CN113239147A - Intelligent conversation method, system and medium based on graph neural network - Google Patents
Intelligent conversation method, system and medium based on graph neural network Download PDFInfo
- Publication number
- CN113239147A CN113239147A CN202110516335.9A CN202110516335A CN113239147A CN 113239147 A CN113239147 A CN 113239147A CN 202110516335 A CN202110516335 A CN 202110516335A CN 113239147 A CN113239147 A CN 113239147A
- Authority
- CN
- China
- Prior art keywords
- client
- nodes
- neural network
- semantic
- session
- 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
- 238000000034 method Methods 0.000 title claims abstract description 95
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 51
- 238000012549 training Methods 0.000 claims abstract description 53
- 238000003062 neural network model Methods 0.000 claims abstract description 52
- 239000011159 matrix material Substances 0.000 claims abstract description 25
- 230000008451 emotion Effects 0.000 claims description 38
- 230000002996 emotional effect Effects 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 13
- 238000003860 storage Methods 0.000 claims description 12
- 238000004458 analytical method Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 19
- 238000005516 engineering process Methods 0.000 description 13
- 230000006870 function Effects 0.000 description 9
- 238000009826 distribution Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 7
- 230000004044 response Effects 0.000 description 7
- 238000004891 communication Methods 0.000 description 6
- 238000003058 natural language processing Methods 0.000 description 6
- 238000009432 framing Methods 0.000 description 5
- 238000012986 modification Methods 0.000 description 5
- 230000004048 modification Effects 0.000 description 5
- 125000004122 cyclic group Chemical group 0.000 description 4
- 238000000605 extraction Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000026676 system process Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 206010028813 Nausea Diseases 0.000 description 1
- 206010067171 Regurgitation Diseases 0.000 description 1
- 206010047700 Vomiting Diseases 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000008693 nausea Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 210000001260 vocal cord Anatomy 0.000 description 1
- 230000008673 vomiting Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Human Computer Interaction (AREA)
- Computing Systems (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Multimedia (AREA)
- Acoustics & Sound (AREA)
- Machine Translation (AREA)
Abstract
The application provides an intelligent conversation method and system based on a graph neural network, which are used for acquiring client semantic information, client application dialogues and client order forming results according to client conversation text data; converting to obtain vector-form customer semantic nodes, customer corresponding dialogue technique nodes, customer single-forming nodes and an adjacent matrix among the nodes through a pre-training language model according to customer semantic information, customer corresponding dialogue techniques and customer single-forming results; inputting the client semantic nodes, the client corresponding dialogue technique nodes, the client single-forming nodes and the adjacency matrixes among the nodes into a graph neural network model for training to obtain the trained graph neural network model; and inputting the current session data into the trained graph neural network model to obtain a corresponding session path. The method analyzes the client phonetics while talking with the client, and predicts the optimal path with the single as the terminal point by combining different phonetics paths of the intelligent robot.
Description
Technical Field
The present application belongs to the field of intelligent session technologies, and in particular, to an intelligent session method, system and medium based on a graph neural network.
Background
The intelligent robot outbound system is a practical application of natural language processing technology, and has been widely applied to service scenes such as marketing, return visit, customer service and the like. Compared with the traditional manual outbound, the intelligent robot outbound system has the advantages of lower operation and management cost, higher working efficiency and the like, and can be close to or even better than the manual outbound in the outbound effect, especially the screening of the intention customers in the electricity sales scene. Therefore, in the field of electric marketing, an industrial mode that customers with stronger intention are extensively called out and screened by using an intelligent robot calling-out system and then follow-up is carried out by a manual seat to form a list is formed.
The existing intelligent calling robot mostly adopts a mode of combining automatic speech recognition ASR, natural language processing NLP and text-to-speech TTS, namely, firstly, speech of a client is converted into text data through ASR, then the text is processed through NLP technology, intention tendency of the client is deduced according to content of the text, corresponding answering dialogues are selected from a corpus, and then the answering dialogues are converted into speech information through TTS technology for broadcasting.
At present, in the process of selecting the dialogs, after the existing system processes the voices of the clients and matches the voices with corresponding features, the robot dialogs are arranged according to the inherent flow worksheets and the dialogs knowledge base, the dialogs are relatively rigid, and the intentions and emotional tendencies of the clients cannot be flexibly changed to enhance the intentions of the potential clients to promote the final completion of the dialogs.
On the other hand, in the prior art, a fixed flow work order and a corpus are mostly adopted to be matched with the traditional machine learning method or the artificial neural network and other technologies, so that the real-time human-machine communication can be realized, but the effect of predicting and constructing structural data such as a conversation strategy and a path is often poor.
Disclosure of Invention
The invention provides an intelligent conversation method, an intelligent conversation system and an intelligent conversation medium based on a graph neural network, and aims to solve the problem that the existing intelligent answer dialogue is relatively rigid and cannot be flexibly changed according to the intention and emotional tendency of a client.
According to a first aspect of the embodiments of the present application, there is provided an intelligent session method based on a graph neural network, specifically including the following steps:
according to the client session text data, acquiring client semantic information, client dialogues and client ordering results;
converting to obtain vector-form customer semantic nodes, customer corresponding dialogue technique nodes, customer single-forming nodes and an adjacent matrix among the nodes through a pre-training language model according to customer semantic information, customer corresponding dialogue techniques and customer single-forming results;
inputting the client semantic nodes, the client corresponding dialogue technique nodes, the client single-forming nodes and the adjacency matrixes among the nodes into a graph neural network model for training to obtain the trained graph neural network model;
and inputting the current session data into the trained graph neural network model to obtain a corresponding session path.
In some embodiments of the present application, obtaining semantic information of a client, a dialogues for the client, and a result of ordering the client according to the text data of a session of the client specifically includes:
according to the client session text data, dividing the client session text data into client side session data and responder side session data through character analysis;
taking the responder session data as the client application dialogue;
acquiring client semantic information according to client side session data;
and searching the client list-forming keywords according to the client side conversation data and/or the responder side conversation data, and taking the list-forming result corresponding to the client list-forming keywords as the client list-forming result.
In some embodiments of the present application, the client semantic information includes client intent information and/or client sentiment information; the client intent includes intent to be a single type, intent to be a singular amount, and intent to be a single time.
In some embodiments of the present application, the obtaining of the client intention information specifically includes:
a client intention language library is established in advance, wherein the client intention language library comprises a plurality of keywords, and each keyword corresponds to intention information;
and searching whether the keywords of the client intention library exist one by one according to the client side conversation data, and taking intention information corresponding to the keywords as client intention information when the keywords exist.
In some embodiments of the present application, the obtaining of the client emotion information specifically includes:
recognizing the conversation data of the client side by using a preset voice excitement recognition model to obtain the emotion excitement of the client;
and obtaining the emotional information of the client according to the corresponding relation between the emotional excitement of the client and the emotion of the client.
In some embodiments of the present application, the converting, according to the client semantic information, the client corresponding dialogues and the client singleton result, the pre-training language model to obtain the client semantic nodes, the client corresponding dialogues nodes, the client singleton nodes and the adjacency matrixes among the nodes in the vector form specifically includes:
according to the client semantic information, the client corresponding dialogues and the client list forming results, a client semantic feature vector, a client corresponding dialogues feature vector and a client list forming feature vector are obtained through a neural network model;
respectively taking the client semantic feature vector, the client dialoging feature vector and the client unigram feature vector as a client semantic node, a client dialoging node and a client unigram node;
and connecting the client semantic nodes, the client corresponding dialogue technical nodes and the client single nodes according to the session corresponding relation to obtain an adjacency matrix among the nodes.
In some embodiments of the present application, before obtaining the semantic information of the client, the dialogues of the client and the results of the unionizing of the client according to the text data of the client session, the method further comprises
Acquiring a client conversation voice sample;
the customer voice samples are converted to customer session text data by speech recognition techniques.
According to a second aspect of the embodiments of the present application, there is provided an intelligent conversational system based on a graph neural network, specifically including:
a client text processing module: the system is used for acquiring client semantic information, client dialogues and client ordering results according to client session text data;
pre-training a language model module: the system comprises a pre-training language model, a client semantic node, a client corresponding dialogue technique node, a client single node and an adjacency matrix among the nodes, wherein the pre-training language model is used for converting to obtain a vector form of the client semantic node, the client corresponding dialogue technique node, the client single node and the adjacency matrix among the nodes according to client semantic information, a client corresponding dialogue technique and a client single result;
the graph neural network training module: the system comprises a graph neural network model, a semantic node, an dialoging node, a client single-forming node and an adjacency matrix among the nodes, wherein the semantic node, the dialoging node, the client single-forming node and the adjacency matrix among the nodes are used for inputting the semantic node, the dialoging node, the client single-forming node and the adjacency matrix among the nodes of the client into the graph neural network model for training to obtain the trained graph neural network model;
a session module: and the method is used for inputting the current session data into the trained graph neural network model to obtain a corresponding session path.
According to a third aspect of the embodiments of the present application, there is provided an intelligent session device based on a graph neural network, including:
a memory: for storing executable instructions; and
the processor is used for connecting with the memory to execute the executable instructions so as to complete the intelligent session method based on the graph neural network.
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 graph neural network based intelligent conversational method.
By adopting the intelligent session method and system based on the graph neural network in the embodiment of the application, client semantic information, client dialogues and client ordering results are obtained according to client session text data; converting to obtain vector-form customer semantic nodes, customer corresponding dialogue technique nodes, customer single-forming nodes and an adjacent matrix among the nodes through a pre-training language model according to customer semantic information, customer corresponding dialogue techniques and customer single-forming results; inputting the client semantic nodes, the client corresponding dialogue technique nodes, the client single-forming nodes and the adjacency matrixes among the nodes into a graph neural network model for training to obtain the trained graph neural network model; and inputting the current session data into the trained graph neural network model to obtain a corresponding session path. After the optimal conversation path prediction technology is adopted, through training of a large amount of conversation data, the robot analyzes the conversation of a client while talking with the client, and an optimal path with a list as a terminal point is predicted by combining different conversation paths of the intelligent robot.
Meanwhile, the conversation quality of the intelligent robot is greatly improved, the customer satisfaction is improved, the robot can screen the purchase intention of the customer while carrying out conversation communication, the order forming of the customer with high intention is facilitated, the willingness of the customer with no intention and the willingness of the customer with low intention are improved, and the maximization of the final order forming possibility is realized.
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:
a schematic diagram of the steps of a graph neural network based intelligent conversational method according to an embodiment of the application is shown in fig. 1;
a flow diagram of a graph neural network based intelligent conversational method according to an embodiment of the application is shown in fig. 2;
a schematic diagram of the neural network of the present application in accordance with an embodiment of the present application is shown in fig. 3;
a schematic structural diagram of an intelligent session system based on a graph neural network according to an embodiment of the present application is shown in fig. 4;
a schematic structural diagram of an intelligent session device based on a graph neural network according to an embodiment of the present application is shown in fig. 5.
Detailed Description
In the process of implementing the application, the inventor finds that in the process of selection for dialogs, after the existing system processes the voice of a client and matches the voice with corresponding characteristics, the robot dialogs are often arranged according to an inherent flow worksheet and a dialogs knowledge base, the dialogs are relatively rigid, and flexible changes cannot be made according to the intention and emotional tendency of the client to enhance the intention of a potential client to cause the problem of final singleness.
Under the condition, the existing technology mostly adopts the fixed flow work order and the corpus to be matched with the traditional machine learning method or the artificial neural network and other technologies, so that the real-time communication of man-machine can be realized, but the effect of predicting and constructing the structural data such as the conversation strategy and the path is not good.
According to the method and the system, the robot analyzes the client phonetics while talking with the client through training of a large amount of talking data, and predicts the optimal path taking the single as the terminal point by combining different phonetics paths of the intelligent robot.
In particular, the method comprises the following steps of,
the method comprises the steps of obtaining client semantic information, client corresponding dialogues and client ordering results according to client session text data by adopting an intelligent session method and system based on a graph neural network; converting to obtain vector-form customer semantic nodes, customer corresponding dialogue technique nodes, customer single-forming nodes and an adjacent matrix among the nodes through a pre-training language model according to customer semantic information, customer corresponding dialogue techniques and customer single-forming results; inputting the client semantic nodes, the client corresponding dialogue technique nodes, the client single-forming nodes and the adjacency matrixes among the nodes into a graph neural network model for training to obtain the trained graph neural network model; and inputting the current session data into the trained graph neural network model to obtain a corresponding session path.
According to the method and the system, the conversation quality of the intelligent robot is greatly improved, the customer satisfaction is improved, the robot can screen the purchase intention of the customer while carrying out conversation communication, the order of the customer with high intention is formed, the willingness of the customer with no intention and the customer with low intention is improved, and the maximization of the final order forming possibility is realized.
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 schematic diagram of steps of a graph neural network-based intelligent conversational method according to an embodiment of the application is shown in fig. 1. A flowchart of an intelligent conversational method based on a graph neural network according to an embodiment of the application is shown in fig. 2.
As shown in fig. 1, the intelligent session method based on the graph neural network in the embodiment of the present application specifically includes the following steps:
s101: and acquiring client semantic information, client dialogues and client ordering results according to the client session text data.
First, a large amount of client session data is acquired, and voice of the client session is converted into client session text data through voice recognition.
The speech-to-text technique in the present embodiment employs an ASR technique.
Among them, ASR (Automatic speech recognition) is a technology for converting human speech into text.
Specifically, it is first required to perform pre-emphasis, framing, and windowing on the original speech to obtain the standard speech.
The process of acquiring the standard voice is as follows:
firstly, the original voice is pre-emphasized by adopting a formula s' n-sn-a sn-1 of pre-emphasis processing, so that the influence of vocal cords and lips of a speaker on the voice of the speaker is eliminated, and the high-frequency resolution of the voice of the speaker is improved.
Wherein s' n is the voice signal amplitude at n moment after pre-emphasis processing, sn is the voice signal amplitude at n moment, sn-1 is the voice signal amplitude at n-1 moment, and a is a pre-emphasis coefficient.
Then, the pre-emphasized original speech is subjected to framing processing, and when framing is performed, discontinuous places appear at the starting point and the end point of each frame of speech, and the more framing, the larger the error with the original speech.
Finally, in order to maintain the frequency characteristic of each frame of speech, windowing is also needed, and the formula of windowing is as follows
And s "N ═ wn × s 'N, where wn is the hamming window at time N, N is the hamming window length, s' N is the signal amplitude in the time domain at time N, and s" N is the signal amplitude in the time domain after the time N has been windowed.
The original customer voice is preprocessed through the steps to obtain the standard voice, and an effective data source is provided for the subsequent endpoint detection processing of the standard voice.
Then, according to the text data of the client conversation, the voice information of the client is obtained through the NLP model, and the voice information comprises information of the intention and emotional tendency of the client.
Meanwhile, the information of the result of the dialogs which the client should use and the corresponding client account result is obtained according to the historical conversation data.
According to the client session text data, acquiring client semantic information, client dialogues and client ordering results, and specifically comprising the following steps:
first, the client session text data is divided into client side session data and responder session data by character analysis based on the client session text data.
Then, taking the session data of the responder as the client application dialogue; and acquiring client semantic information according to the client session data. And simultaneously, searching a client list forming keyword according to the client side conversation data and/or the responder side conversation data, and taking a list forming result corresponding to the client list forming keyword as a client list forming result.
In this embodiment, the client semantic information specifically includes client intention information and/or client emotion information; the customer intent information, in turn, includes intent-to-singleton type, intent-to-singleton amount, and intent-to-singleton time.
The client intention information can be obtained by establishing a client intention language library in advance, and the client emotion information is obtained by a language library of client emotion or simultaneously detecting the emotional excitement of the client.
Specifically, the obtaining of the client intention information specifically includes:
1) a client intention language library is established in advance, the client intention language library comprises a plurality of key words, and each key word corresponds to intention information.
2) And searching whether the keywords of the client intention library exist one by one according to the client side conversation data, and taking intention information corresponding to the keywords as client intention information when the keywords exist.
Specifically, the method for obtaining the client emotion information specifically comprises the following steps:
firstly, recognizing the conversation data of the client side by using a preset voice excitement recognition model to obtain the emotion excitement of the client. And then, obtaining the emotional information of the client according to the corresponding relation between the emotional excitement of the client and the emotion of the client.
Specifically, the emotional excitement of the client needs to be recognized through a voice excitement recognition model.
Further, when the voice contrastiveness recognition model is used for voice contrastiveness recognition, the specific training process of the voice contrastiveness recognition model is as follows:
firstly, original voice information of a client is converted into a spectrogram, and spectrogram feature extraction is carried out, and compared with other feature extraction algorithms, the spectrogram contains more original voice information including time, frequency, amplitude and other information.
Then, training and testing a model, and training a deep neural network model by using the matched input data and the corresponding contrastive degree class label; the trained model predicts unlabeled multi-modal input data, and finally the probability value of the excitement prediction can be obtained according to the trained model.
Specifically, the client emotion information is acquired, and on the other hand, the client emotion information can be acquired through a language library of client emotions. For example, the emotion of the customer includes complaints, anger, disgust or urgency, and the like, and each speech emotion is expressed by various words.
Therefore, firstly, a keyword and speech emotion corresponding library is established; then, identifying keywords in an emotion library according to the character identification data; and finally, obtaining the voice emotion corresponding to the keyword.
For example:
keywords corresponding to complaints of emotions include: bad, junk, what service, no longer, complaint, awkward, and so on.
Keywords corresponding to angry emotions include: dying by thoughts, dying by heat, smelly, rotten and specially rotten.
The keywords corresponding to aversive emotions include: nausea, vomiting, regurgitation, confusion, poor quality, etc.
The keywords corresponding to the urgent emotions include: immediate, bad, fast, urgent, and dying.
S102: and converting to obtain vector-form customer semantic nodes, customer dialogues nodes, customer singleness nodes and an adjacency matrix among the nodes through a pre-training language model according to the customer semantic information, the customer dialogues and the customer singleness results.
Specifically, the method comprises the following steps:
a) and obtaining a client semantic feature vector, a client dialogues feature vector and a client singleton feature vector through a neural network model according to the client semantic information, the client dialogues and the client singleton result.
b) And respectively taking the client semantic feature vector, the client dialoging feature vector and the client unionized feature vector as a client semantic node, a client dialoging node and a client unionized node.
c) And connecting the client semantic nodes, the client corresponding dialogue technical nodes and the client single nodes according to the session corresponding relation to obtain an adjacency matrix among the nodes.
For a particular application, for example, the conversation history { U1, U2, U3.. UM } consists of M words. For each sentence Ui, converting each word in the Ui into a corresponding word vector by using a pre-training Glove method to obtain a vector sequence, sequentially inputting the vector sequence into a lower-layer cyclic neural network, and outputting the final state of the network as a feature vector hi of the current sentence to further obtain a feature code { h1, h2, h 3.. hM }. And inputting the data into an upper-layer cyclic neural network for further coding to obtain a feature vector hD of a conversation level, and finally taking the feature vector hD as a semantic node X of the whole conversation.
And then, constructing an adjacency matrix A between the nodes according to the real data.
In some embodiments of the present application, the method further includes obtaining a client ordering result, where the client ordering result corresponds to the client-based dialog one-to-one.
S103: and inputting the client semantic nodes, the client corresponding dialogue technique nodes, the client single-forming nodes and the adjacency matrixes among the nodes into the graph neural network model for training to obtain the trained graph neural network model.
S104: and inputting the current session data into the trained graph neural network model to obtain a corresponding session path.
According to the client response dialogue historical information, a heteromorphic graph containing client semantic nodes and client response dialogue technical nodes can be constructed, and an optimal conversation response path can be predicted.
A schematic diagram of a neural network according to an embodiment of the present application is shown in fig. 3.
The graph neural network model GNN comprises an encoder and a decoder, wherein input data of the encoder are client semantic nodes, client dialogues nodes and adjacency matrixes among the nodes in a vector form, and output data of the encoder is embedded in a graph; the input data of the decoder is graph embedding, the output data of the decoder is a reconstructed graph, and the reconstructed graph is a corresponding conversation path, namely a conversation path graph.
As shown in fig. 3, the graph neural network model for path connection prediction uses a variational graph self-encoder (VGAE), the network is divided into an encoder and a decoder, the encoder uses a graph formed by graph nodes X and an adjacent matrix a as input data, the output is graph embedding Z, and the decoder uses the graph embedding Z as input, and the output is a reconstructed graph.
Specifically, the mean and variance of the graph-embedded multidimensional gaussian distribution are obtained by passing input data through two multi-layer Graph Convolution Networks (GCNs):
μ＝GCNμ(X,A)；logσ＝GCNσ(X,A)；
wherein X is a client semantic node, a client dialoging node and/or a single result node; a is the adjacency matrix between the nodes.
After obtaining the mean value mu and the variance mu, sampling from the standard Gaussian normal distribution to obtain an element, and embedding the graph obtained by the output of the encoder into a value Z which is mu plus the element sigma; at reconstruction, the decoder is the inner product of the graph embedding, i.e.
Specifically, in the case of the trainer, the graph reconstructed by the variational graph autoencoder VGAE is desired to be as close as possible to the original graph, and the distribution q (Z | X, a) calculated by the multi-layer graph convolution network GCN is desired to be as similar as possible to the standard gaussian distribution p (Z), so that the loss function
From cross entropy
And the relative entropy KL divergence, and the specific calculation mode is as follows:
finally, obtaining a trained graph neural network model, wherein output data obtained through the graph neural network processing is a graph formed by the connection relation among the nodes, wherein the graph may comprise a plurality of paths leading to the unijunction result from the current session state, and the model can output an optimal path leading to the unijunction result from the current session state through the training and the super-parameter fine adjustment of the graph neural network model by using massive training samples.
And finally, generating an optimal path of a subsequent conversation according to model prediction, and selecting the conversation to communicate with the client by the robot according to the predicted optimal path, thereby improving the probability of singleness as much as possible.
According to the intelligent session method based on the graph neural network, client semantic information, client application dialogues and client ordering results are obtained according to client session text data; converting to obtain vector-form customer semantic nodes, customer corresponding dialogue technique nodes, customer single-forming nodes and an adjacent matrix among the nodes through a pre-training language model according to customer semantic information, customer corresponding dialogue techniques and customer single-forming results; inputting the client semantic nodes, the client corresponding dialogue technique nodes, the client single-forming nodes and the adjacency matrixes among the nodes into a graph neural network model for training to obtain the trained graph neural network model; and inputting the current session data into the trained graph neural network model to obtain a corresponding session path.
After the optimal conversation path prediction technology is adopted, through training of a large amount of conversation data, the robot analyzes the conversation of a client while talking with the client, and an optimal path with a list as a terminal point is predicted by combining different conversation paths of the intelligent robot.
Meanwhile, the conversation quality of the intelligent robot is greatly improved, the customer satisfaction is improved, the robot can screen the purchase intention of the customer while carrying out conversation communication, the order forming of the customer with high intention is facilitated, the willingness of the customer with no intention and the willingness of the customer with low intention are improved, and the maximization of the final order forming possibility is realized.
Example 2
For details not disclosed in the intelligent session system based on the graph neural network of this embodiment, please refer to specific implementation contents of the intelligent session method based on the graph neural network in other embodiments.
A schematic structural diagram of an intelligent session system based on a graph neural network according to an embodiment of the present application is shown in fig. 4.
As shown in fig. 4, the intelligent session system based on the graph neural network according to the embodiment of the present application specifically includes a client text processing module 10, a pre-trained language model module 20, a graph neural network training module 30, and a session module 40.
In particular, the method comprises the following steps of,
the client text processing module 10: the system is used for acquiring client semantic information, client dialogues and client ordering results according to client session text data.
First, a large amount of client session data is acquired, and voice of the client session is converted into client session text data through voice recognition.
The speech-to-text technique in the present embodiment employs an ASR technique.
Among them, ASR (Automatic speech recognition) is a technology for converting human speech into text.
Specifically, it is first required to perform pre-emphasis, framing, and windowing on the original speech to obtain the standard speech.
And acquiring the standard voice, and providing an effective data source for the subsequent endpoint detection processing of the standard voice.
Then, according to the text data of the client conversation, the voice information of the client is obtained through the NLP model, and the voice information comprises information of the intention and emotional tendency of the client.
Meanwhile, the information of the result of the dialogs which the client should use and the corresponding client account result is obtained according to the historical conversation data.
According to the client session text data, acquiring client semantic information, client dialogues and client ordering results, and specifically comprising the following steps:
first, the client session text data is divided into client side session data and responder session data by character analysis based on the client session text data.
Then, taking the session data of the responder as the client application dialogue; and acquiring client semantic information according to the client session data. And simultaneously, searching a client list forming keyword according to the client side conversation data and/or the responder side conversation data, and taking a list forming result corresponding to the client list forming keyword as a client list forming result.
In this embodiment, the client semantic information specifically includes client intention information and/or client emotion information; the customer intent information, in turn, includes intent-to-singleton type, intent-to-singleton amount, and intent-to-singleton time.
The client intention information can be obtained by establishing a client intention language library in advance, and the client emotion information is obtained by a language library of client emotion or simultaneously detecting the emotional excitement of the client.
Specifically, the obtaining of the client intention information specifically includes:
1) a client intention language library is established in advance, the client intention language library comprises a plurality of key words, and each key word corresponds to intention information.
2) And searching whether the keywords of the client intention library exist one by one according to the client side conversation data, and taking intention information corresponding to the keywords as client intention information when the keywords exist.
Specifically, the method for obtaining the client emotion information specifically comprises the following steps:
firstly, recognizing the conversation data of the client side by using a preset voice excitement recognition model to obtain the emotion excitement of the client. And then, obtaining the emotional information of the client according to the corresponding relation between the emotional excitement of the client and the emotion of the client.
Specifically, the emotional excitement of the client needs to be recognized through a voice excitement recognition model.
Further, when the voice contrastiveness recognition model is used for voice contrastiveness recognition, the specific training process of the voice contrastiveness recognition model is as follows:
firstly, original voice information of a client is converted into a spectrogram, and spectrogram feature extraction is carried out, and compared with other feature extraction algorithms, the spectrogram contains more original voice information including time, frequency, amplitude and other information.
Then, training and testing a model, and training a deep neural network model by using the matched input data and the corresponding contrastive degree class label; the trained model predicts unlabeled multi-modal input data, and finally the probability value of the excitement prediction can be obtained according to the trained model.
Specifically, the client emotion information is acquired, and on the other hand, the client emotion information can be acquired through a language library of client emotions. For example, the emotion of the customer includes complaints, anger, disgust or urgency, and the like, and each speech emotion is expressed by various words.
Therefore, firstly, a keyword and speech emotion corresponding library is established; then, identifying keywords in an emotion library according to the character identification data; and finally, obtaining the voice emotion corresponding to the keyword.
Pre-training language model module 20: the method is used for converting the client semantic nodes, the client dialoging nodes, the client singleting nodes and the adjacency matrixes among the nodes in the vector form through a pre-training language model according to the client semantic information, the client dialoging and the client singleting results.
Specifically, the method comprises the following steps:
a) and obtaining a client semantic feature vector, a client dialogues feature vector and a client singleton feature vector through a neural network model according to the client semantic information, the client dialogues and the client singleton result.
b) And respectively taking the client semantic feature vector, the client dialoging feature vector and the client unionized feature vector as a client semantic node, a client dialoging node and a client unionized node.
c) And connecting the client semantic nodes, the client corresponding dialogue technical nodes and the client single nodes according to the session corresponding relation to obtain an adjacency matrix among the nodes.
For a particular application, for example, the conversation history { U1, U2, U3.. UM } consists of M words. For each sentence Ui, converting each word in the Ui into a corresponding word vector by using a pre-training Glove method to obtain a vector sequence, sequentially inputting the vector sequence into a lower-layer cyclic neural network, and outputting the final state of the network as a feature vector hi of the current sentence to further obtain a feature code { h1, h2, h 3.. hM }. And inputting the data into an upper-layer cyclic neural network for further coding to obtain a feature vector hD of a conversation level, and finally taking the feature vector hD as a semantic node X of the whole conversation.
And then, constructing an adjacency matrix A between the nodes according to the real data.
In some embodiments of the present application, the method further includes obtaining a client ordering result, where the client ordering result corresponds to the client-based dialog one-to-one.
The neural network training module 30: and the method is used for inputting the client semantic nodes, the client corresponding dialogue technique nodes, the client single-forming nodes and the adjacency matrixes among the nodes into the graph neural network model for training to obtain the trained graph neural network model.
The session module 40: and the method is used for inputting the current session data into the trained graph neural network model to obtain a corresponding session path.
According to the client response dialogue historical information, a heteromorphic graph containing client semantic nodes and client response dialogue technical nodes can be constructed, and an optimal conversation response path can be predicted.
A schematic diagram of a neural network according to an embodiment of the present application is shown in fig. 3.
The graph neural network model GNN comprises an encoder and a decoder, wherein input data of the encoder are client semantic nodes, client dialogues nodes and adjacency matrixes among the nodes in a vector form, and output data of the encoder is embedded in a graph; the input data of the decoder is graph embedding, the output data of the decoder is a reconstructed graph, and the reconstructed graph is a corresponding conversation path, namely a conversation path graph.
As shown in fig. 3, the graph neural network model for path connection prediction uses a variational graph self-encoder (VGAE), the network is divided into an encoder and a decoder, the encoder uses a graph formed by graph nodes X and an adjacent matrix a as input data, the output is graph embedding Z, and the decoder uses the graph embedding Z as input, and the output is a reconstructed graph.
Specifically, the mean and variance of the graph-embedded multidimensional gaussian distribution are obtained by passing input data through two multi-layer Graph Convolution Networks (GCNs):
μ＝GCNμ(X,A)；logσ＝GCNσ(X,A)；
wherein X is a client semantic node, a client dialoging node and/or a single result node; a is the adjacency matrix between the nodes.
After obtaining the mean value mu and the variance mu, sampling from the standard Gaussian normal distribution to obtain an element, and embedding the graph obtained by the output of the encoder into a value Z which is mu plus the element sigma; at reconstruction, the decoder is the inner product of the graph embedding, i.e.
Specifically, in the case of the trainer, the graph reconstructed by the variational graph autoencoder VGAE is desired to be as close as possible to the original graph, and the distribution q (Z | X, a) calculated by the multi-layer graph convolution network GCN is desired to be as similar as possible to the standard gaussian distribution p (Z), so that the loss function
From cross entropy
And the relative entropy KL divergence, and the specific calculation mode is as follows:
finally, obtaining a trained graph neural network model, wherein output data obtained through the graph neural network processing is a graph formed by the connection relation among the nodes, wherein the graph may comprise a plurality of paths leading to the unijunction result from the current session state, and the model can output an optimal path leading to the unijunction result from the current session state through the training and the super-parameter fine adjustment of the graph neural network model by using massive training samples.
And finally, generating an optimal path of a subsequent conversation according to model prediction, and selecting the conversation to communicate with the client by the robot according to the predicted optimal path, thereby improving the probability of singleness as much as possible.
In the intelligent session system based on the graph neural network in the embodiment of the application, the client text processing module 10 acquires client semantic information, client application dialogues and client order forming results according to client session text data; the pre-training language model module 20 converts the client semantic nodes, the client corresponding dialogues nodes, the client list-forming nodes and the adjacent matrixes among the nodes in a vector form through a pre-training language model according to the client semantic information, the client corresponding dialogues and the client list-forming results; the graph neural network training module 30 inputs the client semantic nodes, the client corresponding dialogue technique nodes, the client single nodes and the adjacency matrixes among the nodes into the graph neural network model for training to obtain the trained graph neural network model; the session module 40 inputs the current session data into the trained neural network model to obtain a corresponding session path.
After the optimal conversation path prediction technology is adopted, through training of a large amount of conversation data, the robot analyzes the conversation of a client while talking with the client, and an optimal path with a list as a terminal point is predicted by combining different conversation paths of the intelligent robot.
Meanwhile, the conversation quality of the intelligent robot is greatly improved, the customer satisfaction is improved, the robot can screen the purchase intention of the customer while carrying out conversation communication, the order forming of the customer with high intention is facilitated, the willingness of the customer with no intention and the willingness of the customer with low intention are improved, and the maximization of the final order forming possibility is realized.
Example 3
For details not disclosed in the intelligent session device based on the graph neural network of this embodiment, please refer to specific implementation contents of the intelligent session method or system based on the graph neural network in other embodiments.
A schematic structural diagram of an intelligent conversational device 400 based on a graph neural network according to an embodiment of the application is shown in fig. 5.
As shown in fig. 5, the intelligent conversation 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.
It will be understood by those skilled in the art that the schematic diagram 5 is merely an example of the intelligent session device 400 and does not constitute a limitation of the intelligent session device 400, and may include more or less components than those shown, or combine some components, or different components, for example, the intelligent session 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, and the processor 401 is the control center of the intelligent session device 400 and connects the various parts of the whole intelligent session device 400 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 intelligent session device 400 by executing or executing computer readable instructions or modules stored in the memory 402 and 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 intelligent conversation 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 intelligent conversation apparatus 400, if implemented in the form of software function modules and sold or used as separate products, may be stored in a computer-readable storage medium. 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 a processor to implement the intelligent conversational method based on a graph neural network in other embodiments.
According to the intelligent session equipment and the computer storage medium based on the graph neural network in the embodiment of the application, client semantic information, client application dialogues and client ordering results are obtained according to client session text data; converting to obtain vector-form customer semantic nodes, customer corresponding dialogue technique nodes, customer single-forming nodes and an adjacent matrix among the nodes through a pre-training language model according to customer semantic information, customer corresponding dialogue techniques and customer single-forming results; inputting the client semantic nodes, the client corresponding dialogue technique nodes, the client single-forming nodes and the adjacency matrixes among the nodes into a graph neural network model for training to obtain the trained graph neural network model; and inputting the current session data into the trained graph neural network model to obtain a corresponding session path.
After the optimal conversation path prediction technology is adopted, through training of a large amount of conversation data, the robot analyzes the conversation of a client while talking with the client, and an optimal path with a list as a terminal point is predicted by combining different conversation paths of the intelligent robot.
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. An intelligent conversation method based on a graph neural network is characterized by specifically comprising the following steps:
according to the client session text data, acquiring client semantic information, client dialogues and client ordering results;
converting to obtain vector-form customer semantic nodes, customer dialogues nodes, customer singleness nodes and adjacency matrixes among the nodes through a pre-training language model according to the customer semantic information, the customer dialogues and the customer singleness results;
inputting the client semantic nodes, the client corresponding dialogue technique nodes, the client single-forming nodes and the adjacency matrixes among the nodes into a graph neural network model for training to obtain the trained graph neural network model;
and inputting the current session data into the trained graph neural network model to obtain a corresponding session path.
2. The intelligent conversation method according to claim 1, wherein said obtaining client semantic information, client dialogues and client billing results based on client conversation text data specifically comprises:
according to the client session text data, dividing the client session text data into client side session data and responder side session data through character analysis;
taking the responder session data as a client application dialog;
obtaining client semantic information according to the client side session data;
and searching the client list-forming keywords according to the client side conversation data and/or the responder side conversation data, and taking the list-forming result corresponding to the client list-forming keywords as the client list-forming result.
3. The intelligent conversational method of claim 2, wherein the client semantic information comprises client intent information and/or client sentiment information; the client intent includes intent to be single type, intent to be singular amount, and intent to be single time.
4. The intelligent conversation method according to claim 3, wherein obtaining the client intention information specifically comprises:
the method comprises the steps of establishing a client intention language library in advance, wherein the client intention language library comprises a plurality of key words, and each key word corresponds to intention information;
and searching whether the keywords of the client intention library exist one by one according to the client side conversation data, and taking intention information corresponding to the keywords as client intention information when the keywords exist.
5. The intelligent conversation method according to claim 3, wherein obtaining client emotion information specifically comprises:
recognizing the conversation data of the client side by using a preset voice excitement recognition model to obtain the emotion excitement of the client;
and obtaining the emotional information of the client according to the corresponding relation between the emotional excitement of the client and the emotion of the client.
6. The intelligent conversation method according to claim 1, wherein the converting into vector-form client semantic nodes, client-oriented dialogue technique nodes and adjacency matrixes among nodes according to the client semantic information, the client-oriented dialogue technique and the client-oriented dialogue result through a pre-training language model specifically comprises:
obtaining a client semantic feature vector, a client dialogues feature vector and a client singleton feature vector through a neural network model according to the client semantic information, the client dialogues and the client singleton result;
respectively taking the client semantic feature vector, the client dialoging feature vector and the client unigram feature vector as a client semantic node, a client dialoging node and a client unigram node;
and connecting the client semantic nodes, the client corresponding dialogue technical nodes and the client single nodes according to the session corresponding relation to obtain an adjacency matrix among the nodes.
7. The intelligent conversation method according to claim 1, wherein before obtaining client semantic information, client dialogues and client unionization results based on client conversation text data, further comprising obtaining client conversation voice samples;
the customer voice samples are converted to customer session text data by speech recognition techniques.
8. An intelligent conversation system based on a graph neural network is characterized by specifically comprising:
a client text processing module: the system is used for acquiring client semantic information, client dialogues and client ordering results according to client session text data;
pre-training a language model module: the client semantic nodes, the client corresponding dialogue technique nodes, the client list forming nodes and the adjacency matrixes among the nodes in a vector form are obtained through conversion according to the client semantic information, the client corresponding dialogue technique and the client list forming result through a pre-training language model;
the graph neural network training module: the system comprises a graph neural network model, a semantic node, a dialogue node, a client single-forming node and an adjacency matrix among the nodes, wherein the semantic node, the dialogue node, the client single-forming node and the adjacency matrix are used for inputting the semantic node, the dialogue node, the client single-forming node and the adjacency matrix among the nodes into the graph neural network model for training to obtain the trained graph neural network model;
a session module: and the system is used for inputting the current session data into the trained graph neural network model to obtain a corresponding session path.
9. An intelligent session device based on a graph neural network, comprising:
a memory: for storing executable instructions; and
a processor for interfacing with the memory to execute the executable instructions to perform the graph neural network-based intelligent conversational method of any one 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 intelligent conversational method based on a graph neural network of any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110516335.9A CN113239147A (en) | 2021-05-12 | 2021-05-12 | Intelligent conversation method, system and medium based on graph neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110516335.9A CN113239147A (en) | 2021-05-12 | 2021-05-12 | Intelligent conversation method, system and medium based on graph neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113239147A true CN113239147A (en) | 2021-08-10 |
Family
ID=77133711
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110516335.9A Pending CN113239147A (en) | 2021-05-12 | 2021-05-12 | Intelligent conversation method, system and medium based on graph neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113239147A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113642714A (en) * | 2021-08-27 | 2021-11-12 | 国网湖南省电力有限公司 | Insulator pollution discharge state identification method and system based on small sample learning |
| CN113704440A (en) * | 2021-09-06 | 2021-11-26 | 中国计量大学 | Conversation recommendation method based on path representation in article graph network |
| CN114118080A (en) * | 2021-11-10 | 2022-03-01 | 北京深维智信科技有限公司 | Method and system for automatically identifying client intention from sales session |
| CN114416941A (en) * | 2021-12-28 | 2022-04-29 | 北京百度网讯科技有限公司 | Generation method and device of dialogue knowledge point determination model fusing knowledge graph |
| CN115188374A (en) * | 2022-06-22 | 2022-10-14 | 百融睿诚信息科技有限公司 | Method and device for updating dialect |
| CN115712706A (en) * | 2022-11-07 | 2023-02-24 | 贝壳找房(北京)科技有限公司 | Method and device for determining action decision based on session |
| CN117540829A (en) * | 2023-10-18 | 2024-02-09 | 广西壮族自治区通信产业服务有限公司技术服务分公司 | Knowledge sharing large language model collaborative optimization method and system |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109816101A (en) * | 2019-01-31 | 2019-05-28 | 中科人工智能创新技术研究院(青岛)有限公司 | A kind of session sequence of recommendation method and system based on figure convolutional neural networks |
| CN110490717A (en) * | 2019-09-05 | 2019-11-22 | 齐鲁工业大学 | Method of Commodity Recommendation and system based on user conversation and figure convolutional neural networks |
| CN110609891A (en) * | 2019-09-18 | 2019-12-24 | 合肥工业大学 | Visual dialog generation method based on context awareness graph neural network |
| CN110704594A (en) * | 2019-09-27 | 2020-01-17 | 北京百度网讯科技有限公司 | Task type dialogue interaction processing method and device based on artificial intelligence |
| CN110796313A (en) * | 2019-11-01 | 2020-02-14 | 北京理工大学 | Session recommendation method based on weighted graph volume and item attraction model |
| CN111475637A (en) * | 2020-06-24 | 2020-07-31 | 支付宝(杭州)信息技术有限公司 | Data processing and training method and device for pushing knowledge points |
| CN111966800A (en) * | 2020-07-27 | 2020-11-20 | 腾讯科技（深圳）有限公司 | Emotional dialogue generation method and device and emotional dialogue model training method and device |
| CN112417112A (en) * | 2020-11-10 | 2021-02-26 | 中山大学 | Open domain dialogue system evaluation method based on graph characterization enhancement |
| US20210103706A1 (en) * | 2019-10-04 | 2021-04-08 | Nec Laboratories America, Inc. | Knowledge graph and alignment with uncertainty embedding |
| US20210108939A1 (en) * | 2020-12-22 | 2021-04-15 | Nesreen K. Ahmed | Personalized mobility as a service |
| CN112765978A (en) * | 2021-01-14 | 2021-05-07 | 中山大学 | Dialog diagram reconstruction method and system for multi-person multi-turn dialog scene |
-
2021
- 2021-05-12 CN CN202110516335.9A patent/CN113239147A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109816101A (en) * | 2019-01-31 | 2019-05-28 | 中科人工智能创新技术研究院(青岛)有限公司 | A kind of session sequence of recommendation method and system based on figure convolutional neural networks |
| CN110490717A (en) * | 2019-09-05 | 2019-11-22 | 齐鲁工业大学 | Method of Commodity Recommendation and system based on user conversation and figure convolutional neural networks |
| CN110609891A (en) * | 2019-09-18 | 2019-12-24 | 合肥工业大学 | Visual dialog generation method based on context awareness graph neural network |
| CN110704594A (en) * | 2019-09-27 | 2020-01-17 | 北京百度网讯科技有限公司 | Task type dialogue interaction processing method and device based on artificial intelligence |
| US20210103706A1 (en) * | 2019-10-04 | 2021-04-08 | Nec Laboratories America, Inc. | Knowledge graph and alignment with uncertainty embedding |
| CN110796313A (en) * | 2019-11-01 | 2020-02-14 | 北京理工大学 | Session recommendation method based on weighted graph volume and item attraction model |
| CN111475637A (en) * | 2020-06-24 | 2020-07-31 | 支付宝(杭州)信息技术有限公司 | Data processing and training method and device for pushing knowledge points |
| CN111966800A (en) * | 2020-07-27 | 2020-11-20 | 腾讯科技（深圳）有限公司 | Emotional dialogue generation method and device and emotional dialogue model training method and device |
| CN112417112A (en) * | 2020-11-10 | 2021-02-26 | 中山大学 | Open domain dialogue system evaluation method based on graph characterization enhancement |
| US20210108939A1 (en) * | 2020-12-22 | 2021-04-15 | Nesreen K. Ahmed | Personalized mobility as a service |
| CN112765978A (en) * | 2021-01-14 | 2021-05-07 | 中山大学 | Dialog diagram reconstruction method and system for multi-person multi-turn dialog scene |
Non-Patent Citations (4)
| Title |
|---|
| CHI-MAN WONG ET AL: "Improving Conversational Recommender System by Pretraining Billion-scale Knowledge Graph", 《ARXIV:2104.14899V1 [CS.IR]》, pages 1 - 6 * |
| LIZI LIAO ET AL: "Deep Conversational Recommender in Travel", 《ARXIV:1907.00710V1 [CS.CL]》, pages 1 - 11 * |
| WENPENG HU ET AL: "GSN: A Graph-Structured Network for Multi-Party Dialogues", 《ARXIV:1905.13637V1 [CS.CL]》, pages 1 - 9 * |
| YAN ZENG ET AL: "Multi-Domain Dialogue State Tracking based on State Graph", 《ARXIV:2010.11137V1 [CS.CL]》, pages 1 - 6 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113642714A (en) * | 2021-08-27 | 2021-11-12 | 国网湖南省电力有限公司 | Insulator pollution discharge state identification method and system based on small sample learning |
| CN113642714B (en) * | 2021-08-27 | 2024-02-09 | 国网湖南省电力有限公司 | Insulator pollution discharge state identification method and system based on small sample learning |
| CN113704440A (en) * | 2021-09-06 | 2021-11-26 | 中国计量大学 | Conversation recommendation method based on path representation in article graph network |
| CN113704440B (en) * | 2021-09-06 | 2022-02-18 | 中国计量大学 | Conversation recommendation method based on path representation in article graph network |
| CN114118080A (en) * | 2021-11-10 | 2022-03-01 | 北京深维智信科技有限公司 | Method and system for automatically identifying client intention from sales session |
| CN114118080B (en) * | 2021-11-10 | 2022-09-13 | 北京深维智信科技有限公司 | Method and system for automatically identifying client intention from sales session |
| CN114416941B (en) * | 2021-12-28 | 2023-09-05 | 北京百度网讯科技有限公司 | Knowledge graph-fused dialogue knowledge point determination model generation method and device |
| CN114416941A (en) * | 2021-12-28 | 2022-04-29 | 北京百度网讯科技有限公司 | Generation method and device of dialogue knowledge point determination model fusing knowledge graph |
| CN115188374A (en) * | 2022-06-22 | 2022-10-14 | 百融睿诚信息科技有限公司 | Method and device for updating dialect |
| CN115712706B (en) * | 2022-11-07 | 2023-09-15 | 贝壳找房(北京)科技有限公司 | Method and device for determining action decision based on session |
| CN115712706A (en) * | 2022-11-07 | 2023-02-24 | 贝壳找房(北京)科技有限公司 | Method and device for determining action decision based on session |
| CN117540829A (en) * | 2023-10-18 | 2024-02-09 | 广西壮族自治区通信产业服务有限公司技术服务分公司 | Knowledge sharing large language model collaborative optimization method and system |
| CN117540829B (en) * | 2023-10-18 | 2024-05-17 | 广西壮族自治区通信产业服务有限公司技术服务分公司 | Knowledge sharing large language model collaborative optimization method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112804400B (en) | Customer service call voice quality inspection method and device, electronic equipment and storage medium | |
| CN111028827B (en) | Interaction processing method, device, equipment and storage medium based on emotion recognition | |
| CN113239147A (en) | Intelligent conversation method, system and medium based on graph neural network | |
| US10679613B2 (en) | Spoken language understanding system and method using recurrent neural networks | |
| CN111916111B (en) | Intelligent voice outbound method and device with emotion, server and storage medium | |
| Khare et al. | Self-supervised learning with cross-modal transformers for emotion recognition | |
| CN107818798A (en) | Customer service quality evaluating method, device, equipment and storage medium | |
| CN111081230A (en) | Speech recognition method and apparatus | |
| CN111182162A (en) | Telephone quality inspection method, device, equipment and storage medium based on artificial intelligence | |
| CN112131359A (en) | Intention identification method based on graphical arrangement intelligent strategy and electronic equipment | |
| JP2024502946A (en) | Punctuation and capitalization of speech recognition transcripts | |
| CN112863489B (en) | Speech recognition method, apparatus, device and medium | |
| CN114023300A (en) | Chinese speech synthesis method based on diffusion probability model | |
| CN116049360A (en) | Intelligent voice dialogue scene conversation intervention method and system based on client image | |
| CN112053681B (en) | Telephone customer service quality scoring method and system for ASR and NLU combined training | |
| CN114911932A (en) | Heterogeneous graph structure multi-conversation person emotion analysis method based on theme semantic enhancement | |
| CN115269836A (en) | Intention identification method and device | |
| CN114386426A (en) | Gold medal speaking skill recommendation method and device based on multivariate semantic fusion | |
| CN110931002B (en) | Man-machine interaction method, device, computer equipment and storage medium | |
| CN112102807A (en) | Speech synthesis method, apparatus, computer device and storage medium | |
| CN117150338A (en) | Task processing, automatic question and answer and multimedia data identification model training method | |
| US20220201121A1 (en) | System, method and apparatus for conversational guidance | |
| CN114373443A (en) | Speech synthesis method and apparatus, computing device, storage medium, and program product | |
| CN114328867A (en) | Intelligent interruption method and device in man-machine conversation | |
| CN111310847B (en) | Method and device for training element classification model |
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 |