CN111091011A - Domain prediction method, domain prediction device and electronic equipment - Google Patents
Domain prediction method, domain prediction device and electronic equipment Download PDFInfo
- Publication number
- CN111091011A CN111091011A CN201911327989.6A CN201911327989A CN111091011A CN 111091011 A CN111091011 A CN 111091011A CN 201911327989 A CN201911327989 A CN 201911327989A CN 111091011 A CN111091011 A CN 111091011A
- Authority
- CN
- China
- Prior art keywords
- domain
- current round
- field
- round
- interactive
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 78
- 230000002452 interceptive effect Effects 0.000 claims abstract description 203
- 230000003993 interaction Effects 0.000 claims abstract description 55
- 230000008569 process Effects 0.000 claims abstract description 27
- 238000012549 training Methods 0.000 claims description 25
- 238000004590 computer program Methods 0.000 claims description 10
- 238000012544 monitoring process Methods 0.000 claims description 8
- 238000007781 pre-processing Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 description 58
- 238000004891 communication Methods 0.000 description 10
- 238000013528 artificial neural network Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 5
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 5
- 230000006870 function Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000004091 panning Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
- G10L15/063—Training
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/16—Speech classification or search using artificial neural networks
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/18—Speech classification or search using natural language modelling
- G10L15/1822—Parsing for meaning understanding
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/22—Procedures used during a speech recognition process, e.g. man-machine dialogue
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Acoustics & Sound (AREA)
- Human Computer Interaction (AREA)
- General Engineering & Computer Science (AREA)
- Audiology, Speech & Language Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Machine Translation (AREA)
Abstract
The invention provides a domain prediction method, a domain prediction device and electronic equipment, wherein the domain prediction method comprises the following steps: determining the interactive text of the current round; inputting the interactive text and the supervision information of the current round into a field prediction model to obtain field probability distribution output by the field prediction model and corresponding to the interactive text of the current round, wherein the supervision information is obtained by correcting the field probability distribution output by the field prediction model and corresponding to the interactive text of the previous round based on the field information determined after semantic understanding of the interactive text of the previous round; and determining a domain prediction result based on the domain probability distribution corresponding to the current round of interactive text. The domain prediction method provided by the embodiment of the invention can greatly improve the accuracy of model prediction in the multi-round interaction process, and particularly can obtain an accurate domain prediction result in the face of simplified interaction in the multi-round interaction process.
Description
Technical Field
The present invention relates to the field of voice interaction technologies, and in particular, to a domain prediction method, a domain prediction apparatus, and an electronic device.
Background
In the voice interaction process, in order to better understand the semantics, it is usually necessary to predict which domain the user's expression content belongs to.
In the prior art, one method is to match whether user expression content belongs to a certain field or not based on a grammar rule network or a state machine rule, the generalization capability of the method is poor, and sentence patterns which are not included can not be understood. The other mode is to use a deep neural network to learn sentence pattern information existing in user expression content so as to achieve the purpose of the model prediction field, but the accuracy of the deep neural network on the prediction of single-round interactive expression content is still enough, and once multi-round conversations are involved, the accuracy of the model prediction is greatly reduced.
Disclosure of Invention
Embodiments of the present invention provide a domain prediction method, a domain prediction apparatus, an electronic device, and a readable storage medium that overcome the above problems or at least partially solve the above problems.
In a first aspect, an embodiment of the present invention provides a domain prediction method, including: determining the interactive text of the current round; inputting the current round of interactive texts and supervision information into a field prediction model to obtain field probability distribution output by the field prediction model and corresponding to the current round of interactive texts, wherein the supervision information is obtained by correcting the field probability distribution output by the field prediction model and corresponding to the previous round of interactive texts based on the field information determined after semantic understanding of the previous round of interactive texts; determining a domain prediction result based on the domain probability distribution corresponding to the current round of interactive texts; the field prediction model is obtained by taking multi-round interactive text data as samples in advance and taking predetermined field probability distribution data respectively corresponding to the multi-round interactive text data as sample labels for training.
According to the field prediction method of the embodiment of the invention, the supervision information is obtained according to the following steps: obtaining the domain information determined after semantic understanding of the previous round of interactive text; inputting the previous round of interactive text into the field prediction model to obtain field probability distribution output by the field prediction model and corresponding to the previous round of interactive text; and determining the supervision information based on the domain information and the domain probability distribution corresponding to the previous round of interactive text.
According to the field prediction method provided by the embodiment of the invention, the inputting the interactive text and the supervision information of the current round into a field prediction model to obtain the field probability distribution output by the field prediction model and corresponding to the interactive text of the current round comprises the following steps: inputting the current round of interactive text into a preprocessing layer of the field prediction model to obtain current round content characteristics and current round field word proportion characteristics, wherein the current round content characteristics are used for representing the expression content of the current round of interactive text, and the current round field word proportion characteristics are used for representing the proportion of the length of each field entity of the current round of interactive text in the current round of interactive text; and inputting the content characteristics, the field word proportion characteristics and the supervision information of the current round into an inference layer of the field prediction model to obtain the field probability distribution corresponding to the interactive text of the current round.
According to the domain prediction method of the embodiment of the present invention, the inputting the content features, the domain word proportion features and the supervision information of the current round into the inference layer of the domain prediction model to obtain the domain probability distribution corresponding to the interactive text of the current round includes: inputting the content features and the field word proportion features of the current round into a first layer structure of the reasoning layer to obtain a text expression with field information proportion and a text expression with field classification information, wherein the text expression with field information proportion is used for representing the field information of the interactive text of the current round, and the text expression with field classification information is used for representing the predicted field proportion weight of the interactive text of the current round; and inputting the text expression with the domain information ratio, the text expression with the domain classification information and the supervision information into a second layer structure of the reasoning layer to obtain the domain probability distribution corresponding to the current round of interactive text.
According to the domain prediction method provided by the embodiment of the invention, the text expression with the domain classification information is determined based on the content characteristics of the current round, the word proportion characteristics of the current round and the weight of the domain classification information of the domain prediction model, and the weight of the domain classification information is determined according to each domain information learned by the domain prediction model in the training process.
According to the domain prediction method of the embodiment of the invention, the inputting the current round of interactive text and supervision information into the domain prediction model further comprises: inputting the current round of interaction text, the supervision information and personalized features into the field prediction model, wherein the personalized features are used for representing associated information in the current round of interaction.
According to the field prediction method provided by the embodiment of the invention, the personalized features comprise the foreground field features of the current round of interaction and the background field features of the current round of interaction.
In a second aspect, an embodiment of the present invention provides a domain prediction apparatus, including: the text determining unit is used for determining the interactive text of the current round; a probability distribution determining unit, configured to input the current round of interactive texts and monitoring information into a domain prediction model, and obtain domain probability distribution output by the domain prediction model and corresponding to the current round of interactive texts, where the monitoring information is domain information determined after semantic understanding based on a previous round of interactive texts, and is obtained by correcting the domain probability distribution output by the domain prediction model and corresponding to the previous round of interactive texts; the domain determining unit is used for determining a domain prediction result based on the domain probability distribution corresponding to the current round of interactive texts; the field prediction model is obtained by training a plurality of rounds of interactive text data as samples in advance and field probability distribution which is predetermined and corresponds to the plurality of rounds of interactive text data respectively as sample labels.
In a third aspect, an embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the steps of the domain prediction method provided in the first aspect when executing the program.
In a fourth aspect, an embodiment of the present invention provides a non-transitory computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, implements the steps of the domain prediction method as provided in the first aspect.
According to the domain prediction method, the domain prediction device, the electronic equipment and the readable storage medium, the domain prediction result of the current round is jointly predicted by using the domain probability distribution corresponding to the previous round of interactive text, the domain information determined after semantic understanding of the previous round of interactive text and the current round of interactive text, so that the current round of domain prediction can take the prediction domain of the previous round and the actual context of the previous round into consideration, the accuracy of model prediction in a multi-round interactive process is greatly improved, and particularly, the accurate domain prediction result can be obtained in the face of simplified interaction in the multi-round interactive process.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
FIG. 1 is a flow chart of a domain prediction method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a topology of a domain prediction model in an application phase according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a topology of a domain prediction model in a training phase according to an embodiment of the present invention;
FIG. 4 is a schematic structural diagram of a domain prediction apparatus according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In the prior art, the deep neural network does not completely solve the prediction of short spoken language in multiple rounds of conversation in the interaction process. For example, when a user says 'navigate to science news for the first time', the currently used deep neural network can predict that the current request belongs to the field of mapU navigation; however, when the user then says "Wanda", and then inputs "Wanda" into the deep neural network, since there is no strong sentence pattern information in the expression content of the current round, which domain the expression content of the current round belongs to cannot be predicted by analyzing the entity information alone, so the model has poor comprehension capability for such problems.
A domain prediction method according to an embodiment of the present invention, which may be used in a scenario of multiple rounds of interaction, is described below with reference to fig. 1 to 3.
The field prediction method of the embodiment of the invention comprises the following steps:
and S100, determining the interactive text of the current round.
In actual implementation, determining the current round of interactive text may include: acquiring the interactive voice information of the current round; and transferring the interactive voice information of the current round into the interactive text of the current round.
It should be noted that, multiple rounds of interaction represent that the user performs multiple associated communications with the artificial intelligence, and the current round of interaction includes that the current user speaks or performs meaning expression in other forms, and the artificial intelligence performs corresponding feedback after receiving the round of meaning expression, and correspondingly, the previous round of interaction is the round of interaction before the current round of interaction.
And S200, inputting the interactive text and the supervision information of the current round into a domain prediction model to obtain domain probability distribution which is output by the domain prediction model and corresponds to the interactive text of the current round. And the supervision information is obtained by correcting the domain probability distribution which is output by the domain prediction model and corresponds to the previous round of interactive text based on the domain information which is determined after the previous round of interactive text is semantically understood.
The domain information determined after semantic understanding of the previous round of interactive text refers to the domain information obtained after the last round of expression content of the user is subjected to final semantic understanding. For example: the user says 'i want to listen to Green fairy tales', the model predicts the field of the current round of conversation as the music field, but after semantic understanding, the expressive content of the user is understood as the story field. Wherein the story field is the field information after semantic understanding.
Taking the embodiment shown in fig. 2 as an example, the genre of the previous round of interactive input is named his _ send, and the text of the current round of interactive input is named send.
The above-mentioned supervision information is denoted as his _ attribution _ vec, the domain information determined after semantic understanding of the previous round of interactive text may be represented by a matrix his _ label _ vec, and the matrix his _ label _ vec may be determined by a domain matrix his _ label _ matrix of the previous round of interactive text after a series of semantic understanding.
For example, his _ label _ vec ═ his _ label _ matrixTThe his _ label _ matrix, in other words, the transposed matrix of the his _ label _ matrix is multiplied by the his _ label _ matrix to obtain the his _ label _ vec, so that the dimension of the his _ label _ vec matrix becomes larger, and the information amount becomes rich.
The domain information determined after semantic understanding is used for correcting the domain probability distribution of the previous round, namely the domain information (matrix his _ label _ vec) related to the semantic understanding result is fully connected with the domain probability distribution of the previous round to obtain a supervision matrix (supervision information), and the supervision matrix is used for supervising the domain prediction of the current round.
It should be noted that, when determining the domain probability distribution corresponding to the current round of interactive text by using the domain prediction model, the purpose of the domain information (his _ label _ vec) determined after semantic understanding of the previous round of interactive text is used to supervise the domain probability distribution of the current round of interaction. For example, in the previous round of interactive text prediction process, when the domain predicted by the previous round of domain prediction model is consistent with the domain after the previous round of semantic understanding (assumed to be a music domain), the weight belonging to the music domain in his _ send _ vec will be strengthened, and the weights of other domains will be weakened; when the field predicted by the field prediction model in the previous round is inconsistent with the field subjected to semantic understanding (assuming that the field prediction model is predicted to be a music field and the field subjected to actual semantic understanding is a story field), the weight belonging to the story field in his _ sent _ vec is strengthened, the weights of other fields are weakened, and when the interactive text in the current round is predicted to belong to a certain field, the music field is more unconfirmed, and the interactive text in the current round is more believable to be the story field.
The domain prediction model is obtained by taking multi-round interactive text data as samples in advance and taking predetermined domain probability distribution data respectively corresponding to the multi-round interactive text data as sample labels for training.
Taking fig. 3 as an example, during training, previous round of interactive text data, domain probability distribution data corresponding to the previous round of interactive text data, current round of interactive text data, and domain probability distribution data corresponding to the current round of interactive text data are input, where the previous round of interactive text data and the current round of interactive text data are used as training samples, and the domain probability distribution data corresponding to the previous round of interactive text data and the domain probability distribution data corresponding to the current round of interactive text data are used as training sample labels.
Through training of a certain amount of sample data and sample labels, a trained domain prediction model can be obtained. In the training phase, the domain prediction model does not need to input domain information (his _ label _ vec) or a domain matrix (his _ label _ matrix) determined after semantic understanding of the interactive text.
And S300, determining a domain prediction result based on the domain probability distribution corresponding to the current round of interactive texts.
In an actual implementation, the domain probability distribution may be in the form of a probability distribution matrix, in which the domain corresponding to the maximum value may be determined as the domain prediction result.
According to the domain prediction method provided by the embodiment of the invention, the domain probability distribution corresponding to the previous round of interactive text, the domain information determined after semantic understanding of the previous round of interactive text and the domain prediction result of the current round of interactive text are jointly used for predicting the domain prediction result of the current round, so that the domain prediction of the current round can take the prediction domain of the previous round and the actual context of the previous round into consideration, the accuracy of model prediction in a multi-round interactive process is greatly improved, and especially, the accurate domain prediction result can be obtained in the face of simplified interaction in the multi-round interactive process.
In some embodiments, as shown in fig. 2, the method for predicting the domain of the embodiment of the present invention further includes:
and S010, obtaining the domain information determined after semantic understanding of the previous round of interactive text.
For example, a domain matrix his _ label _ matrix determined after semantic understanding of the previous round of interactive text is obtained, and domain information his _ label _ vec determined after semantic understanding of the previous round of interactive text is determined based on the domain matrix his _ label _ matrix.
If the service after the last round of semantic understanding is the navigation field, setting the parameter of the navigation field to be 1, and setting all the other parameters to be 0; alternatively, the setting may be reversed, that is, the parameter of the navigation field is set to 0, and all others are set to 1.
Wherein his _ label _ vec is his _ label _ matrixTThe his _ label _ matrix, in other words, the transposed matrix of the his _ label _ matrix is multiplied by the his _ label _ matrix to obtain the his _ label _ vec, so that the dimension of the his _ label _ vec matrix becomes larger, and the information amount becomes rich.
And S020, inputting the previous round of interactive text into the domain prediction model to obtain the domain probability distribution which is output by the domain prediction model and corresponds to the previous round of interactive text.
In actual execution, the domain probability distribution corresponding to the previous round of interactive text may be determined and obtained by a domain prediction model during the current round of prediction, or the obtained domain probability distribution may be stored during the previous round of prediction and called during the current round of prediction.
And S030, determining supervision information based on the domain information and the domain probability distribution corresponding to the previous round of interactive text.
In actual execution, the domain information is fully connected with the domain probability distribution corresponding to the previous round of interactive text, so that the supervision information can be obtained.
In some embodiments, step S200, inputting the current round of interactive text and the supervision information into the domain prediction model, and obtaining a domain probability distribution output by the domain prediction model and corresponding to the current round of interactive text, includes:
and step S210, inputting the interactive text of the current round into a preprocessing layer of a field prediction model to obtain the content characteristics and the word proportion characteristics of the field of the current round.
Taking the domain prediction model shown in fig. 2 as an example, the content feature of the current round is "forgetting water", and the domain term proportion feature of the current round is "domain term proportion".
The content characteristics of the current round are used for representing the expression contents of the interactive text of the current round, and the content characteristics of the current round can be in a word-embedding matrix form, for example, the word-embedding matrix can be generated through a neural network or other network models.
The field word proportion feature of the current round is used for representing the proportion of the length of each field entity of the interactive text of the current round in the interactive text of the current round. Specifically, by sorting the domain vocabularies of each domain, the expression content of the user is matched with the domain vocabularies (there are many methods for the algorithm for matching the domain vocabularies at present, such as an AC automaton (Aho-firm automation), etc.), and how many words in the current expression content belong to the domain vocabularies is calculated, so as to obtain the percentage of the number of each domain vocabulary in the whole expression content.
The field word proportion feature of the round can be determined in the form of field dictionary arc-pasting coding, wherein the matrix dimension of the field dictionary arc-pasting coding output is 1 × label, the field dictionary arc-pasting means that all entities in a certain field are collected and sorted together, the method of entity arc-pasting (the implementation mode of the entity arc-pasting can be a method of regular matching, an AC automaton and the like, which is not described herein) is used for arc-pasting on the expression content of the user, and the length ratio of the number of the entities in the field in the expression content of the user to the whole expression content text is obtained.
And S220, inputting the content characteristics, the field word proportion characteristics and the supervision information of the current round into an inference layer of a field prediction model to obtain field probability distribution corresponding to the interactive text of the current round.
Taking fig. 2 as an example, the content features (forgetting water), the domain word proportion features (domain word proportion) and the supervision information (his _ attribution _ vec) of the current round are input into the inference layer of the domain prediction model, so as to obtain the domain probability distribution (domain) corresponding to the interactive text of the current round.
Therefore, when the inference layer carries out prediction, the inference layer simultaneously considers the expression content of the interactive text of the current round, the ratio of the length of each field entity of the interactive text of the current round to the interactive text of the current round and predicts the field probability distribution corresponding to the interactive text of the current round according to the supervision information of the interactive text of the previous round, and when the simplified interaction in the multi-round interaction process is faced, an accurate field prediction result can be obtained.
In some embodiments, step S220, inputting the content features of the current round, the term proportion features of the current round, and the monitoring information into an inference layer of the domain prediction model, to obtain a domain probability distribution corresponding to the interactive text of the current round, includes:
step S221, inputting the content features of the current round and the word proportion features of the current round into a first layer structure of an inference layer to obtain a text expression with the proportion of the domain information and a text expression with the classification information of the domain, wherein the text expression with the proportion of the domain information is used for representing the domain information of the interactive text of the current round, and the text expression with the classification information of the domain is used for representing the predicted domain proportion weight of the interactive text of the current round.
Taking the domain prediction model shown in fig. 2 as an example, the content features (forgetting water) of the current round and the domain word proportion features (domain word proportion) of the current round are input into the first layer structure of the inference layer, so as to obtain the text expression (send _ vec) with the domain information proportion and the text expression (entry _ label _ vec) with the domain classification information.
Further, the inference layer includes: the preprocessing layer is used for determining original text expression according to the content characteristics of the current round; the first layer structure is used for determining the text expression with the domain information ratio according to the original text expression and the current round of domain word ratio characteristics, and the second layer structure is used for determining the text expression with the domain classification information according to the original text expression, the current round of domain word ratio characteristics and the weight of the domain classification information. The preprocessing layer can be an encoding layer, the first layer structure can be an attention layer, and attention processing can be carried out; the second layer structure may be a fully connected layer, which may be fully connected.
Step S221, inputting the content features of the current round and the term proportion features of the current round into the first layer structure of the inference layer, and obtaining the text expression with the domain information proportion and the text expression with the domain classification information may include:
and step S221a, inputting the content characteristics of the round to a preprocessing layer of the reasoning layer to obtain the original text expression. In actual implementation, as shown in fig. 2, the current round of content features (sent, such as forgetting water) are Bi-LSTM encoded to obtain the original text expression (sent veco).
And step S221b, determining the text expression with the domain information ratio according to the original text expression and the domain word ratio characteristics of the current round. In the actual implementation, as shown in fig. 2, the original text expression (sent _ veco) and the feature of the area word ratio (arc _ vec) of the current round are subjected to attention, so as to obtain the text expression (sent _ vec) with the area information ratio.
And step S221c, determining the text expression with the domain classification information according to the original text expression, the feature of the field word ratio in the current round and the weight of the domain classification information. In the actual implementation, as shown in fig. 2, an attribute is performed on the original text expression (sent _ veco), the local round of domain word proportion feature (arc _ vec), and the weight (label _ embedding) of the domain classification information, so as to obtain a text expression (attribute _ label _ vec) with domain classification information. Or, performing an attribute on the original text expression (sent _ veco) and the field word proportion feature (arc _ vec) of the current round, and performing the attribute on the result of the attribute and the weight (label _ embedding) of the field classification information to obtain a text expression (attribute _ label _ vec) with the field classification information.
And S222, inputting the text expression with the domain information proportion, the text expression with the domain classification information and the supervision information into a second layer structure of the reasoning layer to obtain the domain probability distribution corresponding to the interactive text of the current round.
Taking the domain prediction model shown in fig. 2 as an example, the text expression (send _ vec) with domain information proportion, the text expression (attribute _ label _ vec) with domain classification information, and the supervision information (his _ attribute _ vec) are input into the second layer structure of the inference layer, and the domain probability distribution (domain) corresponding to the interactive text of the round is obtained.
Specifically, the text expression (sent _ vec) with the area domain information ratio, the text expression (attribute _ label _ vec) with the area domain classification information and the supervision information (his _ attribute _ vec) are all connected through the second layer structure:
fc_vec＝[attention_label_vec,sent_vec,his_attention_vec]，
fc＝weight*(fc_vec*fc_vecT)+b。
the text expression with the domain classification information is determined based on the content characteristics of the current round, the word proportion characteristics of the current round and the weight of the domain classification information of the domain prediction model, and the weight of the domain classification information is determined according to each domain information learned by the domain prediction model in the training process. In other words, the weight of the domain classification information may be that the domain prediction model remains after the training is completed. Of course, the weight of the domain classification information may also be determined in other models, and then the weight of the domain classification information determined by other models is input to the domain prediction model.
In some preferred embodiments, step S200, inputting the current round of interactive text and supervision information into the domain prediction model, further includes: inputting the current round of interaction text, the supervision information and the personalized features into the domain prediction model, wherein the personalized features are used for representing the associated information in the current round of interaction.
Compared with the above embodiment, during input, the embodiment adds personalized features to deal with the situation that multiple basically same prediction results occur during field prediction, and the personalized features are added to assist the field prediction model to further distinguish the ambiguous results.
The personalized features refer to information which can be extracted from the current interaction of the user according to different model application scenes and has certain correlation with the current interaction, including but not limited to the states of other application equipment in the foreground and background states of the whole interaction equipment in the process of using the interaction equipment by the user. For example, when a user uses an interactive device to navigate related voice interaction, but other application apps such as music and panning are also started in the background of the interactive device of the user; then, at this time, the navigation field belongs to the foreground, and the music field and the shopping field belong to the background.
Further, the personalized features comprise foreground domain features of the current round of interaction and background domain features of the current round of interaction.
The matrix dimension of the personalized data coding output of the personalized features is 1 × 2label (short for personal _ matrix), and the 1 × 2label matrix is formed by splicing 21 × label matrixes. The first 1 × label represents the foreground field in the current round of interaction, if the navigation field is in the foreground in the current interaction process, the parameter of the navigation field is set to be 1, and the other parameters are all 0. The second 1 × cable represents the background field in the current round of interaction, and the parameters of all the background fields in the current interaction process are set to be 1, and the others are all 0.
In some embodiments, step S220 is to input the content features, the term proportion features, the personalized features, and the monitoring information of the current round into an inference layer of the domain prediction model, so as to obtain a domain probability distribution corresponding to the interactive text of the current round.
Taking fig. 2 as an example, the content features (forgetting water) of the current round, the domain word proportion features (domain word proportion) of the current round, the personalized features (personal _ matrix), and the supervision information (his _ attribution _ vec) are input into the inference layer of the domain prediction model, so as to obtain the domain probability distribution (domain) corresponding to the interactive text of the current round.
Step S222 is to input the text expression with the domain information ratio, the text expression with the domain classification information, the personalized feature (personal _ matrix), and the supervision information into the second layer structure of the inference layer, so as to obtain the domain probability distribution corresponding to the current round of interactive text.
Taking the domain prediction model shown in fig. 2 as an example, the personalized features (personal _ matrix) are converted into matrices persjvec, and the conversion method includes, but is not limited to, directly adding a first matrix used for representing the foreground domain in the current round of interaction and a second matrix used for representing the background domain in the current round of interaction; or the first matrix used for representing the foreground field in the current round of interaction and the second matrix used for representing the background field in the current round of interaction are added in a weighting mode.
The text expression with the area information proportion (sent _ vec), the text expression with the area classification information (attribute _ label _ vec), the personalized feature (pers _ vec) and the supervision information (his _ attribute _ vec) are connected in a whole way,
fc_vec＝[attention_label_vec,sent_vec,pers_vec,his_attention_vec]
fc＝weight*(fc_vec*fc_vecT)+b，
and obtaining the domain probability distribution (domain) corresponding to the interactive text of the current round.
In one embodiment of the present invention, a domain prediction method includes: inputting the previous round of interactive text and the domain information determined after semantic understanding of the previous round of interactive text into a first part of a domain prediction model to obtain supervision information, wherein the supervision information is used for representing the corrected domain probability distribution corresponding to the previous round of interactive text; inputting the current round of interactive text, personalized features and supervision information to a second part of the domain prediction model to obtain domain probability distribution corresponding to the current round of interactive text, wherein the personalized features are used for representing associated information in the current round of interaction; determining a domain prediction result based on the domain probability distribution corresponding to the current round of interactive text; the domain prediction model is obtained by taking multi-round interactive text data as a sample in advance and taking the predetermined domain probability distribution of multi-round interaction corresponding to the multi-round interactive text data as a sample label for training.
According to the method, on the basis of fully using the generalization capability of the deep neural network, the field of the expression content of the current round is jointly predicted by using the expression content of the previous round of the user, the final semantic understanding field information of the expression content of the previous round and the personalized features which can be extracted in the interaction process of the current round. Therefore, the problem that the model prediction accuracy is low in the multi-round interaction process in the existing scheme is solved.
The training process of the domain prediction model of the embodiment of the present invention is described below with reference to fig. 3.
As shown in fig. 3, the left part is the text of the previous round of interaction, and the right part is the text of the current round of interaction; in the scheme, the text of the previous round of interactive input is named his _ send, and the text of the current round of interactive input is named send.
In the left part, the his _ send of the previous round of interactive text of the user is firstly coded by Bi-LSTM to obtain his _ send _ veco, the his _ send _ veco and the his _ arc _ vec are directly subjected to attention to obtain sentence expression of the original text, the his _ send _ vec matrix reflects the proportion of the field information in the original text sentence.
his_sent_veco＝Bi-LSM(his_sent)
his_sent_vec＝attention(his_sent_veco,his_arc_vec)
Performing attribute on his _ send _ veco, his _ arc _ vec and his _ label _ embedding (the his _ label _ embedding is a matrix generated by the model in the training process, the matrix represents the information of each learned field when the model predicts, and the matrix dimension is label × 256), calculating the feature matrix of the previous round of interactive text on the field, and generating the his _ attribute _ label _ vec.
Finally, performing full connection operation on his _ attribution _ label _ vec and his _ sent _ vec, namely
In order to make the model better learn the weight of the previous round of interactive text, the above is used with a loss function,
wherein y isiThe score of the previous round of interactive text in each field is predicted, y is a matrix of 1 × label, and the higher the score of one bit of the matrix is, the higher the probability that the previous round of interactive text prediction belongs to the corresponding field is considered by the representation model; y isi-representing the model actual output.
By using the cross entropy loss function, the result predicted by the model is closer to the correct result when the model is trained.
As shown in fig. 3, based on the text expression with domain information ratio (his _ send _ vec) and the text expression with domain classification information (his _ attribution _ label _ vec), a domain probability distribution (his _ attribution _ veco) corresponding to the previous round of interactive text is determined.
In the right part, the previous processing logic is the same as that in the left part in the process of processing the interactive text of the current round. After acquiring the attribute _ label _ vec matrix of the current round of interaction, performing a full connection operation on his _ attribute _ veco, persjvec, sent _ vec and attribute _ label _ vec of the previous round of interaction text. Namely:
fc_vec＝[attention_label_vec,sent_vec,pers_vec,his_attention_veco]
fc＝weight*(fc_vec*fc_vecT)+b
similarly, in order to make the model better learn the wegiht of the interactive text of the current round, the same loss function as that of the previous round is used,
wherein JiRepresenting the score of the predicted interactive text of the current round on each domain, Ji-representing the model actual output.
In the training process, a domain probability distribution label his _ domain _ label corresponding to the previous round of interactive text and a domain probability distribution label domain _ label corresponding to the current round of interactive text need to be input.
The application process of the domain prediction model and the input network topology result of the model training process are slightly different, and the difference lies in that in the network topology structure of the previous round of interaction, the determined domain matrix his _ label _ matrix of the previous round of interaction text after semantic understanding needs to be input.
In an application network of a domain prediction model, when monitoring information his _ attribution _ vec is calculated, a his _ label _ vec matrix is added compared with the determination of his _ attribution _ veco in a training model. During actual model testing, the domain matrix of the previous round of interactive text after a series of semantic understanding is his _ label _ matrix, and a specific his _ label _ vec calculation formula is as follows:
his_label_vec＝his_label_matrixT*his_label_matrix
his_attention_vec＝[his_sent_vec,his_attention_label_vec,his_label_vec]
fc_vec＝[attention_label_vec,sent_vec,pers_vec,his_attention_vec]
fc＝weight*(fc_vec*fc_vecT)+b
multiplying the transposed matrix of the his _ label _ matrix with the his _ label _ matrix to obtain the his _ label _ vec, wherein the dimension of the matrix of the his _ label _ vec is increased, and the information quantity is enriched. When the domain prediction model predicts the domain of the interactive text of the current round, the purpose of his _ label _ vec is used for monitoring the model prediction result of the interaction of the current round. For example, in the previous round of interactive text prediction process, when the domain predicted by the previous round of domain prediction model is consistent with the domain after the previous round of semantic understanding (assumed to be a music domain), the weight belonging to the music domain in his _ send _ vec will be strengthened, and the weights of other domains will be weakened; when the field predicted by the previous round of field prediction model is inconsistent with the field after the previous round of semantic understanding (the model prediction is a music field, and the field of actual semantic understanding is a story field), the weight belonging to the story field in his _ sent _ vec is strengthened, the weight of other fields is weakened, and when the current round of interactive text is predicted to belong to a certain field, the music field is more unconfirmed, and the current round of interactive text is more believable to be the story field.
The domain prediction method provided by the embodiment of the invention can be used for jointly predicting the domain of the expression content of the current round by using the expression content of the previous round of the user, the domain information of the final semantic understanding of the expression content of the previous round and the personalized features which can be extracted in the interaction process of the current round on the basis of fully using the generalization capability of the deep neural network, so that the problem of low model prediction accuracy in the multi-round interaction process of the conventional scheme can be solved.
The domain prediction apparatus provided by the embodiment of the present invention is described below, and the domain prediction apparatus described below and the domain prediction method described above may be referred to correspondingly.
As shown in fig. 4, the domain prediction apparatus according to the embodiment of the present invention includes: a text determining unit 710, configured to determine a current round of interactive text; a probability distribution determining unit 720, configured to input the current round of interactive texts and the supervision information into the domain prediction model, and obtain domain probability distribution output by the domain prediction model and corresponding to the current round of interactive texts, where the supervision information is domain information determined after semantic understanding based on a previous round of interactive texts, and is obtained by correcting the domain probability distribution output by the domain prediction model and corresponding to the previous round of interactive texts; the domain determining unit 730 is configured to determine a domain prediction result based on a domain probability distribution corresponding to the current round of interactive text; the domain prediction model is obtained by taking multi-round interactive text data as samples in advance and taking predetermined domain probability distribution respectively corresponding to the multi-round interactive text data as sample labels for training.
Fig. 5 illustrates a physical structure diagram of an electronic device, which may include, as shown in fig. 5: a processor (processor)810, a communication Interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication Interface 820 and the memory 830 communicate with each other via the communication bus 840. The processor 810 may call logic instructions in the memory 830 to perform a domain prediction method, the method comprising: determining the interactive text of the current round; inputting the current round of interactive texts and supervision information into a field prediction model to obtain field probability distribution output by the field prediction model and corresponding to the current round of interactive texts, wherein the supervision information is obtained by correcting the field probability distribution output by the field prediction model and corresponding to the previous round of interactive texts based on the field information determined after semantic understanding of the previous round of interactive texts; determining a domain prediction result based on the domain probability distribution corresponding to the current round of interactive texts; the field prediction model is obtained by taking multi-round interactive text data as samples in advance and taking predetermined field probability distribution data respectively corresponding to the multi-round interactive text data as sample labels for training.
It should be noted that, when being implemented specifically, the electronic device in this embodiment may be a server, a PC, or other devices, as long as the structure includes the processor 810, the communication interface 820, the memory 830, and the communication bus 840 shown in fig. 5, where the processor 810, the communication interface 820, and the memory 830 complete mutual communication through the communication bus 840, and the processor 810 may call the logic instructions in the memory 830 to execute the above method. The embodiment does not limit the specific implementation form of the electronic device.
In addition, the logic instructions in the memory 830 may be implemented in software functional units and stored in a computer readable storage medium when the logic instructions are sold or used as independent products. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
Further, an embodiment of the present invention discloses a computer program product, the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, when the program instructions are executed by a computer, the computer can execute the execution domain prediction method provided by the above method embodiments, the method includes: determining the interactive text of the current round; inputting the current round of interactive texts and supervision information into a field prediction model to obtain field probability distribution output by the field prediction model and corresponding to the current round of interactive texts, wherein the supervision information is obtained by correcting the field probability distribution output by the field prediction model and corresponding to the previous round of interactive texts based on the field information determined after semantic understanding of the previous round of interactive texts; determining a domain prediction result based on the domain probability distribution corresponding to the current round of interactive texts; the field prediction model is obtained by taking multi-round interactive text data as samples in advance and taking predetermined field probability distribution data respectively corresponding to the multi-round interactive text data as sample labels for training.
In another aspect, an embodiment of the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored, where the computer program is implemented to, when executed by a processor, perform the execution domain prediction method provided in the foregoing embodiments, where the method includes: determining the interactive text of the current round; inputting the current round of interactive texts and supervision information into a field prediction model to obtain field probability distribution output by the field prediction model and corresponding to the current round of interactive texts, wherein the supervision information is obtained by correcting the field probability distribution output by the field prediction model and corresponding to the previous round of interactive texts based on the field information determined after semantic understanding of the previous round of interactive texts; determining a domain prediction result based on the domain probability distribution corresponding to the current round of interactive texts; the field prediction model is obtained by taking multi-round interactive text data as samples in advance and taking predetermined field probability distribution data respectively corresponding to the multi-round interactive text data as sample labels for training.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A domain prediction method, comprising:
determining the interactive text of the current round;
inputting the current round of interactive texts and supervision information into a field prediction model to obtain field probability distribution output by the field prediction model and corresponding to the current round of interactive texts, wherein the supervision information is obtained by correcting the field probability distribution output by the field prediction model and corresponding to the previous round of interactive texts based on the field information determined after semantic understanding of the previous round of interactive texts;
determining a domain prediction result based on the domain probability distribution corresponding to the current round of interactive texts;
the domain prediction model is obtained by taking multi-round interactive text data as samples and taking predetermined domain probability distribution data respectively corresponding to the multi-round interactive text data as sample labels for training.
2. The domain prediction method of claim 1, wherein the supervisory information is obtained by:
obtaining the domain information determined after semantic understanding of the previous round of interactive text;
inputting the previous round of interactive text into the field prediction model to obtain field probability distribution output by the field prediction model and corresponding to the previous round of interactive text;
and determining the supervision information based on the domain information and the domain probability distribution corresponding to the previous round of interactive text.
3. The method of claim 1, wherein the inputting the current round of interactive texts and the supervision information into a domain prediction model to obtain a domain probability distribution output by the domain prediction model and corresponding to the current round of interactive texts comprises:
inputting the current round of interactive text into a preprocessing layer of the field prediction model to obtain current round content characteristics and current round field word proportion characteristics, wherein the current round content characteristics are used for representing the expression content of the current round of interactive text, and the current round field word proportion characteristics are used for representing the proportion of the length of each field entity of the current round of interactive text in the current round of interactive text;
and inputting the content characteristics, the field word proportion characteristics and the supervision information of the current round into an inference layer of the field prediction model to obtain the field probability distribution corresponding to the interactive text of the current round.
4. The domain prediction method of claim 3, wherein the inputting the content features of the current round, the domain word proportion features of the current round and the supervision information into an inference layer of the domain prediction model to obtain the domain probability distribution corresponding to the interactive text of the current round comprises:
inputting the content features and the field word proportion features of the current round into a first layer structure of the reasoning layer to obtain a text expression with field information proportion and a text expression with field classification information, wherein the text expression with field information proportion is used for representing the field information of the interactive text of the current round, and the text expression with field classification information is used for representing the predicted field proportion weight of the interactive text of the current round;
and inputting the text expression with the domain information ratio, the text expression with the domain classification information and the supervision information into a second layer structure of the reasoning layer to obtain the domain probability distribution corresponding to the current round of interactive text.
5. The domain prediction method according to claim 4, wherein the text expression with the domain classification information is determined based on the content features of the current round, the domain word proportion features of the current round, and the weight of the domain classification information of the domain prediction model, and the weight of the domain classification information is determined according to each domain information learned by the domain prediction model in a training process.
6. The domain prediction method of any of claims 1-5, wherein the inputting the current round of interactive text and supervisory information into a domain prediction model comprises:
inputting the current round of interaction text, the supervision information and personalized features into the field prediction model, wherein the personalized features are used for representing associated information in the current round of interaction.
7. The domain prediction method of claim 6, wherein the personalized features comprise foreground domain features of the current round of interaction and background domain features of the current round of interaction.
8. A domain prediction apparatus, comprising:
the text determining unit is used for determining the interactive text of the current round;
a probability distribution determining unit, configured to input the current round of interactive texts and monitoring information into a domain prediction model, and obtain domain probability distribution output by the domain prediction model and corresponding to the current round of interactive texts, where the monitoring information is domain information determined after semantic understanding based on a previous round of interactive texts, and is obtained by correcting the domain probability distribution output by the domain prediction model and corresponding to the previous round of interactive texts;
the domain determining unit is used for determining a domain prediction result based on the domain probability distribution corresponding to the current round of interactive texts; wherein
The field prediction model is obtained by taking multi-round interactive text data as samples in advance and taking predetermined field probability distribution respectively corresponding to the multi-round interactive text data as sample labels for training.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the domain prediction method according to any of claims 1 to 8 when executing the program.
10. A non-transitory computer readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the domain prediction method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911327989.6A CN111091011B (en) | 2019-12-20 | 2019-12-20 | Domain prediction method, domain prediction device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911327989.6A CN111091011B (en) | 2019-12-20 | 2019-12-20 | Domain prediction method, domain prediction device and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111091011A true CN111091011A (en) | 2020-05-01 |
| CN111091011B CN111091011B (en) | 2023-07-28 |
Family
ID=70395878
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911327989.6A Active CN111091011B (en) | 2019-12-20 | 2019-12-20 | Domain prediction method, domain prediction device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111091011B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111639160A (en) * | 2020-05-29 | 2020-09-08 | 达闼机器人有限公司 | Domain identification method, interaction method, electronic device and storage medium |
Citations (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050177414A1 (en) * | 2004-02-11 | 2005-08-11 | Sigma Dynamics, Inc. | Method and apparatus for automatically and continuously pruning prediction models in real time based on data mining |
| US20160364382A1 (en) * | 2015-06-15 | 2016-12-15 | Microsoft Technology Licensing, Llc | Contextual language generation by leveraging language understanding |
| CN106530305A (en) * | 2016-09-23 | 2017-03-22 | 北京市商汤科技开发有限公司 | Semantic segmentation model training and image segmentation method and device, and calculating equipment |
| CN106649739A (en) * | 2016-12-23 | 2017-05-10 | 深圳市空谷幽兰人工智能科技有限公司 | Multi-round interactive information inheritance recognition method, apparatus and interactive system |
| CN106776578A (en) * | 2017-01-03 | 2017-05-31 | 竹间智能科技（上海）有限公司 | Talk with the method and device of performance for lifting conversational system |
| CN107632987A (en) * | 2016-07-19 | 2018-01-26 | 腾讯科技（深圳）有限公司 | One kind dialogue generation method and device |
| CN108153780A (en) * | 2016-12-05 | 2018-06-12 | 阿里巴巴集团控股有限公司 | A kind of human-computer dialogue device and its interactive method of realization |
| CN109582767A (en) * | 2018-11-21 | 2019-04-05 | 北京京东尚科信息技术有限公司 | Conversational system processing method, device, equipment and readable storage medium storing program for executing |
| CN109725961A (en) * | 2017-10-31 | 2019-05-07 | 百度（美国）有限责任公司 | The system and method that execution task is inputted based on user using natural language processing |
| CN109857843A (en) * | 2018-12-25 | 2019-06-07 | 科大讯飞股份有限公司 | Exchange method and system based on document |
| CN109960749A (en) * | 2019-02-22 | 2019-07-02 | 清华大学 | Model acquisition methods, keyword generation method, device, medium and calculating equipment |
| CN109977209A (en) * | 2019-03-22 | 2019-07-05 | 深圳狗尾草智能科技有限公司 | More wheel man-machine interaction methods, system, computer and medium |
| WO2019136946A1 (en) * | 2018-01-15 | 2019-07-18 | 中山大学 | Deep learning-based weakly supervised salient object detection method and system |
| US20190236132A1 (en) * | 2018-01-26 | 2019-08-01 | Ge Inspection Technologies, Lp | Generating natural language recommendations based on an industrial language model |
| CN110188204A (en) * | 2019-06-11 | 2019-08-30 | 腾讯科技（深圳）有限公司 | A kind of extension corpora mining method, apparatus, server and storage medium |
| CN110209791A (en) * | 2019-06-12 | 2019-09-06 | 百融云创科技股份有限公司 | It is a kind of to take turns dialogue intelligent speech interactive system and device more |
| US20190294973A1 (en) * | 2018-03-23 | 2019-09-26 | Google Llc | Conversational turn analysis neural networks |
| US20190384810A1 (en) * | 2018-06-15 | 2019-12-19 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Method of training a descriptive text generating model, and method and apparatus for generating descriptive text |
-
2019
- 2019-12-20 CN CN201911327989.6A patent/CN111091011B/en active Active
Patent Citations (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050177414A1 (en) * | 2004-02-11 | 2005-08-11 | Sigma Dynamics, Inc. | Method and apparatus for automatically and continuously pruning prediction models in real time based on data mining |
| US20160364382A1 (en) * | 2015-06-15 | 2016-12-15 | Microsoft Technology Licensing, Llc | Contextual language generation by leveraging language understanding |
| US20180285348A1 (en) * | 2016-07-19 | 2018-10-04 | Tencent Technology (Shenzhen) Company Limited | Dialog generation method, apparatus, and device, and storage medium |
| CN107632987A (en) * | 2016-07-19 | 2018-01-26 | 腾讯科技（深圳）有限公司 | One kind dialogue generation method and device |
| CN106530305A (en) * | 2016-09-23 | 2017-03-22 | 北京市商汤科技开发有限公司 | Semantic segmentation model training and image segmentation method and device, and calculating equipment |
| CN108153780A (en) * | 2016-12-05 | 2018-06-12 | 阿里巴巴集团控股有限公司 | A kind of human-computer dialogue device and its interactive method of realization |
| CN106649739A (en) * | 2016-12-23 | 2017-05-10 | 深圳市空谷幽兰人工智能科技有限公司 | Multi-round interactive information inheritance recognition method, apparatus and interactive system |
| CN106776578A (en) * | 2017-01-03 | 2017-05-31 | 竹间智能科技（上海）有限公司 | Talk with the method and device of performance for lifting conversational system |
| CN109725961A (en) * | 2017-10-31 | 2019-05-07 | 百度（美国）有限责任公司 | The system and method that execution task is inputted based on user using natural language processing |
| WO2019136946A1 (en) * | 2018-01-15 | 2019-07-18 | 中山大学 | Deep learning-based weakly supervised salient object detection method and system |
| US20190236132A1 (en) * | 2018-01-26 | 2019-08-01 | Ge Inspection Technologies, Lp | Generating natural language recommendations based on an industrial language model |
| US20190294973A1 (en) * | 2018-03-23 | 2019-09-26 | Google Llc | Conversational turn analysis neural networks |
| US20190384810A1 (en) * | 2018-06-15 | 2019-12-19 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Method of training a descriptive text generating model, and method and apparatus for generating descriptive text |
| CN109582767A (en) * | 2018-11-21 | 2019-04-05 | 北京京东尚科信息技术有限公司 | Conversational system processing method, device, equipment and readable storage medium storing program for executing |
| CN109857843A (en) * | 2018-12-25 | 2019-06-07 | 科大讯飞股份有限公司 | Exchange method and system based on document |
| CN109960749A (en) * | 2019-02-22 | 2019-07-02 | 清华大学 | Model acquisition methods, keyword generation method, device, medium and calculating equipment |
| CN109977209A (en) * | 2019-03-22 | 2019-07-05 | 深圳狗尾草智能科技有限公司 | More wheel man-machine interaction methods, system, computer and medium |
| CN110188204A (en) * | 2019-06-11 | 2019-08-30 | 腾讯科技（深圳）有限公司 | A kind of extension corpora mining method, apparatus, server and storage medium |
| CN110209791A (en) * | 2019-06-12 | 2019-09-06 | 百融云创科技股份有限公司 | It is a kind of to take turns dialogue intelligent speech interactive system and device more |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111639160A (en) * | 2020-05-29 | 2020-09-08 | 达闼机器人有限公司 | Domain identification method, interaction method, electronic device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111091011B (en) | 2023-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110704641B (en) | Ten-thousand-level intention classification method and device, storage medium and electronic equipment | |
| CN111858854B (en) | Question-answer matching method and relevant device based on historical dialogue information | |
| CN111339305A (en) | Text classification method and device, electronic equipment and storage medium | |
| US11636272B2 (en) | Hybrid natural language understanding | |
| EP4113357A1 (en) | Method and apparatus for recognizing entity, electronic device and storage medium | |
| CN108710704A (en) | Determination method, apparatus, electronic equipment and the storage medium of dialogue state | |
| CN113094481A (en) | Intention recognition method and device, electronic equipment and computer readable storage medium | |
| CN111027292B (en) | Method and system for generating limited sampling text sequence | |
| CN117332072B (en) | Dialogue processing, voice abstract extraction and target dialogue model training method | |
| CN110942774A (en) | Man-machine interaction system, and dialogue method, medium and equipment thereof | |
| CN115497465A (en) | Voice interaction method and device, electronic equipment and storage medium | |
| CN113111190A (en) | Knowledge-driven dialog generation method and device | |
| CN114239607A (en) | Conversation reply method and device | |
| CN111241843A (en) | Semantic relation inference system and method based on composite neural network | |
| CN111091011A (en) | Domain prediction method, domain prediction device and electronic equipment | |
| CN117271745A (en) | Information processing method and device, computing equipment and storage medium | |
| CN109002498B (en) | Man-machine conversation method, device, equipment and storage medium | |
| CN112818688B (en) | Text processing method, device, equipment and storage medium | |
| CN114333790A (en) | Data processing method, device, equipment, storage medium and program product | |
| CN115934920B (en) | Model training method for man-machine conversation and related device | |
| CN114942980B (en) | Method and device for determining text matching | |
| CN117453899B (en) | Intelligent dialogue system and method based on large model and electronic equipment | |
| CN117453895B (en) | Intelligent customer service response method, device, equipment and readable storage medium | |
| CN118136020A (en) | Voice recognition method and device | |
| CN115617997A (en) | Dialog state tracking method, device, equipment and medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |