CN116860972A - Interactive information classification method, device, apparatus, storage medium and program product - Google Patents
Interactive information classification method, device, apparatus, storage medium and program product Download PDFInfo
- Publication number
- CN116860972A CN116860972A CN202310826323.5A CN202310826323A CN116860972A CN 116860972 A CN116860972 A CN 116860972A CN 202310826323 A CN202310826323 A CN 202310826323A CN 116860972 A CN116860972 A CN 116860972A
- Authority
- CN
- China
- Prior art keywords
- vector
- dialogue
- data
- sample
- word
- 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 48
- 230000002452 interceptive effect Effects 0.000 title claims abstract description 24
- 239000013598 vector Substances 0.000 claims abstract description 587
- 238000012545 processing Methods 0.000 claims abstract description 69
- 238000004590 computer program Methods 0.000 claims abstract description 26
- 238000011176 pooling Methods 0.000 claims description 29
- 238000013528 artificial neural network Methods 0.000 claims description 25
- 230000006870 function Effects 0.000 claims description 24
- 238000013527 convolutional neural network Methods 0.000 claims description 11
- 230000004913 activation Effects 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 6
- 125000004122 cyclic group Chemical group 0.000 claims description 5
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 11
- 238000012549 training Methods 0.000 description 11
- 239000011159 matrix material Substances 0.000 description 8
- 230000008569 process Effects 0.000 description 8
- 238000004891 communication Methods 0.000 description 5
- 239000003795 chemical substances by application Substances 0.000 description 4
- 230000011218 segmentation Effects 0.000 description 4
- 238000013145 classification model Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000003993 interaction Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000013500 data storage Methods 0.000 description 2
- 101150086656 dim1 gene Proteins 0.000 description 2
- 239000010970 precious metal Substances 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 102100032202 Cornulin Human genes 0.000 description 1
- 101000920981 Homo sapiens Cornulin Proteins 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 229910021389 graphene Inorganic materials 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/353—Clustering; Classification into predefined classes
-
- 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
- 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
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Machine Translation (AREA)
Abstract
The application relates to an interactive information classification method, an interactive information classification device, a computer device, a storage medium and a computer program product. Relates to the technical field of artificial intelligence. The method comprises the following steps: acquiring dialogue data comprising dialogue text data, dialogue attribution data and dialogue object data; acquiring a first word vector and a first word vector for dialogue text data; vector splicing processing is carried out on the first word vector and the first word vector to obtain a first spliced vector; acquiring a first dialogue attribution vector and a first dialogue object vector according to dialogue attribution data and dialogue object data; acquiring a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector; inputting the first vector into the trained intention recognition model to obtain a predictive probability vector; and determining the category label of the dialogue data according to the predictive probability vector. By adopting the method, the accuracy of the category label prediction of the dialogue data can be improved.
Description
Technical Field
The present application relates to the field of artificial intelligence technology, and in particular, to an interactive information classification method, apparatus, computer device, storage medium, and computer program product.
Background
In the conventional technology, a word set with word frequency larger than a certain threshold is defined, and dialog text is identified according to the word set and a model to identify dialog category, so that dialog intention is known.
However, once words and sentences which are not included in the word set are input, and only the word vector is used for recognition, recognition errors are easy to cause, and the accuracy of dialog class recognition is low.
Disclosure of Invention
In view of the foregoing, it is desirable to provide an interactive information classification method, apparatus, computer device, computer readable storage medium, and computer program product that can improve accuracy of category label prediction for dialogue data.
In a first aspect, the present application provides a method for classifying interaction information. The method comprises the following steps:
acquiring dialogue data; the dialogue data comprises dialogue text data, dialogue attribution data and dialogue object data;
acquiring a first word vector and a first word vector for the dialogue text data;
vector splicing processing is carried out on the first word vector and the first word vector to obtain a first spliced vector;
acquiring a first dialogue attribution vector and a first dialogue object vector according to the dialogue attribution data and the dialogue object data;
Acquiring a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector;
inputting the first vector into a trained intention recognition model to obtain a predictive probability vector; the predictive probability vector comprises probability values corresponding to a plurality of tag information;
and determining the category label of the dialogue data according to the predictive probability vector.
In one embodiment, the obtaining the first word vector and the first word vector for the dialog text data includes:
inputting the dialogue text data into a pre-trained word embedding model to obtain a first word vector aiming at the dialogue text data;
and inputting the dialogue text data into a word embedding model trained in advance to obtain a first word vector aiming at the dialogue text data.
In one embodiment, the intent recognition model includes a neural network layer, a pooling layer, and a vector processing layer; the neural network layer includes a convolutional neural network CNN or a cyclic neural network RNN, and the inputting the first vector into the trained intent recognition model to obtain a predictive probability vector includes:
Inputting the first vector to the neural network layer to obtain a second vector;
inputting the second vector to the pooling layer to respectively carry out maximum pooling treatment and average pooling treatment to obtain a third vector and a fourth vector;
and inputting the third vector and the fourth vector into the vector processing layer to perform addition processing linear processing and activation function processing to obtain the predictive probability vector.
In one embodiment, the determining the category label of the dialogue data according to the predictive probability vector includes:
and under the condition that the probability value corresponding to the label information exists in the predictive probability vector and is larger than a first threshold value, determining the corresponding label information as the class label of the dialogue data.
In one embodiment, the method further comprises:
acquiring a dialogue sample data set; the dialogue sample data set comprises a plurality of dialogue samples and category labels corresponding to the dialogue samples; the dialogue samples comprise dialogue text samples, dialogue attribution samples and dialogue object samples;
preprocessing the dialogue text sample, the dialogue attribution sample and the dialogue object sample to obtain a dialogue vector sample;
And adjusting parameters of the intention recognition model according to the dialogue vector sample and the category label corresponding to the dialogue sample until the intention recognition model converges to obtain a trained intention recognition model.
In one embodiment, the preprocessing the dialogue text sample, the dialogue attribution sample and the dialogue object sample to obtain a dialogue vector sample includes:
according to the dialogue text sample, acquiring a word vector sample and a word vector sample aiming at the dialogue text sample;
vector splicing processing is carried out on the word vector samples and the word vector samples to obtain spliced vector samples;
according to the dialogue attribution sample and the dialogue object sample, a dialogue attribution vector sample and a dialogue object vector sample are obtained;
and obtaining a dialogue vector sample according to the spliced vector sample, the dialogue attribution vector sample and the dialogue object vector sample.
In a second aspect, the application further provides an interactive information classification device. The device comprises:
the dialogue data acquisition module is used for acquiring dialogue data; the dialogue data comprises dialogue text data, dialogue attribution data and dialogue object data;
The first vector acquisition module is used for acquiring a first word vector and a first word vector aiming at the dialogue text data according to the dialogue text data;
the first splicing vector obtaining module is used for carrying out vector splicing processing on the first word vector and the first word vector to obtain a first splicing vector;
the second vector acquisition module is used for acquiring a first dialogue attribution vector and a first dialogue object vector according to the dialogue attribution data and the dialogue object data;
the first vector obtaining module is used for obtaining a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector;
the prediction probability vector obtaining module is used for inputting the first vector into the trained intention recognition model to obtain a prediction probability vector; the predictive probability vector comprises probability values corresponding to a plurality of tag information;
and the category label determining module is used for determining the category label of the dialogue data according to the prediction probability vector.
In a third aspect, the present application also provides a computer device. The computer device comprises a memory storing a computer program and a processor which when executing the computer program performs the steps of:
Acquiring dialogue data; the dialogue data comprises dialogue text data, dialogue attribution data and dialogue object data;
acquiring a first word vector and a first word vector for the dialogue text data;
vector splicing processing is carried out on the first word vector and the first word vector to obtain a first spliced vector;
acquiring a first dialogue attribution vector and a first dialogue object vector according to the dialogue attribution data and the dialogue object data;
acquiring a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector;
inputting the first vector into a trained intention recognition model to obtain a predictive probability vector; the predictive probability vector comprises probability values corresponding to a plurality of tag information;
and determining the category label of the dialogue data according to the predictive probability vector.
In a fourth aspect, the present application also provides a computer-readable storage medium. The computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:
acquiring dialogue data; the dialogue data comprises dialogue text data, dialogue attribution data and dialogue object data;
Acquiring a first word vector and a first word vector for the dialogue text data;
vector splicing processing is carried out on the first word vector and the first word vector to obtain a first spliced vector;
acquiring a first dialogue attribution vector and a first dialogue object vector according to the dialogue attribution data and the dialogue object data;
acquiring a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector;
inputting the first vector into a trained intention recognition model to obtain a predictive probability vector; the predictive probability vector comprises probability values corresponding to a plurality of tag information;
and determining the category label of the dialogue data according to the predictive probability vector.
In a fifth aspect, the present application also provides a computer program product. The computer program product comprises a computer program which, when executed by a processor, implements the steps of:
acquiring dialogue data; the dialogue data comprises dialogue text data, dialogue attribution data and dialogue object data;
acquiring a first word vector and a first word vector for the dialogue text data;
Vector splicing processing is carried out on the first word vector and the first word vector to obtain a first spliced vector;
acquiring a first dialogue attribution vector and a first dialogue object vector according to the dialogue attribution data and the dialogue object data;
acquiring a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector;
inputting the first vector into a trained intention recognition model to obtain a predictive probability vector; the predictive probability vector comprises probability values corresponding to a plurality of tag information;
and determining the category label of the dialogue data according to the predictive probability vector.
The interactive information classification method, the interactive information classification device, the computer equipment, the storage medium and the computer program product acquire dialogue data comprising dialogue text data, dialogue attribution data and dialogue object data; acquiring a first word vector and a first word vector for dialogue text data; vector splicing processing is carried out on the first word vector and the first word vector to obtain a first spliced vector; acquiring a first dialogue attribution vector and a first dialogue object vector according to dialogue attribution data and dialogue object data; acquiring a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector; inputting the first vector into the trained intention recognition model to obtain a predictive probability vector; and determining the category label of the dialogue data according to the predictive probability vector. By adopting the method, the category prediction accuracy of the dialogue data can be improved by respectively converting the dialogue text data into the first word vector and the first word vector; after conversation attribution data is converted into a first conversation attribution vector, the first vector is obtained, the influence of different dialect habits possibly existing in clients in different areas on the prediction precision of conversation data categories is effectively improved, the semantic understanding of a model on wrongly written words which are wrongly translated due to the influence of the dialect is facilitated, and the accuracy of predicting the category of the conversation data by using the model is improved.
Drawings
FIG. 1 is an application environment diagram of an interactive information classification method in one embodiment;
FIG. 2 is a flow diagram of a method of classifying interaction information in one embodiment;
FIG. 3 is a block diagram of a model that is intended to be identified in one embodiment;
FIG. 4 is a training flow diagram of an intent recognition model in one embodiment;
FIG. 5 is a block diagram of a classification model in one embodiment;
FIG. 6 is a flow chart of an interactive information classification method according to another embodiment;
FIG. 7 is a block diagram of an interactive information classification device in one embodiment;
fig. 8 is an internal structural diagram of a computer device in one embodiment.
Detailed Description
The present application will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.
The interactive information classification method provided by the embodiment of the application can be applied to an application environment shown in fig. 1. Wherein the terminal 102 communicates with the server 104 via a network. The data storage system may store data that the server 104 needs to process. The data storage system may be integrated on the server 104 or may be located on a cloud or other network server.
The terminal 102 acquires dialogue data including dialogue text data, dialogue attribution data, and dialogue object data from the server 104 to acquire a first word vector and a first word vector for the dialogue text data; the terminal 102 performs vector splicing processing on the first word vector and the first word vector to obtain a first spliced vector; acquiring a first dialogue attribution vector and a first dialogue object vector according to dialogue attribution data and dialogue object data; acquiring a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector; inputting the first vector into the trained intention recognition model to obtain a predictive probability vector; and determining the category label of the dialogue data according to the predictive probability vector.
The terminal 102 may be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, internet of things devices, and portable wearable devices, where the internet of things devices may be smart speakers, smart televisions, smart air conditioners, smart vehicle devices, and the like. The portable wearable device may be a smart watch, smart bracelet, headset, or the like. The server 104 may be implemented as a stand-alone server or as a server cluster of multiple servers.
In one embodiment, as shown in fig. 2, a method for classifying interaction information is provided, which is illustrated by taking the application of the method to the terminal 102 in fig. 1 as an example, and includes the following steps:
step 202, obtaining dialogue data; the dialogue data includes dialogue text data, dialogue attribution data, and dialogue object data.
Wherein the dialogue data includes dialogue text data, dialogue attribution data and dialogue object data. The dialogue data may be customer call data, mainly dialogue records of customers and agents. Specifically, dialogue text data can be obtained by acquiring dialogue audio information, converting the audio information into text information, and denoising the text information.
The conversation home location data may be a customer's incoming call home location. The dialog object data may include users and agents.
For example, the terminal may acquire the audio information of a session and the home location data of the session to acquire session data of the session.
Step 204, obtaining a first word vector and a first word vector for the dialog text data.
Wherein the first word vector is derived based on the dialog text data and the first word vector is derived based on the dialog text data.
Illustratively, the manner of converting the word vector may be converted using parameters of the word embedded portion of ERNIE3.0 (Enhanced Language Representation with Informative Entities, pre-trained model); in particular, the dialog text data is converted into a word id sequence and a word matrix is formed on the basis of the word id sequence and the parameters of the word embedding part.
The word vector may be converted using the word-embedded parameters of an N-Gram (language model) model. The dialogue text data is separated by a word segmentation tool and converted into word id sequences, and then a word matrix is formed according to the word id sequences and parameters of a word embedding part. The word segmentation tool may be a jieba word segmentation tool or the like.
In one example, for example, "hello, i want to consult. This sentence is converted into a word id sequence [1891, 748, 37, 1175, 1214, 922, 2068, 18, 175, 4711], which, based on the parameters of the word id sequence and the word embedding part, forms a 10 xdim 0 word matrix, dim0 being the corresponding dimension of the word embedding, 1024 in ERNIE 3.0. The sentence is divided by word segmentation tools and converted into word id sequences [25800, 25800,0, 71, 462, 3205, 3205, 1443, 1443,2], and then corresponding line vectors are taken according to the word id sequences to form a word matrix of 10 x dim1, wherein dim1 is a dimension corresponding to word embedding, and is generally 300. Note that when two words constitute a word, their corresponding word vectors are copied twice.
For example, the terminal may acquire a first word vector and a first word vector for the dialog text data, respectively.
And 206, performing vector splicing processing on the first word vector and the first word vector to obtain a first spliced vector.
The first splicing vector is obtained by carrying out vector splicing processing on the first word vector and the first word vector. The first splice vector is a splice vector for dialog text data. The specific operation of the vector splicing process may be to splice the matrix corresponding to the first word vector and the matrix corresponding to the first word vector together by columns.
For example, the terminal may perform vector concatenation processing on the first word vector and the first word vector for the dialog text data to obtain a first concatenated vector.
Step 208, obtaining a first dialogue attribution vector and a first dialogue object vector according to the dialogue attribution data and the dialogue object data.
The first dialogue attribution vector is obtained by vectorizing dialogue attribution data. The first dialog object vector is vectorized by dialog object data.
For example, the terminal may vectorize the dialogue home data and the dialogue object data to obtain a first dialogue home vector and a first dialogue object vector.
Step 210, obtaining a first vector according to the first stitching vector, the first dialogue attribution vector and the first dialogue object vector.
The first vector is a vector for dialogue data, and can be obtained by vector addition of a first stitching vector, a first dialogue attribution vector and a first dialogue object vector.
For example, the terminal may obtain the first vector according to the first concatenation vector, the first session attribution vector, and the first session object vector.
Step 212, inputting the first vector into the trained intention recognition model to obtain a predictive probability vector; the predictive probability vector includes probability values corresponding to a plurality of tag information.
The intention recognition model is obtained through deep learning training and is used for carrying out a prediction model of a dialogue type. The predictive probability vector may include a plurality of tag information and a probability value corresponding to each tag information. The predictive probability vector is used to reflect tags that may exist for the dialog, as well as types that may not exist. The probability value is a number between 0 and 1.
For example, the terminal may input the first vector to the trained intent recognition model to obtain the predictive probability vector.
Step 214, determining the category label of the dialogue data according to the prediction probability vector.
The category labels of the dialogue data can comprise a plurality of category labels, and particularly, regarding banking, the category labels can be adjusted according to business requirements, and 14 category labels mainly exist, wherein the category labels comprise other category labels, personal mobile phone banking, account and debit cards, work and bank messengers, accounting queries, account opening rows and site information queries, funds, personal credit, transfer funds, deposit, personal internet banking, financial accounting, self-service machines and precious metals. Since a customer may consult multiple questions in one call, there may be more than one category label corresponding to one session data.
For example, the terminal may determine a category label of the dialogue data based on the predictive probability vector. Specifically, a probability threshold may be set to determine a category label for the dialogue data based on the probability threshold and the predictive probability vector.
In the interactive information classification method, dialogue data comprising dialogue text data, dialogue attribution data and dialogue object data is obtained; acquiring a first word vector and a first word vector for dialogue text data; vector splicing processing is carried out on the first word vector and the first word vector to obtain a first spliced vector; acquiring a first dialogue attribution vector and a first dialogue object vector according to dialogue attribution data and dialogue object data; acquiring a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector; inputting the first vector into the trained intention recognition model to obtain a predictive probability vector; and determining the category label of the dialogue data according to the predictive probability vector. By adopting the method, the category prediction accuracy of the dialogue data can be improved by respectively converting the dialogue text data into the first word vector and the first word vector; after conversation attribution data is converted into a first conversation attribution vector, the first vector is obtained, the influence of different dialect habits possibly existing in clients in different areas on the prediction precision of conversation data categories is effectively improved, the semantic understanding of a model on wrongly written words which are wrongly translated due to the influence of the dialect is facilitated, and the accuracy of predicting the category of the conversation data by using the model is improved.
In one embodiment, step 204 includes:
step 2042, inputting the dialogue text data into a pre-trained word embedding model, and obtaining a first word vector for the dialogue text data.
And 2044, inputting the dialogue text data into a pre-trained word embedding model to obtain a first word vector aiming at the dialogue text data.
The word embedding model may use word embedding parameters of the ERNIE3.0 model, which may be used to convert dialog text data into a first word vector. The word embedding model can use word vector parameters of an N-Gram model with hundred degrees trained based on corpus such as hundred degrees encyclopedia and the like, and can be used for converting dialogue text data into a first word vector.
The terminal inputs dialogue text data to a word embedding model trained in advance to obtain a first word vector for the dialogue text data, and inputs the dialogue text data to a word embedding model trained in advance to obtain a first word vector for the dialogue text data.
In the above embodiment, the dialogue text data is respectively converted into the first word vector and the first word vector, so that the accuracy of category prediction of the dialogue data can be improved. In addition, the word embedding model and the word embedding model directly use part of parameters in the existing model, so that the problem that the existing model occupies more hardware resources due to large parameter quantity is avoided; and the existing model uses position coding, the position coding is limited to the length of the character input at one time, and only partial parameters of the existing model are used, so that the input of any character length can be supported.
In one embodiment, the intent recognition model includes a neural network layer, a pooling layer, and a vector processing layer; the neural network layer includes a convolutional neural network CNN or a cyclic neural network RNN, and step 212 includes:
step 2122, inputting the first vector to the neural network layer to obtain a second vector.
The first vector may be obtained by vector addition of the first stitching vector, the first session home vector, and the first session object vector. The second vector is a text feature vector.
And 2124, inputting the second vector to the pooling layer to respectively perform maximum pooling treatment and average pooling treatment to obtain a third vector and a fourth vector.
Wherein the third vector is obtained by performing a maximum pooling process on the second vector. The fourth vector is averaged and pooled from the second vector.
And step 2126, inputting the third vector and the fourth vector to the vector processing layer to perform addition processing linear processing and activation function processing to obtain the prediction probability vector.
Wherein the activation function may be a tanh function. The predictive probability vector includes probability values corresponding to the plurality of tag information.
Illustratively, referring to FIG. 3, a block diagram of an intent recognition model is shown. The intent recognition model includes a neural network layer 302, a pooling layer 304, and a vector processing layer 306.
The neural network layer 302 may be composed of a network structure such as CNN (Convolutional Neural Networks, convolutional neural network) or RNN (Recurrent Neural Networks, cyclic neural network) or RCNN or CRNN. For converting the first vector into a second vector having text characteristics.
The pooling layer 304 is configured to perform a maximum pooling process and an average pooling process on the second vector output through the neural network layer.
And the vector processing layer 306 is configured to perform addition processing and linear processing on the third vector and the fourth vector, and then perform activation function processing to obtain a prediction probability vector.
The intent recognition model can further comprise a loss function layer, wherein the loss function layer is used for minimizing and fitting the loss function in the loss function layer through an AdamW optimizer in model training, so that the loss function layer converges on a training set and is sufficiently small.
For example, a first vector may be input to the neural network layer to obtain a second vector; inputting the second vector into a pooling layer for maximum pooling treatment and average pooling treatment to obtain a third vector and a fourth vector; and inputting the third vector and the fourth vector into a vector processing layer to perform addition processing linear processing and activation function processing to obtain a predictive probability vector.
In the embodiment, the neural network layer, the pooling layer and the vector processing layer in the intention recognition model are used for obtaining the prediction probability vector, and the method can be used for improving the category prediction accuracy of the dialogue data by respectively converting the dialogue text data into the first word vector and the first word vector; after conversation attribution data is converted into a first conversation attribution vector, the first vector is obtained, the influence of different dialect habits possibly existing in clients in different areas on the prediction precision of conversation data categories is effectively improved, the semantic understanding of a model on wrongly written words which are wrongly translated due to the influence of the dialect is facilitated, and the accuracy of predicting the category of the conversation data by using the model is improved.
In one embodiment, step 214 includes:
step 2142, determining that the corresponding tag information is a category tag of the dialogue data when the probability value corresponding to the tag information exists in the predictive probability vector is greater than a first threshold.
Wherein the first threshold may be an empirical value. The category label of the dialogue data may include a plurality of.
In the above embodiment, when the probability value corresponding to the tag information in the predictive probability vector is greater than the first threshold, the corresponding tag information is determined to be the category tag of the dialogue data. After conversation attribution data is converted into a first conversation attribution vector, the first vector is obtained, the influence of different dialect habits possibly existing in clients in different areas on the prediction precision of conversation data categories is effectively improved, the semantic understanding of a model on wrongly written words which are wrongly translated due to the influence of the dialect is facilitated, and the accuracy of predicting the category of the conversation data by using the model is improved.
In one embodiment, referring to FIG. 4, a training flow diagram of an intent recognition model is shown, comprising:
step 402, obtaining a dialogue sample data set; the dialogue sample data set comprises a plurality of dialogue samples and category labels corresponding to the dialogue samples; the dialog samples include a dialog text sample, a dialog home sample, and a dialog object sample.
Wherein the session sample dataset may be a historical session dataset. The dialog sample data set may include a plurality of dialog samples and category labels corresponding to the dialog samples. The dialog samples include a dialog text sample, a dialog home sample, and a dialog object sample.
The dialogue text sample can be obtained through historical incoming call data of the client, specifically, the historical dialogue audio information can be obtained through obtaining the historical dialogue audio information, the historical dialogue audio information is converted into text information, and denoising processing is carried out on the text information, so that the dialogue text sample is obtained.
The dialogue attribution sample may be a historical customer's incoming call attribution. The dialog object sample may include a user and an agent.
Specifically, in banking, the dialogue sample is a dialogue record of a customer and an agent, and the category label includes "other, personal mobile banking, account and debit card, bank messenger, accounting query, account opening and website information query, funds, personal credit, transfer of money, deposit, personal online banking, financial management, self-service equipment, and precious metal. In the conversation process, a label record can be made for the conversation by the seat so as to record the category label of the conversation.
And step 404, preprocessing the dialogue text sample, the dialogue attribution sample and the dialogue object sample to obtain a dialogue vector sample.
The preprocessing mode can carry out vectorization processing on the dialogue text sample, the dialogue attribution sample and the dialogue object sample so as to obtain a dialogue vector sample. Specifically, the dialog text sample, the dialog home sample and the dialog object sample may be vectorized, and these vectors may be added to obtain a dialog vector sample.
And step 406, adjusting parameters of the intention recognition model according to the dialogue vector sample and the category label corresponding to the dialogue sample until the intention recognition model converges to obtain a trained intention recognition model.
Illustratively, referring to FIG. 5, a block diagram of a classification model is shown. The classification model may be collectively composed of a word embedding model 502, a word embedding model 504, and an intent recognition model 506, wherein the intent recognition model 506 may include a neural network layer 5062, a pooling layer 5064, a vector processing layer 5066, and a loss function layer 5068.
In the loss function layer 4068, the output prediction vector and the label information are minimally fitted using weighted loss functions for two classifications, so that the intent recognition model can be converged and the value of the loss function is sufficiently small when training using the dialogue sample data set.
Specifically, the weighted loss function for two classifications (sigmoid focalloss) can be expressed as:
wherein L is a loss function, y is input label information, y' is predicted label information, delta is a Sigmoid activation function, alpha is a super parameter for balancing positive and negative samples, the value range is [0,1], and the general value is 0.25; χ is a superparameter that balances the easily separable and difficult separable samples, and defaults to 2.0.
For example, when the category label processing is performed manually, each piece of sample data may be marked as multiple times, and multiple labels may exist in one sample, and the same label may be marked multiple times, so it may be considered that the more labels are marked, the greater the weight corresponding to the labels should be.
Wherein Label represents a Label, label sum Indicating the total number of times the sample was labeled with the label.
Illustratively, the final representation of the loss function may be:
Loss＝SigmoidFocalLoss(predict，label/label.max(1，keepdim))；
wherein, the prediction represents the predicted tag information, and the label/label. Max (1, keeper) represents the predicted tag information.
Illustratively, in the loss function layer, a weighted loss function for two classifications is minimally fitted using an AdamW optimizer, so that the intent recognition model can be made to converge when it is trained.
In the embodiment, the historical dialogue sample data is adopted to train the pre-training intention recognition model, so that the pre-training intention recognition model is converged, the trained intention recognition model is suitable for prediction of the dialogue class label in the real environment, and accuracy of class label prediction is improved.
In one embodiment, step 404 includes:
step 4042, acquiring a word vector sample and a word vector sample for the dialogue text sample according to the dialogue text sample.
And 4044, performing vector splicing processing on the word vector samples and the word vector samples to obtain spliced vector samples.
Step 4046, obtaining a dialogue attribution vector sample and a dialogue object vector sample according to the dialogue attribution sample and the dialogue object sample.
Step 4048, obtaining a dialogue vector sample according to the concatenation vector sample, the dialogue attribution vector sample and the dialogue object vector sample.
Wherein the word vector samples may be obtained based on a word embedding model. The word vector samples may be obtained based on a word embedding model. The splice vector samples are splice vectors for dialog text samples. The specific operation of the vector splicing process may be to splice the matrix corresponding to the word vector and the matrix corresponding to the word vector together by columns. The dialogue attribution vector sample is obtained by vectorizing the dialogue attribution sample. The dialog object vector samples are vectorized from dialog object samples.
Illustratively, the terminal may obtain a dialog vector sample from the concatenation vector sample, the dialog home vector sample, and the dialog object vector sample.
In the above embodiment, the word vector sample and the word vector sample for the dialogue text sample are obtained from the dialogue text sample; carrying out vector splicing processing on the word vector samples and the word vector samples to obtain spliced vector samples; according to the dialogue attribution sample and the dialogue object sample, acquiring a dialogue attribution vector sample and a dialogue object vector sample; and obtaining a dialogue vector sample according to the spliced vector sample, the dialogue attribution vector sample and the dialogue object vector sample. Parameters of the intention recognition model are adjusted through the dialogue vector sample and the category label corresponding to the dialogue sample until the intention recognition model converges to obtain a trained intention recognition model, historical dialogue sample data is adopted to train the intention recognition model before training, so that the intention recognition model before training converges, the intention recognition model after training is suitable for prediction of the category label of the dialogue in a real environment, and accuracy of category label prediction is improved.
For a better understanding of the complete process of classification of interactive information in an embodiment of the present invention, a complete example is described, referring to fig. 6, which shows a schematic flow chart of a method of classification of interactive information in another embodiment, comprising the steps of:
Step 602, a dialog sample dataset is obtained.
Step 604, obtaining a word vector sample and a word vector sample for the dialogue text sample according to the dialogue text sample; carrying out vector splicing processing on the word vector samples and the word vector samples to obtain spliced vector samples; according to the dialogue attribution sample and the dialogue object sample, acquiring a dialogue attribution vector sample and a dialogue object vector sample; and obtaining a dialogue vector sample according to the spliced vector sample, the dialogue attribution vector sample and the dialogue object vector sample.
Step 606, adjusting parameters of the intention recognition model according to the dialogue vector sample and the category label corresponding to the dialogue sample until the intention recognition model converges, and obtaining the trained intention recognition model.
In step 608, dialogue data is obtained.
Step 610, inputting the dialogue text data into a pre-trained word embedding model to obtain a first word vector for the dialogue text data; and inputting the dialogue text data into a word embedding model trained in advance to obtain a first word vector aiming at the dialogue text data.
Step 612, performing vector splicing processing on the first word vector and the first word vector to obtain a first spliced vector; acquiring a first dialogue attribution vector and a first dialogue object vector according to dialogue attribution data and dialogue object data; and acquiring a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector.
Step 614, inputting the first vector to the neural network layer to obtain a second vector; inputting the second vector to a pooling layer to respectively carry out maximum pooling treatment and average pooling treatment to obtain a third vector and a fourth vector; and inputting the third vector and the fourth vector into a vector processing layer to perform addition processing linear processing and activation function processing to obtain a predictive probability vector.
In step 616, when the probability value corresponding to the label information in the predictive probability vector is greater than the first threshold, the corresponding label information is determined to be a category label of the dialogue data.
In this embodiment, dialogue data including dialogue text data, dialogue attribution data, and dialogue object data is acquired; acquiring a first word vector and a first word vector for dialogue text data; vector splicing processing is carried out on the first word vector and the first word vector to obtain a first spliced vector; acquiring a first dialogue attribution vector and a first dialogue object vector according to dialogue attribution data and dialogue object data; acquiring a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector; inputting the first vector into the trained intention recognition model to obtain a predictive probability vector; and determining the category label of the dialogue data according to the predictive probability vector. By adopting the method, the category prediction accuracy of the dialogue data can be improved by respectively converting the dialogue text data into the first word vector and the first word vector; after conversation attribution data is converted into a first conversation attribution vector, the first vector is obtained, the influence of different dialect habits possibly existing in clients in different areas on the prediction precision of conversation data categories is effectively improved, the semantic understanding of a model on wrongly written words which are wrongly translated due to the influence of the dialect is facilitated, and the accuracy of predicting the category of the conversation data by using the model is improved.
It should be understood that, although the steps in the flowcharts related to the embodiments described above are sequentially shown as indicated by arrows, these steps are not necessarily sequentially performed in the order indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in the flowcharts described in the above embodiments may include a plurality of steps or a plurality of stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of the steps or stages is not necessarily performed sequentially, but may be performed alternately or alternately with at least some of the other steps or stages.
Based on the same inventive concept, the embodiment of the application also provides an interactive information classification device for realizing the interactive information classification method. The implementation of the solution provided by the device is similar to the implementation described in the above method, so the specific limitation in the embodiments of the device for classifying interactive information provided in the following may be referred to the limitation of the method for classifying interactive information hereinabove, and will not be described herein.
In one embodiment, as shown in fig. 7, there is provided an interactive information classification apparatus, including: a dialogue data acquisition module 702, a first vector acquisition module 704, a first stitching vector acquisition module 706, a second vector acquisition module 708, a first vector acquisition module 710, a predictive probability vector acquisition module 712, and a category label determination module 714, wherein:
a dialogue data acquisition module 702, configured to acquire dialogue data; the dialogue data comprises dialogue text data, dialogue attribution data and dialogue object data;
a first vector obtaining module 704, configured to obtain a first word vector and a first word vector for dialogue text data according to the dialogue text data;
a first spliced vector obtaining module 706, configured to perform vector splicing processing on the first word vector and the first word vector to obtain a first spliced vector;
a second vector obtaining module 708, configured to obtain a first dialogue home vector and a first dialogue object vector according to the dialogue home data and the dialogue object data;
a first vector obtaining module 710, configured to obtain a first vector according to the first stitching vector, the first session home vector, and the first session object vector;
A predictive probability vector obtaining module 712, configured to input the first vector into a trained intent recognition model to obtain a predictive probability vector; the predictive probability vector comprises probability values corresponding to a plurality of tag information;
a category label determining module 714, configured to determine a category label of the dialogue data according to the prediction probability vector.
In some embodiments, the first vector acquisition module 704 includes:
a first word vector obtaining unit, configured to input the dialog text data into a word embedding model trained in advance, to obtain a first word vector for the dialog text data;
and the first word vector obtaining unit is used for inputting the dialogue text data into a pre-trained word embedding model to obtain a first word vector aiming at the dialogue text data.
In some embodiments, the intent recognition model includes a neural network layer, a pooling layer, and a vector processing layer; the neural network layer includes a convolutional neural network CNN or a cyclic neural network RNN, and the prediction probability vector obtaining module 712 includes:
a second vector obtaining unit for inputting the first vector to the neural network layer to obtain a second vector;
A third vector and fourth vector obtaining unit, configured to input the second vector to the pooling layer to perform maximum pooling processing and average pooling processing, respectively, to obtain a third vector and a fourth vector;
and the predictive probability vector obtaining unit is used for inputting the third vector and the fourth vector into the vector processing layer to perform addition processing linear processing and activation function processing so as to obtain the predictive probability vector.
In some embodiments, category label determination module 714 includes:
and the category label determining unit is used for determining the corresponding label information as the category label of the dialogue data under the condition that the probability value corresponding to the label information exists in the predictive probability vector and is larger than a first threshold value.
In some embodiments, the apparatus further comprises:
the sample data set acquisition module is used for acquiring a dialogue sample data set; the dialogue sample data set comprises a plurality of dialogue samples and category labels corresponding to the dialogue samples; the dialogue samples comprise dialogue text samples, dialogue attribution samples and dialogue object samples;
a dialogue vector sample obtaining module, configured to preprocess the dialogue text sample, the dialogue attribution sample, and the dialogue object sample to obtain a dialogue vector sample;
And the intention recognition model obtaining module is used for adjusting parameters of the intention recognition model according to the dialogue vector sample and the category label corresponding to the dialogue sample until the intention recognition model converges to obtain a trained intention recognition model.
In some embodiments, the dialog vector sample derivation module includes:
the word vector and word vector sample obtaining unit is used for obtaining a word vector sample and a word vector sample aiming at the dialogue text sample according to the dialogue text sample;
the vector sample splicing unit is used for carrying out vector splicing processing on the character vector samples and the word vector samples to obtain spliced vector samples;
a home location and object vector sample obtaining unit, configured to obtain a dialogue home location vector sample and a dialogue object vector sample according to the dialogue home location sample and the dialogue object sample;
and the dialogue vector sample obtaining unit is used for obtaining dialogue vector samples according to the spliced vector samples, the dialogue attribution vector samples and the dialogue object vector samples.
The above-mentioned respective modules in the mutual information classification device may be implemented in whole or in part by software, hardware, and a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
In one embodiment, a computer device is provided, which may be a terminal, and the internal structure thereof may be as shown in fig. 8. The computer device includes a processor, a memory, an input/output interface, a communication interface, a display unit, and an input means. The processor, the memory and the input/output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input/output interface. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input/output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used for carrying out wired or wireless communication with an external terminal, and the wireless mode can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. The computer program is executed by a processor to implement a method of interactive information classification. The display unit of the computer device is used for forming a visual picture, and can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, can also be a key, a track ball or a touch pad arranged on the shell of the computer equipment, and can also be an external keyboard, a touch pad or a mouse and the like.
It will be appreciated by those skilled in the art that the structure shown in FIG. 8 is merely a block diagram of some of the structures associated with the present inventive arrangements and is not limiting of the computer device to which the present inventive arrangements may be applied, and that a particular computer device may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.
In one embodiment, a computer device is provided comprising a memory and a processor, the memory having stored therein a computer program, the processor when executing the computer program performing the steps of:
acquiring dialogue data; the dialogue data comprises dialogue text data, dialogue attribution data and dialogue object data;
acquiring a first word vector and a first word vector for the dialogue text data;
vector splicing processing is carried out on the first word vector and the first word vector to obtain a first spliced vector;
acquiring a first dialogue attribution vector and a first dialogue object vector according to the dialogue attribution data and the dialogue object data;
acquiring a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector;
Inputting the first vector into a trained intention recognition model to obtain a predictive probability vector; the predictive probability vector comprises probability values corresponding to a plurality of tag information;
and determining the category label of the dialogue data according to the predictive probability vector.
In one embodiment, a computer readable storage medium is provided having a computer program stored thereon, which when executed by a processor, performs the steps of:
acquiring dialogue data; the dialogue data comprises dialogue text data, dialogue attribution data and dialogue object data;
acquiring a first word vector and a first word vector for the dialogue text data;
vector splicing processing is carried out on the first word vector and the first word vector to obtain a first spliced vector;
acquiring a first dialogue attribution vector and a first dialogue object vector according to the dialogue attribution data and the dialogue object data;
acquiring a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector;
inputting the first vector into a trained intention recognition model to obtain a predictive probability vector; the predictive probability vector comprises probability values corresponding to a plurality of tag information;
And determining the category label of the dialogue data according to the predictive probability vector.
In one embodiment, a computer program product is provided comprising a computer program which, when executed by a processor, performs the steps of:
acquiring dialogue data; the dialogue data comprises dialogue text data, dialogue attribution data and dialogue object data;
acquiring a first word vector and a first word vector for the dialogue text data;
vector splicing processing is carried out on the first word vector and the first word vector to obtain a first spliced vector;
acquiring a first dialogue attribution vector and a first dialogue object vector according to the dialogue attribution data and the dialogue object data;
acquiring a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector;
inputting the first vector into a trained intention recognition model to obtain a predictive probability vector; the predictive probability vector comprises probability values corresponding to a plurality of tag information;
and determining the category label of the dialogue data according to the predictive probability vector.
It should be noted that, the user information (including but not limited to user equipment information, user personal information, etc.) and the data (including but not limited to data for analysis, stored data, presented data, etc.) related to the present application are information and data authorized by the user or sufficiently authorized by each party, and the collection, use and processing of the related data need to comply with the related laws and regulations and standards of the related country and region.
Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, database, or other medium used in embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high density embedded nonvolatile Memory, resistive random access Memory (ReRAM), magnetic random access Memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric Memory (Ferroelectric Random Access Memory, FRAM), phase change Memory (Phase Change Memory, PCM), graphene Memory, and the like. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, and the like. By way of illustration, and not limitation, RAM can be in the form of a variety of forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), and the like. The databases referred to in the embodiments provided herein may include at least one of a relational database and a non-relational database. The non-relational database may include, but is not limited to, a blockchain-based distributed database, and the like. The processor referred to in the embodiments provided in the present application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, or the like, but is not limited thereto.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The foregoing examples illustrate only a few embodiments of the application and are described in detail herein without thereby limiting the scope of the application. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the application, which are all within the scope of the application. Accordingly, the scope of the application should be assessed as that of the appended claims.
Claims (10)
1. An interactive information classification method, characterized in that the method comprises:
acquiring dialogue data; the dialogue data comprises dialogue text data, dialogue attribution data and dialogue object data;
acquiring a first word vector and a first word vector for the dialogue text data;
vector splicing processing is carried out on the first word vector and the first word vector to obtain a first spliced vector;
Acquiring a first dialogue attribution vector and a first dialogue object vector according to the dialogue attribution data and the dialogue object data;
acquiring a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector;
inputting the first vector into a trained intention recognition model to obtain a predictive probability vector; the predictive probability vector comprises probability values corresponding to a plurality of tag information;
and determining the category label of the dialogue data according to the predictive probability vector.
2. The method of claim 1, wherein the obtaining the first word vector and the first word vector for the dialog text data comprises:
inputting the dialogue text data into a pre-trained word embedding model to obtain a first word vector aiming at the dialogue text data;
and inputting the dialogue text data into a word embedding model trained in advance to obtain a first word vector aiming at the dialogue text data.
3. The method of claim 1, wherein the intent recognition model comprises a neural network layer, a pooling layer, and a vector processing layer; the neural network layer includes a convolutional neural network CNN or a cyclic neural network RNN, and the inputting the first vector into the trained intent recognition model to obtain a predictive probability vector includes:
Inputting the first vector to the neural network layer to obtain a second vector;
inputting the second vector to the pooling layer to respectively carry out maximum pooling treatment and average pooling treatment to obtain a third vector and a fourth vector;
and inputting the third vector and the fourth vector into the vector processing layer to perform addition processing linear processing and activation function processing to obtain the predictive probability vector.
4. The method of claim 1, wherein said determining a category label for the dialogue data based on the predictive probability vector comprises:
and under the condition that the probability value corresponding to the label information exists in the predictive probability vector and is larger than a first threshold value, determining the corresponding label information as the class label of the dialogue data.
5. The method according to claim 1, wherein the method further comprises:
acquiring a dialogue sample data set; the dialogue sample data set comprises a plurality of dialogue samples and category labels corresponding to the dialogue samples; the dialogue samples comprise dialogue text samples, dialogue attribution samples and dialogue object samples;
preprocessing the dialogue text sample, the dialogue attribution sample and the dialogue object sample to obtain a dialogue vector sample;
And adjusting parameters of the intention recognition model according to the dialogue vector sample and the category label corresponding to the dialogue sample until the intention recognition model converges to obtain a trained intention recognition model.
6. The method of claim 5, wherein preprocessing the dialog text sample, the dialog home sample, and the dialog object sample to obtain a dialog vector sample comprises:
according to the dialogue text sample, acquiring a word vector sample and a word vector sample aiming at the dialogue text sample;
vector splicing processing is carried out on the word vector samples and the word vector samples to obtain spliced vector samples;
according to the dialogue attribution sample and the dialogue object sample, a dialogue attribution vector sample and a dialogue object vector sample are obtained;
and obtaining a dialogue vector sample according to the spliced vector sample, the dialogue attribution vector sample and the dialogue object vector sample.
7. An interactive information classification apparatus, the apparatus comprising:
the dialogue data acquisition module is used for acquiring dialogue data; the dialogue data comprises dialogue text data, dialogue attribution data and dialogue object data;
The first vector acquisition module is used for acquiring a first word vector and a first word vector aiming at the dialogue text data according to the dialogue text data;
the first splicing vector obtaining module is used for carrying out vector splicing processing on the first word vector and the first word vector to obtain a first splicing vector;
the second vector acquisition module is used for acquiring a first dialogue attribution vector and a first dialogue object vector according to the dialogue attribution data and the dialogue object data;
the first vector obtaining module is used for obtaining a first vector according to the first splicing vector, the first dialogue attribution vector and the first dialogue object vector;
the prediction probability vector obtaining module is used for inputting the first vector into the trained intention recognition model to obtain a prediction probability vector; the predictive probability vector comprises probability values corresponding to a plurality of tag information;
and the category label determining module is used for determining the category label of the dialogue data according to the prediction probability vector.
8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the mutual information classification method according to any of claims 1-6 when the computer program is executed.
9. A computer readable storage medium having stored thereon a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the mutual information classification method according to any of claims 1 to 6.
10. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the mutual information classification method as claimed in any of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310826323.5A CN116860972A (en) | 2023-07-06 | 2023-07-06 | Interactive information classification method, device, apparatus, storage medium and program product |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310826323.5A CN116860972A (en) | 2023-07-06 | 2023-07-06 | Interactive information classification method, device, apparatus, storage medium and program product |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116860972A true CN116860972A (en) | 2023-10-10 |
Family
ID=88229785
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310826323.5A Pending CN116860972A (en) | 2023-07-06 | 2023-07-06 | Interactive information classification method, device, apparatus, storage medium and program product |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116860972A (en) |
-
2023
- 2023-07-06 CN CN202310826323.5A patent/CN116860972A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112732911B (en) | Semantic recognition-based speaking recommendation method, device, equipment and storage medium | |
| CN110046648B (en) | Method and device for classifying business based on at least one business classification model | |
| CN114841161A (en) | Event element extraction method, device, equipment, storage medium and program product | |
| CN112995414A (en) | Behavior quality inspection method, device, equipment and storage medium based on voice call | |
| CN111310462A (en) | User attribute determination method, device, equipment and storage medium | |
| CN114117048A (en) | Text classification method and device, computer equipment and storage medium | |
| CN113887214A (en) | Artificial intelligence based wish presumption method and related equipment thereof | |
| CN115129863A (en) | Intention recognition method, device, equipment, storage medium and computer program product | |
| CN113869068A (en) | Scene service recommendation method, device, equipment and storage medium | |
| CN113886539A (en) | Method and device for recommending dialect, customer service equipment and storage medium | |
| CN114676237A (en) | Sentence similarity determining method and device, computer equipment and storage medium | |
| CN113807920A (en) | Artificial intelligence based product recommendation method, device, equipment and storage medium | |
| CN116860972A (en) | Interactive information classification method, device, apparatus, storage medium and program product | |
| CN114911926A (en) | Text classification method combining convolutional neural network with attention mechanism and support vector machine classifier | |
| Sindhu et al. | Aspect based opinion mining leveraging weighted bigru and CNN module in parallel | |
| CN114490996B (en) | Intention recognition method and device, computer equipment and storage medium | |
| CN117931858B (en) | Data query method, device, computer equipment and storage medium | |
| CN115082245A (en) | Operation data processing method and device, electronic equipment and storage medium | |
| CN118331973A (en) | Data processing method, device, computer equipment and storage medium | |
| CN116975315A (en) | Text matching method, device, computer equipment and storage medium | |
| CN116798414A (en) | Method, device, computer equipment and storage medium for classifying call voice | |
| CN117312508A (en) | Image-based question answering method, apparatus, device, storage medium, and program product | |
| CN116401321A (en) | Data asset classification grading method, system, equipment and medium | |
| CN117493493A (en) | Keyword definition method, keyword definition device, computer equipment and storage medium | |
| CN114239603A (en) | Business requirement matching method and device, computer equipment and storage 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 |