:
Pmn=Cn/Cm;
where Cm is the number of intended sequences on the intended directed graph node m that pass through m but are not the end point, Cn is the number of intended sequences that pass from the intended directed graph node m to the intended directed graph node n, and Pmn is the probability of a transition between the intended directed graph node m and the intended directed graph node n.
In some embodiments of the present application, after obtaining the client intention model by embedding a plurality of intention sequence training words of a plurality of clients into a network, the method further includes:
and inputting a sequence of the client intention of the client in order success to the client intention model to obtain a success intention vector.
In some embodiments of the present application, the Word embedding network is a Word2vec model, and the Word2vec model combines a CBOW model and a skip-gram model and adopts a negative sampling and hierarchical softmax method.
In some embodiments of the present application, the method for identifying a client intention of a client call interaction information to obtain a plurality of intention sequences of a plurality of clients specifically includes:
when the customer call interaction information is voice, converting the customer call interaction information into character interaction information;
identifying the intention of the client through a classification model of the intention of the client to obtain a plurality of intention sequences;
through the steps, the intention of the clients is identified according to the conversation interaction information of the clients, and a plurality of intention sequences of the clients are obtained.
In some embodiments of the present application, before the identifying the client intention through the client intention classification model and obtaining the plurality of intention sequences, the method further includes:
acquiring a client intention data set;
and inputting the client intention data set into a client intention classification network for training to obtain a client intention classification model.
According to a second aspect of an embodiment of the present application, there is provided a customer intention analysis system, specifically including:
a client intent classification module: the system comprises a plurality of users, a plurality of communication terminals and a plurality of communication terminals, wherein the communication terminals are used for receiving the communication interaction information of the users; the sequence of intentions includes a plurality of customer intentions arranged in time;
a customer intent model module: the method comprises the steps of embedding multiple intention sequence training words of multiple clients into a network to obtain a client intention model;
a to-be-predicted customer intention vector module: the system comprises a client intention model, a plurality of client intention vectors and a plurality of data processing units, wherein the client intention model is used for inputting at least one time of interaction information of a client to be predicted to the client intention model to obtain a plurality of client intention vectors;
a client intent prediction module: the system comprises a plurality of client intention vectors, a plurality of success intention vectors and a plurality of cosine distances, wherein the client intention vectors are used for calculating and comparing cosine distances between the plurality of client intention vectors and the success intention vectors, and the client intention vector corresponding to the smallest cosine distance is selected; the corresponding obtained client intention is a client intention prediction result; the success intention vector is a pre-stored success intention vector of a service or a single client.
According to a third aspect of embodiments of the present application, there is provided a client intention analysis device including:
a memory: for storing executable instructions; and
a processor for interfacing with the memory to execute the executable instructions to accomplish the above customer intent analysis method.
According to a fourth aspect of embodiments of the present application, there is provided a computer-readable storage medium having a computer program stored thereon; the computer program is executed by a processor to implement a client intent analysis method.
By adopting the method, the system, the equipment and the storage medium for analyzing the client intention in the embodiment of the application, the client intention is identified according to the conversation interaction information of the client, and a plurality of intention sequences of a plurality of clients are obtained; the sequence of intentions includes a plurality of customer intentions arranged in time; embedding a plurality of intention sequence training words of a plurality of clients into a network to obtain a client intention model; inputting at least one time of interaction information of a customer to be predicted into a customer intention model to obtain a plurality of customer intention vectors; calculating cosine distances between the plurality of customer intention vectors and the success intention vectors, comparing the cosine distances, and selecting the customer intention vector corresponding to the smallest cosine distance; the corresponding obtained client intention is a client intention prediction result; the success intention vector is a pre-stored success intention vector of a service or a single client. According to the method and the device, the client intention modeling is carried out according to the client intention sequence, so that the accuracy of the client intention can be improved, and the business can be predicted in a single direction.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the application and together with the description serve to explain the application and not to limit the application. In the drawings:
FIG. 1 is a flow chart illustrating steps of a method for analyzing a customer intention according to an embodiment of the present application;
a simple schematic diagram of the structure of the CBOW model is shown in fig. 2;
a data flow diagram of the CBOW model is shown in fig. 3;
a schematic structural diagram of a customer intention analysis system according to an embodiment of the application is shown in fig. 4;
a schematic structural diagram of a customer intention analysis device according to an embodiment of the present application is shown in fig. 5.
Detailed Description
In the process of implementing the present application, the inventor finds that the client intentions generated by the NLP system are generally independently analyzed as isolated components, and neglects the relationship between the client intentions, such as time information, so that it is difficult to capture the continuous intention change of the client in the client intention analysis, and thus accurate intention analysis and modeling cannot be performed. It is even less possible to obtain an intention recognition result of intervening in a direction toward a business advantage.
In view of the above disadvantages, in particular, the method, system, device and storage medium for analyzing the intention of a client perform intention classification on the interactive information of the client call to obtain a plurality of intention sequences of a plurality of clients; the sequence of intentions includes a plurality of customer intentions arranged in time; embedding a plurality of intention sequence training words of a plurality of clients into a network to obtain a client intention model; inputting at least one time of interaction information of a customer to be predicted into a customer intention model to obtain a plurality of customer intention vectors; calculating cosine distances between the plurality of customer intention vectors and the success intention vectors, comparing the cosine distances, and selecting the customer intention vector corresponding to the smallest cosine distance; and predicting the result for the client intention according to the obtained client intention.
According to the method and the device, the client intention modeling is carried out according to the client intention sequence, so that the accuracy of the client intention can be improved, and the business can be predicted in a single direction.
In order to make the technical solutions and advantages of the embodiments of the present application more apparent, the following further detailed description of the exemplary embodiments of the present application with reference to the accompanying drawings makes it clear that the described embodiments are only a part of the embodiments of the present application, and are not exhaustive of all embodiments. It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict.
Example 1
A flow chart of the steps of a method of analyzing a customer intention according to an embodiment of the present application is shown in fig. 1.
As shown in fig. 1, the method for analyzing the client intention of the embodiment specifically includes the following steps:
s101: performing intention identification according to the conversation interactive information of the customers to obtain a plurality of intention sequences of the customers; the sequence of intentions includes a plurality of customer intents arranged in time.
The method comprises the following steps of identifying the intention of a customer by using the conversation interactive information of the customer to obtain a plurality of intention sequences of a plurality of customers, and specifically comprises the following steps: firstly, when the client call interaction information is voice, converting the client call interaction information into character interaction information; then, identifying the intention of the client through a classification model of the intention of the client to obtain a plurality of intention sequences; finally, the intention of the clients is identified according to the conversation interactive information of the clients, and a plurality of intention sequences of the clients are obtained.
Further, before the client intention classification model identifies the client intention and obtains a plurality of intention sequences, the method further includes: acquiring a client intention data set; and inputting the client intention data set into a client intention classification network for training to obtain a client intention classification model.
S102: and embedding a plurality of intention sequence training words of a plurality of clients into the network to obtain a client intention model.
The method includes the following steps that a plurality of intention sequence training words of a plurality of clients are embedded into a network to obtain a client intention model, and specifically includes the following steps:
1) summarizing the plurality of intention sequences to obtain an intention directed graph; the nodes of the intention directed graph are client intents, and the weight of the edges of the intention directed graph is the probability of transition between the nodes of the intention directed graph;
specifically, the calculation formula of the probability of transition between nodes of the intention directed graph is as follows:
Pmn=Cn/Cm;
where Cm is the number of intended sequences on the intended directed graph node m that pass through m but are not the end point, Cn is the number of intended sequences that pass from the intended directed graph node m to the intended directed graph node n, and Pmn is the probability of a transition between the intended directed graph node m and the intended directed graph node n.
2) And traversing the intention directed graph to obtain a plurality of random intention sequences, and embedding training words into the network according to the plurality of random intention sequences to obtain a client intention model.
During specific training, dividing a plurality of random intention sequences into a training set and a test set; seventy percent of the total number of sequences is typically used as the training set, and the remaining thirty percent is used as the test set.
Inputting training set data to a word embedding network for training, calculating a loss function, feeding back the loss function to the network through a function value, and continuously adjusting network parameters until the training network converges.
And finally, inputting a test set to a word embedding network for testing to obtain a final client intention model.
Specifically, in some embodiments of the present application, the Word embedding network is a Word2vec model, the Word2vec model combines a CBOW model and a skip-gram model, and adopts a negative sampling and hierarchical softmax method.
S103: and inputting at least one piece of interaction information of the customer to be predicted into the customer intention model to obtain a plurality of customer intention vectors.
S104: calculating cosine distances between the plurality of customer intention vectors and the success intention vectors, comparing the cosine distances, and selecting the customer intention vector corresponding to the smallest cosine distance; and predicting the result for the client intention according to the obtained client intention.
Preferably, after obtaining the client intention model by embedding the training words into the network through the multiple intention sequences of multiple clients, the method further includes: and inputting a sequence of the client intention of the client in order success to the client intention model to obtain a success intention vector.
Further illustrating the client intent analysis method of the present embodiment, the identification of intent plays an important role in modern AI dialog systems, especially in controlled rounds of dialog. The system needs to match replies in a targeted manner according to the intention of each task node client.
However, the client intention flow in the general dialog system is set in advance, and it may be difficult to select an intention more favorable for the business in the case of multi-intention.
For example, in the insurance sales scenario, if the client says "i am driving and you say", the sentence includes two intentions, namely "busy" and "definite positive", respectively. The general system will give matching intents based on the confidence of each intention, and the general probability of the intention result will be "busy", ignoring the "definite positive" intention that can be utilized, which is favorable for business development into business.
In the embodiment of the application, in order to enable the system to identify the client intention with higher business value, such as 'clear positive' and the like, the client intention itself is modeled to obtain the client intention model.
The method comprises the following specific steps:
firstly, the intention classification model is used for carrying out preliminary judgment and intention classification on the intention of the client of each call, and a plurality of intention sequences of a plurality of clients are obtained.
Suppose that the sequence of intentions for client I is I11,I12,I13…IijWhere i denotes the customer number and j denotes the intention of the j-th occurrence of the call. In this way, by analyzing all known calls, several sequences of intentions can be obtained.
Then, because some intentions of the multiple customers are overlapping, multiple sequences of intentions are aggregated into an intent directed graph D.
Each node is a different customer intention, and the weight of each edge is the probability of transition between the customer intentions.
The probability Pmn of the transition from the customer intention node m to the customer intention node n is calculated as:
Pmn=Cn/Cm;
where Cm is the number of intended sequences on the intended directed graph node m that pass through m but are not the end point, Cn is the number of intended sequences that pass from the intended directed graph node m to the intended directed graph node n, and Pmn is the probability of a transition between the intended directed graph node m and the intended directed graph node n.
Secondly, traversing the intention directed graph to obtain a plurality of random walk sequences.
For convenience of presentation, we assume that one of the intent sequences is w (t-2), w (t-1), w (t), w (t +1), w (t + 2). And t is the current time.
After the plurality of random walk sequence training words are embedded into the network to obtain a client intention model, namely, a node in the intention sequence is predicted according to client interaction information through the trained neural network.
The Word embedding network is a Word2vec model, the Word2vec model is formed by combining a CBOW model and a skip-gram model and adopting a negative sampling and hierarchical softmax method.
A simplified schematic of the structure of the CBOW model is shown in fig. 2.
As shown in fig. 2, the CBOW model calculates the probability of occurrence of a certain central word according to a plurality of consecutive words before and after the central word, i.e. predicts the target word by context.
The model has three layers, an input layer, a hidden layer and an output layer. W (t-2), w (t-1), w (t +1), w (t +2) of the input layer are the context of the core word w (t).
A data flow diagram of the CBOW model is shown in fig. 3.
As shown in fig. 3, the original corpus assumes a total of V intents. And in the sliding window A of the model, the selected context intention number C is 2A.
First, at the input layer, one-hot of a plurality of contextual intents is input.
Since there are V intents in total, the dimension 1 × V of one-hot of each intention, then there are C vectors of 1 × V input at the input level, so the data dimension is C × V at the input level.
Then, setting the dimension of the finally obtained intention vector to be N, initializing a weight matrix w between the input layer and the hidden layer, wherein the dimension of w is V x N. One-hot (C × V) of the context intention is multiplied by the input weight matrix w (V × N) of the network, resulting in C vectors of 1 × N. They are summed and then averaged to give a hidden layer vector h with dimension 1 x N.
The specific formula of the hidden layer vector h is as follows:
h=C1(x1+x2+...+xc)·w;
where x1, x2.. xc is the one-hot of the context word.
Next, a weight matrix w between the hidden layer and the output layer is initialized, with dimension N × V.
Next, the hidden layer vector h (dimension 1 × N) is multiplied by w '(dimension N × V) to obtain a vector u of 1 × V, where u is h · w'.
For convenience of probabilistic representation, the vector u is passed through softmax, where each dimension of the vector softmax (u) represents an intent in the corpus. The vector softmax (u) represents the most probable location as the intermediate intent predicted by the model.
Finally, the 1 × V vector output from the previous step is compared with one hot in the group route.
The aim of the training is to maximize the probability of the actual center intention occurring, based on which a loss function is defined, by minimizing the loss function, updating W and W' with a gradient descent algorithm. When the network converges, the training is completed, and the matrix W is the intent vector to be predicted.
When a vector representation of a certain intention in the corpus is to be obtained, multiplying the one-hot of the intention by a weight matrix w to obtain a vector of 1 × N, wherein the vector is the vector representation of the intention.
Since only one position in the intended one-hot representation is 1 and the rest are 0, then after multiplication with w, a certain column of vectors in w is obtained. Since the position of 1 in one-hot of each intention is different, one-hot of different intentions is multiplied by w, and the resultant intention vector is a different column vector in w.
It can be seen that each column vector in the weight matrix can correspondingly and uniquely represent each intention in the corpus. The desired intent vector is the weight matrix between the neural network input layer and the hidden layer.
By training the obtained customer intention model, vector representation can be obtained for each intention, and addition and subtraction between vectors can obtain representation of a customer intention path.
Based on this, the vector representation of the successful intent sequence can be obtained by adding the customer intent vectors of the successful business order. The successful customer intent vector is represented by Si, where i represents the ith successful intent sequence. And retaining Si in memory.
In particular, if we have M candidate intentions in the nth round of interaction, we will select the intention with the smallest average cosine distance. The average cosine distance is calculated by comparing the average cosine distance between the vector of N + m1 and N + m2 … N + mm and Si.
Finally, choosing the minimum cosine distance of the customer intent, since Si is from a successful sequence of intentions, we can increase the probability of success to be single. That is, it is more likely to select "definite positive" among the choices of "busy" and "definite positive".
The customer intention analysis method can model a customer intention sequence, and can select the intention which is more favorable for promoting the business effect when predicting on an actual line. By comparing vectors of different customer intentions, the stream of customer intentions can be clustered to find similar customer-attention for more efficient and targeted sales.
In addition, the vector of the intention is calculated off-line, so that the method has extremely high actual landing value, does not increase the burden of a system, can perform effective feature calculation with extremely low calculation cost, and simultaneously integrates more feature information.
According to the method for analyzing the intention of the customer in the embodiment of the application, intention classification is carried out on the conversation interactive information of the customer to obtain a plurality of intention sequences of a plurality of customers; the sequence of intentions includes a plurality of customer intentions arranged in time; embedding a plurality of intention sequence training words of a plurality of clients into a network to obtain a client intention model; inputting at least one time of interaction information of a customer to be predicted into a customer intention model to obtain a plurality of customer intention vectors; calculating cosine distances between the plurality of customer intention vectors and the success intention vectors, comparing the cosine distances, and selecting the customer intention vector corresponding to the smallest cosine distance; and predicting the result for the client intention according to the obtained client intention.
According to the method and the device, the client intention modeling is carried out according to the client intention sequence, so that the accuracy of the client intention can be improved, and the business can be predicted in a single direction.
Example 2
For details not disclosed in the customer intention analysis system of this embodiment, please refer to specific implementation contents of the customer intention analysis method in other embodiments.
A schematic structural diagram of a customer intention analysis system according to an embodiment of the present application is shown in fig. 4.
As shown in fig. 4, the customer intention analysis system specifically includes:
the client intention classification module 10: the system comprises a plurality of users, a plurality of communication terminals and a plurality of communication terminals, wherein the communication terminals are used for receiving the communication interaction information of the users; the sequence of intentions includes a plurality of customer intents arranged in time.
The method comprises the following steps of identifying the intention of a customer by using the conversation interactive information of the customer to obtain a plurality of intention sequences of a plurality of customers, and specifically comprises the following steps: firstly, when the client call interaction information is voice, converting the client call interaction information into character interaction information; then, identifying the intention of the client through a classification model of the intention of the client to obtain a plurality of intention sequences; finally, the intention of the clients is identified according to the conversation interactive information of the clients, and a plurality of intention sequences of the clients are obtained.
Further, before the client intention classification model identifies the client intention and obtains a plurality of intention sequences, the method further includes: acquiring a client intention data set; and inputting the client intention data set into a client intention classification network for training to obtain a client intention classification model.
The customer intent model module 20: the method is used for embedding the training words into the network through a plurality of intention sequences of a plurality of clients to obtain a client intention model.
The method includes the following steps that a plurality of intention sequence training words of a plurality of clients are embedded into a network to obtain a client intention model, and specifically includes the following steps:
1) summarizing the plurality of intention sequences to obtain an intention directed graph; the nodes of the intention directed graph are client intents, and the weight of the edges of the intention directed graph is the probability of transition between the nodes of the intention directed graph;
specifically, the calculation formula of the probability of transition between nodes of the intention directed graph is as follows:
Pmn=Cn/Cm;
where Cm is the number of intended sequences on the intended directed graph node m that pass through m but are not the end point, Cn is the number of intended sequences that pass from the intended directed graph node m to the intended directed graph node n, and Pmn is the probability of a transition between the intended directed graph node m and the intended directed graph node n.
2) And traversing the intention directed graph to obtain a plurality of random intention sequences, and embedding training words into the network according to the plurality of random intention sequences to obtain a client intention model.
Specifically, in some embodiments of the present application, the Word embedding network is a Word2vec model, the Word2vec model combines a CBOW model and a skip-gram model, and adopts a negative sampling and hierarchical softmax method.
The to-be-predicted customer intention vector module 30: the system is used for inputting at least one time of interaction information of a customer to be predicted into a customer intention model to obtain a plurality of customer intention vectors.
The customer intent prediction module 40: the system comprises a plurality of client intention vectors, a plurality of success intention vectors and a plurality of cosine distances, wherein the client intention vectors are used for calculating and comparing cosine distances between the plurality of client intention vectors and the success intention vectors, and the client intention vector corresponding to the smallest cosine distance is selected; and predicting the result for the client intention according to the obtained client intention.
Preferably, after obtaining the client intention model by embedding the training words into the network through the multiple intention sequences of multiple clients, the method further includes: and inputting a sequence of the client intention of the client in order success to the client intention model to obtain a success intention vector.
In the customer intention analysis system in the embodiment of the application, a customer intention classification module 10 identifies customer intentions according to the conversation interaction information of the customers to obtain a plurality of intention sequences of the customers; the sequence of intentions includes a plurality of customer intentions arranged in time; the client intention model module 20 embeds a plurality of intention sequence training words of a plurality of clients into a network to obtain a client intention model; the to-be-predicted customer intention vector module 30 inputs at least one piece of interaction information of the to-be-predicted customer into the customer intention model to obtain a plurality of customer intention vectors; the customer intention predicting module 40 calculates cosine distances between a plurality of customer intention vectors and success intention vectors, compares the cosine distances and selects a customer intention vector corresponding to the smallest cosine distance; the corresponding obtained client intention is a client intention prediction result; the success intention vector is a pre-stored success intention vector of a service or a single client.
According to the method and the device, the client intention modeling is carried out according to the client intention sequence, so that the accuracy of the client intention can be improved, and the business can be predicted in a single direction.
Example 3
For details that are not disclosed in the apparatus for analyzing a client intention of this embodiment, please refer to specific implementation contents of a method or a system for analyzing a client intention in other embodiments.
A schematic structural diagram of a customer intention analysis device 400 according to an embodiment of the present application is shown in fig. 5.
As shown in fig. 5, the customer intention analyzing apparatus 400 includes:
the memory 402: for storing executable instructions; and
a processor 401 is coupled to the memory 402 to execute executable instructions to perform the motion vector prediction method.
Those skilled in the art will appreciate that the schematic diagram 5 is merely an example of the customer intention analysis device 400 and does not constitute a limitation of the customer intention analysis device 400 and may include more or less components than those shown, or combine certain components, or different components, e.g., the customer intention analysis device 400 may also include input output devices, network access devices, buses, etc.
The Processor 401 (CPU) may be other general-purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA) or other Programmable logic device, a discrete Gate or transistor logic device, a discrete hardware component, or the like. The general purpose processor may be a microprocessor or the processor 401 may be any conventional processor or the like, the processor 401 being the control center of the customer intention analyzing apparatus 400 and the various parts of the entire customer intention analyzing apparatus 400 being connected by various interfaces and lines.
The memory 402 may be used to store computer readable instructions and the processor 401 may implement the various functions of the customer intent analysis device 400 by executing or executing computer readable instructions or modules stored in the memory 402 and by invoking data stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data created according to the use of the customer intention analysis apparatus 400, and the like. In addition, the Memory 402 may include a hard disk, a Memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash Memory Card (Flash Card), at least one disk storage device, a Flash Memory device, a Read-Only Memory (ROM), a Random Access Memory (RAM), or other non-volatile/volatile storage devices.
The modules integrated by the client intention analyzing apparatus 400 may be stored in a computer-readable storage medium if they are implemented in the form of software functional modules and sold or used as separate products. Based on such understanding, all or part of the flow of the method according to the embodiments of the present invention may also be implemented by hardware related to computer readable instructions, which may be stored in a computer readable storage medium, and when the computer readable instructions are executed by a processor, the steps of the method embodiments may be implemented.
Example 4
The present embodiment provides a computer-readable storage medium having stored thereon a computer program; the computer program is executed by the processor to implement the customer intention analysis method in other embodiments.
According to the client intention analysis equipment and the computer storage medium in the embodiment of the application, the client intention is identified according to the client call interaction information, and a plurality of intention sequences of a plurality of clients are obtained; the sequence of intentions includes a plurality of customer intentions arranged in time; embedding a plurality of intention sequence training words of a plurality of clients into a network to obtain a client intention model; inputting at least one time of interaction information of a customer to be predicted into a customer intention model to obtain a plurality of customer intention vectors; calculating cosine distances between the plurality of customer intention vectors and the success intention vectors, comparing the cosine distances, and selecting the customer intention vector corresponding to the smallest cosine distance; the corresponding obtained client intention is a client intention prediction result; the success intention vector is a pre-stored success intention vector of a service or a single client. According to the method and the device, the client intention modeling is carried out according to the client intention sequence, so that the accuracy of the client intention can be improved, and the business can be predicted in a single direction.
According to the method and the device, the client intention modeling is carried out according to the client intention sequence, so that the accuracy of the client intention can be improved, and the business can be predicted in a single direction.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in this specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.
It is to be understood that although the terms first, second, third, etc. may be used herein to describe various information, these information should not be limited to these terms. These terms are only used to distinguish one type of information from another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of the present invention. The word "if" as used herein may be interpreted as "at … …" or "when … …" or "in response to a determination", depending on the context.
While the preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all alterations and modifications as fall within the scope of the application.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present application without departing from the spirit and scope of the application. Thus, if such modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include such modifications and variations as well.
Claims (10)
1. A customer intention analysis method, characterized by comprising the steps of:
identifying the intention of the client according to the conversation interaction information of the client to obtain a plurality of intention sequences of a plurality of clients; the sequence of intentions includes a plurality of customer intentions arranged in time;
embedding a plurality of intention sequence training words of the plurality of clients into a network to obtain a client intention model;
inputting at least one time of interaction information of a customer to be predicted into the customer intention model to obtain a plurality of customer intention vectors;
calculating cosine distances between the plurality of customer intention vectors and success intention vectors, comparing the cosine distances, and selecting the customer intention vector corresponding to the smallest cosine distance; the corresponding obtained client intention is a client intention prediction result; the success intention vector is a pre-stored success intention vector of a service or a single client.
2. The method for analyzing the intention of the customer as claimed in claim 1, wherein the obtaining of the model of the intention of the customer by embedding the training words into the network through the plurality of intention sequences of the customers specifically comprises:
arranging the plurality of intention sequences to obtain an intention directed graph; the intention directed graph nodes are client intents, and the weights of the edges of the intention directed graph are probabilities of transition among the intention directed graph nodes;
and traversing the intention directed graph to obtain a plurality of random intention sequences, and embedding training words into a network according to the plurality of random intention sequences to obtain a client intention model.
3. The customer intent analysis method according to claim 2, wherein the probability of transition between the intent directed graph nodes is calculated by the formula:
Pmn=Cn/Cm;
where Cm is the number of intended sequences on the intended directed graph node m that pass through m but are not the end point, Cn is the number of intended sequences that pass from the intended directed graph node m to the intended directed graph node n, and Pmn is the probability of a transition between the intended directed graph node m and the intended directed graph node n.
4. The method for analyzing client intention according to claim 1, wherein after obtaining the client intention model by embedding the training words into the network through the plurality of intention sequences of the plurality of clients, the method further comprises:
and inputting a sequence of client intention of the client to the client intention model to obtain a success intention vector.
5. The client intention analysis method of claim 1, wherein the Word embedding network is a Word2vec model, and the Word2vec model adopts a CBOW model and a skip-gram model in combination, and adopts a negative sampling and hierarchical softmax method.
6. The method for analyzing customer intention according to claim 1, wherein the identifying the customer intention from the customer call interaction information to obtain a plurality of intention sequences of a plurality of customers specifically comprises:
when the customer call interaction information is voice, converting the customer call interaction information into character interaction information;
identifying the intention of the client through a classification model of the intention of the client to obtain a plurality of intention sequences;
through the steps, the intention of the clients is identified according to the conversation interaction information of the clients, and a plurality of intention sequences of the clients are obtained.
7. The method of claim 6, wherein before the identifying the client intention through the client intention classification model to obtain the plurality of intention sequences, the method further comprises:
acquiring a client intention data set;
and inputting the customer intention data set into a customer intention classification network for training to obtain the customer intention classification model.
8. A customer intention analysis device is characterized by specifically comprising:
a client intent classification module: the system is used for identifying the intention of the customer according to the conversation interaction information of the customer to obtain a plurality of intention sequences of a plurality of customers; the sequence of intentions includes a plurality of customer intentions arranged in time;
a customer intent model module: the method comprises the steps of embedding a plurality of intention sequence training words of a plurality of clients into a network to obtain a client intention model;
a to-be-predicted customer intention vector module: the system comprises a client intention model, a plurality of client intention vectors and a plurality of data processing units, wherein the client intention model is used for inputting at least one piece of interaction information of a client to be predicted to the client intention model to obtain a plurality of client intention vectors;
a client intent prediction module: the system is used for calculating cosine distances between the plurality of customer intention vectors and success intention vectors, comparing the cosine distances, and selecting the customer intention vector corresponding to the smallest cosine distance; the corresponding obtained client intention is a client intention prediction result; the success intention vector is a pre-stored success intention vector of a service or a single client.
9. A customer intention analysis apparatus characterized by comprising:
a memory: for storing executable instructions; and
a processor for interfacing with the memory to execute the executable instructions to perform the customer intent analysis method of any of claims 1-7.
10. A computer-readable storage medium, having stored thereon a computer program; the computer program is executed by a processor to implement the customer intention analysis method according to any one of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210039128.3A CN114416985A (en) | 2022-01-13 | 2022-01-13 | Customer intention analysis method, system, device and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210039128.3A CN114416985A (en) | 2022-01-13 | 2022-01-13 | Customer intention analysis method, system, device and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114416985A true CN114416985A (en) | 2022-04-29 |
Family
ID=81274062
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210039128.3A Pending CN114416985A (en) | 2022-01-13 | 2022-01-13 | Customer intention analysis method, system, device and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114416985A (en) |
-
2022
- 2022-01-13 CN CN202210039128.3A patent/CN114416985A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111291183B (en) | Method and device for carrying out classification prediction by using text classification model | |
CN108346436B (en) | Voice emotion detection method and device, computer equipment and storage medium | |
KR101699720B1 (en) | Apparatus for voice command recognition and method thereof | |
JP2022525702A (en) | Systems and methods for model fairness | |
CN110377916B (en) | Word prediction method, word prediction device, computer equipment and storage medium | |
CN108062954B (en) | Speech recognition method and device | |
JP2775140B2 (en) | Pattern recognition method, voice recognition method, and voice recognition device | |
KR20190129580A (en) | Device and method to personlize voice recognition model | |
CN112435673B (en) | Model training method and electronic terminal | |
KR102133825B1 (en) | Voice conversation method and system of enhanced word features | |
CN111274789B (en) | Training method and device of text prediction model | |
CN111783474A (en) | Comment text viewpoint information processing method and device and storage medium | |
JP6892606B2 (en) | Positioning device, position identification method and computer program | |
CN110717027B (en) | Multi-round intelligent question-answering method, system, controller and medium | |
CN110008332B (en) | Method and device for extracting main words through reinforcement learning | |
KR102543698B1 (en) | Computing system and method for data labeling thereon | |
CN111199149B (en) | Sentence intelligent clarification method and system for dialogue system | |
KR20190136578A (en) | Method and apparatus for speech recognition | |
CN113537630A (en) | Training method and device of business prediction model | |
CN115130711A (en) | Data processing method and device, computer and readable storage medium | |
CN112036954A (en) | Item recommendation method and device, computer-readable storage medium and electronic device | |
CN111557010A (en) | Learning device and method, and program | |
CN114492601A (en) | Resource classification model training method and device, electronic equipment and storage medium | |
CN114692972A (en) | Training method and device of behavior prediction system | |
CN113806501B (en) | Training method of intention recognition model, intention recognition method and equipment |
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 |