CN112699679B - Emotion recognition method and device, electronic equipment and storage medium - Google Patents
Emotion recognition method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN112699679B CN112699679B CN202110317324.8A CN202110317324A CN112699679B CN 112699679 B CN112699679 B CN 112699679B CN 202110317324 A CN202110317324 A CN 202110317324A CN 112699679 B CN112699679 B CN 112699679B
- Authority
- CN
- China
- Prior art keywords
- sentence
- word
- emotion
- processed
- sentences
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000008909 emotion recognition Effects 0.000 title claims abstract description 77
- 238000000034 method Methods 0.000 title claims abstract description 51
- 230000008451 emotion Effects 0.000 claims abstract description 185
- 239000013598 vector Substances 0.000 claims abstract description 166
- 238000013145 classification model Methods 0.000 claims abstract description 69
- 238000000605 extraction Methods 0.000 claims abstract description 64
- 238000012549 training Methods 0.000 claims abstract description 45
- 238000002372 labelling Methods 0.000 claims abstract description 40
- 230000011218 segmentation Effects 0.000 claims abstract description 40
- 238000011176 pooling Methods 0.000 claims description 108
- 239000011159 matrix material Substances 0.000 claims description 32
- 238000012545 processing Methods 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 12
- 230000007935 neutral effect Effects 0.000 claims description 12
- 238000010586 diagram Methods 0.000 description 8
- 238000006243 chemical reaction Methods 0.000 description 6
- 238000013508 migration Methods 0.000 description 5
- 230000005012 migration Effects 0.000 description 5
- 238000013526 transfer learning Methods 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000003058 natural language processing Methods 0.000 description 2
- PCTMTFRHKVHKIS-BMFZQQSSSA-N (1s,3r,4e,6e,8e,10e,12e,14e,16e,18s,19r,20r,21s,25r,27r,30r,31r,33s,35r,37s,38r)-3-[(2r,3s,4s,5s,6r)-4-amino-3,5-dihydroxy-6-methyloxan-2-yl]oxy-19,25,27,30,31,33,35,37-octahydroxy-18,20,21-trimethyl-23-oxo-22,39-dioxabicyclo[33.3.1]nonatriaconta-4,6,8,10 Chemical compound C1C=C2C[C@@H](OS(O)(=O)=O)CC[C@]2(C)[C@@H]2[C@@H]1[C@@H]1CC[C@H]([C@H](C)CCCC(C)C)[C@@]1(C)CC2.O[C@H]1[C@@H](N)[C@H](O)[C@@H](C)O[C@H]1O[C@H]1/C=C/C=C/C=C/C=C/C=C/C=C/C=C/[C@H](C)[C@@H](O)[C@@H](C)[C@H](C)OC(=O)C[C@H](O)C[C@H](O)CC[C@@H](O)[C@H](O)C[C@H](O)C[C@](O)(C[C@H](O)[C@H]2C(O)=O)O[C@H]2C1 PCTMTFRHKVHKIS-BMFZQQSSSA-N 0.000 description 1
- WPWLFFMSSOAORQ-UHFFFAOYSA-N 5-bromo-4-chloro-3-indolyl acetate Chemical compound C1=C(Br)C(Cl)=C2C(OC(=O)C)=CNC2=C1 WPWLFFMSSOAORQ-UHFFFAOYSA-N 0.000 description 1
- 238000012935 Averaging Methods 0.000 description 1
- 206010063659 Aversion Diseases 0.000 description 1
- 208000020401 Depressive disease Diseases 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000013507 mapping 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
- 238000005457 optimization Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/284—Lexical analysis, e.g. tokenisation or collocates
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Machine Translation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides an emotion recognition method, an emotion recognition device, electronic equipment and a storage medium, wherein the method comprises the following steps: obtaining a statement to be processed; performing word segmentation on the to-be-processed sentence, and acquiring a word vector of each word of the to-be-processed sentence based on a preset word vector model; inputting the word vector of each word of the sentence to be processed into a pre-trained emotion classification model to obtain the probability distribution of the sentence to be processed on each emotion category; the emotion classification model is obtained by training based on the sample sentences and the labeling information of the sample sentences; the emotion classification model is used for extracting semantic information of sentence levels, multi-element levels and word levels of word vectors of the to-be-processed sentences, and obtaining probability distribution of the to-be-processed sentences on each emotion category based on semantic information extraction results. The emotion recognition method provided by the invention improves the accuracy of emotion recognition results and improves user experience.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to an emotion recognition method and device, electronic equipment and a storage medium.
Background
With the continuous popularization of artificial intelligence, people have higher and higher requirements on communication services of intelligent customer service robots, and the fact that the emotion of a customer can be accurately identified is particularly important.
The problem of emotion recognition essentially belongs to an emotion classification problem, namely, the judgment of emotion classification is carried out according to the input content. Most of the current general solutions also focus on the training of emotion classification models, including the generation of training data and the design of classification models. However, the existing intelligent customer service robot is not accurate enough in the aspect of recognizing the emotion of a client, so that the user experience is poor.
In the prior art, the training data of the emotion classification model is single, and the design of the emotion classification model is often considered from a single level, such as the representation of an entire sentence, or the representation of a multi-tuple, or the representation of a word after the sentence is segmented. Although the emotion classification model designed from a single information level can well fit training data based on strong representation capability of a deep neural network in the training process, the accuracy is lost in comparison with the training process in generalization.
Disclosure of Invention
The invention provides an emotion recognition method, an emotion recognition device, electronic equipment and a storage medium, which are used for solving the technical problem that in the prior art, emotion recognition training data is single and model design is only considered from a single information level, so that the accuracy rate of model application is low, and the aim of improving the emotion recognition accuracy rate is fulfilled.
In a first aspect, the present invention provides a method for emotion recognition, comprising:
obtaining a statement to be processed;
performing word segmentation on the statement to be processed, and acquiring a word vector of each word of the statement to be processed based on a preset word vector model;
inputting the word vector of each word of the sentence to be processed into a pre-trained emotion classification model to obtain probability distribution of the sentence to be processed on each emotion category;
the emotion classification model is obtained by training based on the sample sentences and the labeling information of the sample sentences; the emotion classification model is used for extracting semantic information of sentence level, multi-element level and word level of word vectors of the to-be-processed sentences, and obtaining probability distribution of the to-be-processed sentences on each emotion category based on semantic information extraction results.
According to the emotion recognition method provided by the invention, the emotion classification model comprises a sentence level semantic information extraction layer, a multi-component level semantic information extraction layer, a word level semantic information extraction layer and an emotion recognition layer; wherein,
the sentence level semantic information extraction layer is used for respectively carrying out average pooling and maximum pooling on the word vectors of the sentences to be processed, and splicing the pooling results of the average pooling and the maximum pooling to obtain sentence level semantic information;
the multi-group level semantic information extraction layer is used for respectively combining word vectors of the sentences to be processed according to a plurality of preset combined word lengths to obtain a plurality of combined word vector sets, respectively performing convolution operation on each combined word vector in the plurality of combined word vector sets, respectively performing average pooling and maximum pooling on convolution results, and splicing the average pooling and maximum pooling results to obtain multi-group level semantic information; the combination word vectors in the different combination word vector sets have different combination word lengths;
the word level semantic information extraction layer is used for performing attention distribution calculation on the word vectors and the emotion category matrixes of the sentences to be processed to obtain an attention score matrix, performing average pooling and maximum pooling on calculation results between the word vectors and the attention score matrix of the sentences to be processed respectively, and splicing the average pooling and maximum pooling results to obtain word level semantic information; and the emotion recognition layer obtains probability distribution of the sentence to be processed on each emotion category based on the sentence level semantic information, the multi-group level semantic information and the word level semantic information.
According to the emotion recognition method provided by the invention, after the probability distribution of the sentence to be processed on each emotion category is obtained, the method further comprises the following steps:
acquiring the maximum probability value of the sentence to be processed from the probability distribution of the sentence to be processed on each emotion category;
when the maximum probability value of the sentence to be processed is larger than a preset threshold value, taking the emotion category corresponding to the maximum probability value as the emotion category of the sentence to be processed;
and when the maximum probability value of the sentence to be processed is smaller than or equal to a preset threshold value, taking a neutral emotion category as the emotion category of the sentence to be processed.
According to the emotion recognition method provided by the invention, before the sentence to be processed is obtained, the method further comprises the following steps:
acquiring a sample sentence and marking information of the sample sentence;
performing word segmentation on the sample sentence to obtain a word vector of each word of the sample sentence;
and training an emotion classification model based on the word vectors of the sample sentences and the labeling information of the sample sentences.
According to the emotion recognition method provided by the invention, the training of the emotion classification model based on the word vectors of the sample sentences and the labeling information of the sample sentences comprises the following steps:
step S1, extracting semantic information of sentence level, multi-element level and word level from word vectors of sample sentences by using the emotion classification model to be trained;
step S2, extracting a result based on the semantic information to obtain an emotion recognition result of the sample sentence;
step S3, judging whether a model training termination condition is met or not according to the emotion recognition result of the sample sentence and the labeling information of the sample sentence, adjusting the emotion classification model to be trained when the model training termination condition is not met, and executing step S1 again by using the adjusted emotion classification model; and when the model training termination condition is met, obtaining a trained emotion classification model.
According to the emotion recognition method provided by the invention, the acquiring of the sample sentence and the labeling information of the sample sentence comprises the following steps:
obtaining a statement in a first field, wherein the first field is the same field as the field of the statement to be processed;
extracting a first type of sentences with emotion keywords from the sentences in the first field, and labeling information for the first type of sentences;
extracting second sentences without emotion keywords from the sentences in the first field according to a preset proportion, and labeling information for the second sentences; wherein the second type statement marking information is neutral emotion;
and taking the first type of statement and the second type of statement as sample statements, and taking the information labeled for the first type of statement and the information labeled for the second type of statement as labeling information of the sample statements.
According to the emotion recognition method provided by the invention, the acquiring of the sample sentence and the labeling information of the sample sentence comprises the following steps:
acquiring statements in a second field and label information of the statements in the second field, wherein the second field is a field different from the field of the statements to be processed;
taking the sentences in the second field as sample sentences, and taking the labeling information of the sentences in the second field as the labeling information of the sample sentences;
correspondingly, the obtaining of the emotion recognition result of the sample sentence based on the semantic information extraction result includes:
migrating the semantic information extraction result;
and extracting a result based on the migrated semantic information to obtain an emotion recognition result of the sample sentence.
In a second aspect, the present invention provides an emotion recognition apparatus comprising:
the acquisition module is used for acquiring the statement to be processed;
the processing module is used for segmenting the statement to be processed and acquiring a word vector of each word of the statement to be processed based on a preset word vector model;
the input module is used for inputting the word vector of each word of the sentence to be processed into a pre-trained emotion classification model to obtain the probability distribution of the sentence to be processed on each emotion category;
the training module is used for training the emotion classification model based on the sample sentences and the labeling information of the sample sentences;
and the extraction module is used for extracting semantic information of sentence level, multi-group level and word level from the word vector of the sentence to be processed, and obtaining probability distribution of the sentence to be processed on each emotion category based on a semantic information extraction result.
In a third aspect, the present invention provides an electronic device comprising: the system comprises a processor, a memory and a bus, wherein the processor and the memory are communicated with each other through the bus; the memory stores program instructions executable by the processor, the processor invoking the program instructions to perform a method as described in any of the above.
In a fourth aspect, the invention provides a non-transitory computer readable storage medium storing computer instructions that cause the computer to perform the method of any one of the above.
According to the emotion recognition method, the emotion recognition device, the electronic equipment and the storage medium, the obtained sentence to be processed is subjected to word segmentation, word vector conversion is carried out on the segmented words based on the preset word vector model, then semantic information extraction of sentence levels, multi-group levels and word levels is carried out on the word vectors of the sentence to be processed through the pre-trained emotion classification model, probability distribution of the sentence to be processed on each emotion category is obtained based on the semantic information extraction result, the emotion classification model recognition accuracy is improved, and user experience is improved.
Drawings
In order to more clearly illustrate the technical solutions of the present invention or the prior art, the drawings needed for 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 schematic flow chart of a method for emotion recognition provided by the present invention;
FIG. 2 is a schematic diagram of a flow of hierarchical semantic extraction in emotion recognition provided by the present invention;
FIG. 3 is a schematic diagram of a process for training an emotion classification model according to the present invention;
FIG. 4 is a diagram illustrating the migration of semantic information extraction results in a second domain in emotion recognition according to the present invention;
fig. 5 is a schematic structural diagram of an emotion recognition apparatus provided in the present invention;
fig. 6 is a schematic structural diagram of an electronic device provided in the present invention.
Detailed Description
In order to more clearly illustrate the technical solutions of the present invention or the prior art, the drawings needed for 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 schematic flow diagram of an emotion recognition method provided by the present invention, and as shown in fig. 1, the emotion recognition method provided by the present invention specifically includes the following steps:
step 101: obtaining a statement to be processed;
step 102: performing word segmentation on the statement to be processed, and acquiring a word vector of each word of the statement to be processed based on a preset word vector model;
step 103: inputting the word vector of each word of the sentence to be processed into a pre-trained emotion classification model to obtain probability distribution of the sentence to be processed on each emotion category;
the emotion classification model is obtained by training based on the sample sentences and the labeling information of the sample sentences; the emotion classification model is used for extracting semantic information of sentence level, multi-element level and word level of word vectors of the to-be-processed sentences, and obtaining probability distribution of the to-be-processed sentences on each emotion category based on semantic information extraction results.
Specifically, the to-be-processed sentence refers to a sentence that needs emotion recognition, and in this embodiment, the to-be-processed sentence is a sentence in the customer service field. In other embodiments, the pending statement may also be a statement in other fields, such as the hotel field, the safe driving field, the depression treatment field, and the like.
The word segmentation means that a word segmentation tool is adopted to perform word segmentation on the sentence to be processed, the word segmentation result is determined by the word segmentation tool, and the selection of the word segmentation tool is not specifically limited herein.
The word Vector model (one-hot Vector) is used to convert each word in the sentence text into a corresponding word Vector, so that each word has a mapping relation with the word Vector. Word vectors are a general term for a set of language modeling and feature learning techniques in Word embedded Natural Language Processing (NLP), and refer to vectors in which words or phrases from a vocabulary are mapped to real numbers.
The probability distribution refers to the magnitude of probability values of the sentences to be processed on each emotion category. The categories of emotions include a variety of basic types of emotions, such as happy, angry, fear, sadness, disgust, surprise, angry, and the like.
In this embodiment, the obtained to-be-processed sentence is subjected to word segmentation first, then each word segmentation is subjected to word segmentation conversion based on a word vector model, a word vector corresponding to each word segmentation is obtained, and the word vector is input into a pre-trained emotion classification model, so as to obtain a probability value of the to-be-processed sentence on each emotion category.
In this embodiment, the emotion classification model needs to be trained in advance, and the word vector model is also preset, where the emotion classification model is obtained by training based on the sample sentence and the label information of the sample sentence, and can extract semantic information of sentence level, tuple level and word level from the word vector of the sentence to be processed, and obtain probability values of the sentence to be processed in each emotion category according to the multi-level speech information extraction results. In this embodiment, the tuple layer is formed by combining three tuples, i.e. binary, ternary and quaternary.
In step S102, a word segmentation process is performed on the sentence to be processed, and a word vector of the segmented word is obtained based on a preset word vector model. The word segmentation result is determined by a word segmentation tool, and the word segmentation results obtained by different word segmentation tools are different. If the sentence to be processed is that "i just eat dinner i happy today", the word segmentation result can be: i/just/eat/dinner/i/today/very/happy, and based on a preset word vector model obtains: the word vector corresponding to 'I' is A, the word vector corresponding to 'just' is B, the word vector corresponding to 'eat up' is C, the word vector corresponding to 'dinner' is 'D', the word vector corresponding to 'I' is A, the word vector corresponding to 'today' is 'E', the word vector corresponding to 'very' is 'F', the word vector corresponding to 'happy' is 'G', and finally the sentence to be processed 'I just eat up dinner and I are happy today' is formed into a word vector expression of [ A B C D A E F G ]. The word segmentation tool can be selected according to actual needs, and is not particularly limited herein.
In the above step S103, the emotion types may be set to different kinds of emotions according to actual requirements, for example, 4 emotion types may be set, and 6 emotions may also be set. In the present embodiment, 5 emotion types are set. Such as happy, sad, angry, disgust, and neutral. Wherein, neutral emotions refer to emotions other than the former four emotions.
In the embodiment of the invention, the obtained sentence to be processed is subjected to word segmentation, the word vector conversion is carried out on the segmentation based on a preset word vector model, semantic information extraction of sentence level, multi-group level and word level is carried out on the word vector based on a pre-trained emotion classification model, and probability distribution of the sentence to be processed on each emotion category is obtained according to the extraction result. The emotion recognition method provided by the invention improves the accuracy of emotion recognition results and improves user experience.
Fig. 2 is a schematic view of a flow of emotion recognition hierarchical semantic extraction according to an embodiment of the present invention, and as shown in fig. 2, in an embodiment of the present invention, the emotion classification model includes a sentence hierarchical semantic information extraction layer, a tuple hierarchical semantic information extraction layer, a word hierarchical semantic information extraction layer, and an emotion recognition layer; wherein,
the sentence level semantic information extraction layer is used for respectively carrying out average pooling and maximum pooling on the word vectors of the sentences to be processed, and splicing the pooling results of the average pooling and the maximum pooling to obtain sentence level semantic information;
the multi-group level semantic information extraction layer is used for respectively combining word vectors of the sentences to be processed according to a plurality of preset combined word lengths to obtain a plurality of combined word vector sets, respectively performing convolution operation on each combined word vector in the plurality of combined word vector sets, respectively performing average pooling and maximum pooling on convolution results, and splicing the average pooling and maximum pooling results to obtain multi-group level semantic information; the combination word vectors in the different combination word vector sets have different combination word lengths;
the word level semantic information extraction layer is used for performing attention distribution calculation on the word vectors and the emotion category matrixes of the sentences to be processed to obtain an attention score matrix, performing average pooling and maximum pooling on calculation results between the word vectors and the attention score matrix of the sentences to be processed respectively, and splicing the average pooling and maximum pooling results to obtain word level semantic information;
and the emotion recognition layer obtains probability distribution of the sentence to be processed on each emotion category based on the sentence level semantic information, the multi-group level semantic information and the word level semantic information.
Specifically, pooling (pooling) is used to remove miscellaneous information, retain critical information, and achieve optimization of data. The average pooling (mean pooling) can retain the characteristics of the overall data and better highlight background information, and the average pooling expression formula is as follows:
wherein,v sthe average pooling result is shown as a result of,v ia word vector value representing each participle.
Maximum pooling (max pooling) allows for better retention of texture features, with the maximum pooling expressed by the formula:v M=Max-pooling(v1,v2,…,vL) Whereinv MThe maximum pooling result is indicated.
In this embodiment, the pooling results of the average pooling and the maximum pooling of the semantic information of the same level are first spliced, and then the semantic information after the pooling processing of different levels is spliced, so as to obtain the semantic information extraction result including multiple levels.
The n-gram multi-element level semantic information extraction layer extracts a corresponding multi-element combination word according to a plurality of preset combination word lengths, wherein the combination word length is determined by an n value, when n =2, the combination word is expressed as a binary group, the combination word length is 2, that is, the multi-element combination word is formed by combining a plurality of word groups composed of two words, and the convolution operation is to set a convolution kernel through a neural network (CNN) to acquire information of each multi-element group.
The emotion category matrix is a matrix obtained by performing word vector conversion processing on emotion categories and processing acquired word vectors. If 5 emotion categories are selected in this embodiment, which are happy, sad, angry, disgust and neutral, the word vector corresponding to each emotion is the following word vectors:
the word vector corresponding to happiness is H, the word vector corresponding to sadness is I, the word vector corresponding to anger is J, the word vector corresponding to disgust is K, the word vector corresponding to neutrality is M, a word vector expression form of [ H I J K M ] is formed, and the emotion category matrix is obtained from the obtained word vector expression form according to needs.
In this embodiment, a long-sequence semantic information representation is generated by splicing the obtained sentence-level semantic information, tuple-level semantic information, and word-level semantic information, and probability distribution of the to-be-processed sentence in each emotion category is obtained by full-connection matrix processing. In this embodiment, it is preferable to extract sentence-level semantic information, tuple-level semantic information, and word-level semantic information, where the tuples are binary tuples, ternary tuples, and quaternary tuples, respectively. This is explained in detail below by means of a specific example.
If the obtained sentence to be processed is that 'I just eat dinner and I are happy', a word vector expression of [ AB C D A E F G ] is generated through the processing of word segmentation and a word vector model.
The sentence level semantic information extraction layer is used for respectively performing average pooling and maximum pooling on the word vectors of the to-be-processed sentences, and splicing the pooling results of the average pooling and the maximum pooling to obtain sentence level semantic information, and the specific implementation result is shown in the following table 1.
TABLE 1
The average pooling is realized by averaging a few vectors A to G in each dimension, and the maximum pooling is realized by taking the maximum value in each dimension, and it can be seen from the above Table 1 that the final average pooling result is the vector VsThe maximum pooling result is the vector VMAnd splicing the average pooling result and the maximum pooling result to obtain the expression of the semantic information of sentence level.
The multi-group level semantic information extraction layer is used for combining word vectors of the to-be-processed sentences according to a plurality of preset combined word lengths to obtain a plurality of combined word vector sets, performing convolution operation on each combined word vector in the plurality of combined word vector sets, performing average pooling and maximum pooling on convolution results respectively, and splicing the average pooling and maximum pooling results to obtain multi-group level semantic information. The n gram tuples in this embodiment are preferably n =2, 3 and 4.
When n =2, the binary information is acquired in a manner of overlapping and dividing with the previous word, so that missing tuple information is avoided. Respectively as follows: AB BC CD DA AE EF FG, which constitutes a vector set of binary combination words, and the specific implementation results are shown in Table 2 below.
TABLE 2
The average pooling is realized by the vector AB after convolutionZTo FGZThe vectors are averaged in each dimension, the maximum pooling is the maximum value in each dimension, and it can be seen from table 2 that the final averaged pooling result is vector VZS2The maximum pooling result is the vector VZM2And splicing the average pooling result and the maximum pooling result to obtain the representation of the semantic information of the binary group level.
When n =3, the triple information is obtained in a manner of overlapping and dividing with the previous word, so that missing tuple information is avoided, and the method comprises the following steps: the ABC BCD CDA DAE AEF EFG forms a vector set of the triple combination words, and the specific implementation results are shown in the following table 3.
TABLE 3
The average pooling is realized by the vector ABC after convolutionZTo the EFGZThe vectors are averaged in each dimension, and the maximum pooling is the maximum value in each dimension, and it can be seen from table 3 that the final averaged pooling result is vector VZS3The maximum pooling result is the vector VZM3And splicing the average pooling result and the maximum pooling result to obtain the representation of the three-tuple-level semantic information.
When n =4, the four-tuple information is obtained in a manner of overlapping and dividing with the previous word, so that missing tuple information is avoided, and the four-tuple information is respectively: ABCD BCDA CDAE DAEF AEFG, which forms a vector set of the four-tuple compound words, and the specific implementation results are shown in the following table 4.
TABLE 4
The average pooling is realized by the convolution vector ABCDZTo AEFGZThe vectors are averaged in each dimension, the maximum pooling is the maximum value in each dimension, and it can be seen from table 4 that the final averaged pooling result is vector VZS4The maximum pooling result is the vector VZM4And splicing the average pooling result and the maximum pooling result to obtain the expression of the semantic information of the quadruple hierarchy.
And the word level semantic information extraction layer is used for performing attention distribution calculation on the word vector and the emotion category matrix of the sentence to be processed to obtain an attention score matrix, performing average pooling and maximum pooling on calculation results between the word vector and the attention score matrix of the sentence to be processed respectively, and splicing the pooled results of the average pooling and the maximum pooling to obtain word level semantic information.
The emotion category matrix is a matrix obtained by performing word vector conversion processing on emotion categories and processing the acquired word vectors; as shown in fig. 2, the Attention score matrix is a matrix representation obtained by multiplying the word matrix and the emotion classification matrix to obtain an emotion-sentence similarity matrix, and performing an Attention operation on the emotion-sentence similarity matrix; and then multiplying the attention score matrix and the word matrix to obtain a calculation result of the multiplication, performing average pooling and maximum pooling on the calculation result, and finally splicing the average pooling and maximum pooling results to obtain semantic information representation of the word level.
And finally, the emotion recognition layer obtains probability distribution of the sentence to be processed on each emotion category based on the sentence level semantic information, the multi-group level semantic information and the word level semantic information. The obtained sentence level semantic information, the multi-element group level semantic information and the word level semantic information are spliced to generate a long sequence matrix representation, and the long sequence matrix is processed through a full connection matrix to obtain the probability distribution condition of the sentence to be processed on each emotion category.
In the embodiment, the accuracy of the recognition result of the emotion classification model and the user experience are improved by extracting the semantic information of the sentence level, the multi-group level and the word level of the sentence to be processed through the pre-trained emotion classification model.
In another embodiment of the present invention, after obtaining the probability distribution of the sentence to be processed on each emotion category, the method further includes:
acquiring the maximum probability value of the sentence to be processed from the probability distribution of the sentence to be processed on each emotion category;
when the maximum probability value of the sentence to be processed is larger than a preset threshold value, taking the emotion category corresponding to the maximum probability value as the emotion category of the sentence to be processed;
and when the maximum probability value of the sentence to be processed is smaller than or equal to a preset threshold value, taking a neutral emotion category as the emotion category of the sentence to be processed.
Specifically, the size of the preset threshold may be set according to a parameter obtained from the training result. Supposing that probability distribution of the sentences to be processed in each emotion category is obtained, the probability value of the happy emotion corresponding to the sentences to be processed is the largest and is equal to 0.6, if the preset threshold value is 0.5, it is obvious that the maximum probability value of the emotion category is greater than the preset threshold value, and the emotion category of the sentences to be processed is output as happy; and if the maximum probability value is 0.4 and is obviously smaller than the preset threshold value, outputting the emotion type of the sentence to be processed as a neutral emotion.
According to the embodiment, by comparing the judged maximum probability value with the preset threshold value, the emotion category corresponding to the sentence to be processed can be well acquired, and the accuracy and efficiency of emotion recognition are improved.
In an embodiment of the present invention, before the obtaining the statement to be processed, the method further includes:
acquiring a sample sentence and marking information of the sample sentence;
performing word segmentation on the sample sentence to obtain a word vector of each word of the sample sentence;
and training an emotion classification model based on the word vectors of the sample sentences and the labeling information of the sample sentences.
Specifically, the sample sentence refers to annotation data used for training the emotion classification model, and the annotation information refers to an emotion type corresponding to the sample sentence. In this embodiment, the setting manner of the emotion type may be as shown in table 5 below, a word segmentation process is performed on a sample sentence based on a word segmentation tool, a preset word vector model obtains a word vector of the sample sentence, and then an emotion classification model is trained based on the word vector and the label information of the sample sentence.
TABLE 5
| Categories of emotions | Happy | Sadness and sorrow | Anger and anger | Aversion to | Neutral property |
| Numbering | 1 | 2 | 3 | 4 | 5 |
If the sample sentence is "i just eat dinner i are happy today", the annotation information corresponding to the obtained sample sentence is 1, that is, the corresponding emotion type of the sample sentence is happy. Through the specific implementation manner in the embodiment, word segmentation processing and word vector acquisition are performed on the sample sentence in the same manner, and the emotion classification model is trained through the word vector and the label information of the sample sentence to acquire each undetermined parameter of the model. The setting mode of the emotion type can be set according to actual needs, and is not particularly limited.
According to the embodiment of the invention, the emotion classification model is trained through a large number of sample sentences and marking information, each undetermined parameter of the emotion recognition model can be obtained, and the accuracy and generalization capability of the emotion recognition model are improved.
In another embodiment of the present invention, as shown in fig. 3, the training of the emotion classification model based on the word vector of the sample sentence and the label information of the sample sentence specifically includes the following steps:
step S1, extracting semantic information of sentence level, multi-element level and word level from word vectors of sample sentences by using the emotion classification model to be trained;
step S2, extracting a result based on the semantic information to obtain an emotion recognition result of the sample sentence;
step S3, judging whether a model training termination condition is met or not according to the emotion recognition result of the sample sentence and the labeling information of the sample sentence, adjusting the emotion classification model to be trained when the model training termination condition is not met, and executing step S1 again by using the adjusted emotion classification model; and when the model training termination condition is met, obtaining a trained emotion classification model.
Specifically, the model training termination condition may be set to terminate training when the emotion recognition result of the sample sentence is consistent with the annotation information of the sample sentence. And if the labeling information of the sample sentence is happy, the recognition result of the sample sentence obtained through model training is also happy, the model training termination condition is met, and otherwise, the emotion classification model is adjusted and trained again. In this embodiment, semantic information extraction of a sentence level, a tuple level, and a word level needs to be performed on a word vector of a sample sentence, where the tuple level may be set to be extraction of information of a tuple level and a triple level. And is not particularly limited herein. The semantic information extraction steps of each layer are described in the above embodiments, and are not described herein again.
In the embodiment of the invention, sentence level, multi-group level and word level semantic information are extracted from the sample sentence, the emotion recognition result of the sample sentence is obtained according to the extraction result, and the emotion recognition result and the labeling information of the sample sentence are judged to judge whether the emotion recognition result meets the termination condition of model training or not to obtain the emotion classification model, so that the accuracy of the recognition result of the emotion classification model can be improved.
In an embodiment of the present invention, the obtaining of the sample statement and the annotation information of the sample statement includes:
obtaining a statement in a first field, wherein the first field is the same field as the field of the statement to be processed;
extracting a first type of sentences with emotion keywords from the sentences in the first field, and labeling information for the first type of sentences;
extracting second sentences without emotion keywords from the sentences in the first field according to a preset proportion, and labeling information for the second sentences; wherein the second type statement marking information is neutral emotion;
and taking the first type of statement and the second type of statement as sample statements, and taking the information labeled for the first type of statement and the information labeled for the second type of statement as labeling information of the sample statements.
Specifically, in this embodiment, the first field is a customer service field, and the sample statement is a statement in the customer service field, and is obtained by sampling and screening the acquired session data between the customer service and the client. In other embodiments, the first field may also be other fields, such as a hotel field, a safe driving field, and the like, and is not limited herein.
The first type of sentences refer to sample sentences with emotion keywords and label the sample sentences, the second type of sentences refer to sample sentences without emotion keywords, the second type of sentences are also labeled, and the emotion types corresponding to the sample sentences without emotion keywords are labeled as neutral emotions. Namely, the emotion type of the sample sentence with the emotion keyword is labeled as the emotion type corresponding to the keyword, and corresponding labeling is performed according to the setting mode of the emotion type. If i'm happy, the emotion keyword is "happy", and the emotion type is set to be happy 1, the emotion category corresponding to "i'm happy" is labeled as 1, that is, the emotion type of the sentence is happy.
In the embodiment, the sample sentences in the customer service field are labeled, and the sample sentences with emotion keywords and the sample sentences without emotion keywords are labeled respectively, so that the efficiency of classifying emotion types by the emotion classification model can be improved, and the accuracy of emotion recognition of the model can be improved.
In an embodiment of the present invention, as shown in fig. 4, the acquiring the sample statement and the annotation information of the sample statement includes:
acquiring statements in a second field and label information of the statements in the second field, wherein the second field is a field different from the field of the statements to be processed;
taking the sentences in the second field as sample sentences, and taking the labeling information of the sentences in the second field as the labeling information of the sample sentences;
correspondingly, the obtaining of the emotion recognition result of the sample sentence based on the semantic information extraction result includes:
migrating the semantic information extraction result;
and extracting a result based on the migrated semantic information to obtain an emotion recognition result of the sample sentence.
Specifically, Transfer Learning (Transfer Learning) is a machine Learning method, which is to Transfer knowledge in one domain (i.e., a source domain) to another domain (i.e., a target domain) so that the target domain can obtain a better Learning effect.
In this embodiment, the sample sentence not only includes text data in the customer service field, but also includes text data in other application fields, where the second field refers to other application fields except the customer service field, such as e-commerce field, hotel field, and the like.
And realizing data migration on the semantic information extraction result obtained in the second field through the migration matrix to obtain the emotion recognition result of the sample sentence in the second field. In the training process of the emotion recognition model, the emotion recognition model can be trained through a comparison result between the emotion recognition result of the sample sentence in the second field and the annotation information of the sample sentence in the second field, wherein the annotation information of the sample sentence in the second field is the published annotation text data.
In the embodiment, the acquired sample sentences in other application fields are subjected to semantic information extraction, and the acquired semantic extraction result is subjected to migration processing through the migration matrix, so that the influence of the labeling sample sentences in other application fields on the labeling sample sentences in the customer service field is weakened, the shortage of the number of the labeling sample sentences in the customer service field is compensated, and the accuracy of model training is improved.
Fig. 5 is a schematic structural diagram of an emotion recognition apparatus according to an embodiment of the present invention, and as shown in fig. 5, the present invention provides an emotion recognition apparatus including:
an obtaining module 501, configured to obtain a statement to be processed;
a processing module 502, configured to perform word segmentation on the to-be-processed sentence, and obtain a word vector of each word of the to-be-processed sentence based on a preset word vector model;
an input module 503, configured to input a word vector of each word of the to-be-processed sentence into a pre-trained emotion classification model, so as to obtain probability distribution of the to-be-processed sentence on each emotion category;
a training module 504, configured to train the emotion classification model based on the sample sentence and the labeling information of the sample sentence;
the extraction module 505 is configured to extract semantic information of the sentence level, the tuple level, and the word level from the word vector of the to-be-processed sentence, and obtain probability distribution of the to-be-processed sentence in each emotion category based on a semantic information extraction result.
Specifically, the sentence to be processed refers to text data acquired in the customer service field; the word segmentation processing is to perform word segmentation on the to-be-processed sentence according to the word segmentation tool, and the word segmentation result is determined by the word segmentation tool and is not specifically limited. The word vector model is preset in advance, and the obtained participles are subjected to word vector conversion, so that each participle is converted into a corresponding word vector.
In the embodiment of the invention, the obtained sentence to be processed is segmented through the processing module and the segmentation is converted into word vectors based on the preset word vector model, and the input module is used for inputting the word vectors of the sentence to be processed into the pre-trained emotion classification model to obtain the probability distribution of the sentence to be processed on each emotion category. The emotion recognition device provided by the invention improves the accuracy of emotion recognition results and improves user experience.
Since the principle of the apparatus according to the embodiment of the present invention is the same as that of the method according to the above embodiment, further details are not described herein for further explanation.
In another embodiment of the present invention, the method further comprises: the transfer learning module is used for transferring the semantic information extraction result of the second field; and extracting a result based on the migrated semantic information to obtain an emotion recognition result of the sample sentence. According to the embodiment, the number of the sample sentences in the customer service field can be expanded through the arrangement of the transfer learning module, and the training accuracy of the emotion classification model is improved.
Fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, and as shown in fig. 6, the present invention provides an electronic device, including: a processor (processor)601, a memory (memory)602, and a bus 603;
the processor 601 and the memory 602 complete communication with each other through the bus 603;
The present embodiments provide a non-transitory computer-readable storage medium storing computer instructions that cause the computer to perform the methods provided by the above method embodiments, for example, including: obtaining a statement to be processed; performing word segmentation on the statement to be processed, and acquiring a word vector of each word of the statement to be processed based on a preset word vector model; inputting the word vector of each word of the sentence to be processed into a pre-trained emotion classification model to obtain probability distribution of the sentence to be processed on each emotion category; the emotion classification model is obtained by training based on the sample sentences and the labeling information of the sample sentences; the emotion classification model is used for extracting semantic information of sentence level, multi-element level and word level of word vectors of the to-be-processed sentences, and obtaining probability distribution of the to-be-processed sentences on each emotion category based on semantic information extraction results.
Those of ordinary skill in the art will understand that: all or part of the steps for implementing the method embodiments may be implemented by hardware related to program instructions, and the program may be stored in a computer readable storage medium, and when executed, the program performs the steps including the method embodiments; and the aforementioned storage medium includes: various media that can store program codes, such as ROM, RAM, magnetic or optical disks.
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 (9)
1. A method of emotion recognition, comprising:
obtaining a statement to be processed;
performing word segmentation on the statement to be processed, and acquiring a word vector of each word of the statement to be processed based on a preset word vector model;
inputting the word vector of each word of the sentence to be processed into a pre-trained emotion classification model to obtain probability distribution of the sentence to be processed on each emotion category; wherein the emotion categories are a plurality of basic emotion types;
the emotion classification model is obtained by training based on the sample sentences and the labeling information of the sample sentences;
the emotion classification model is used for extracting semantic information of sentence levels, multi-element levels and word levels from word vectors of the to-be-processed sentences, and obtaining probability distribution of the to-be-processed sentences on each emotion category based on semantic information extraction results;
the emotion classification model comprises a sentence level semantic information extraction layer, a multi-group level semantic information extraction layer, a word level semantic information extraction layer and an emotion recognition layer; wherein,
the sentence level semantic information extraction layer is used for respectively carrying out average pooling and maximum pooling on the word vectors of the sentences to be processed, and splicing the pooling results of the average pooling and the maximum pooling to obtain sentence level semantic information;
the multi-group level semantic information extraction layer is used for respectively combining word vectors of the sentences to be processed according to a plurality of preset combined word lengths to obtain a plurality of combined word vector sets, respectively performing convolution operation on each combined word vector in the plurality of combined word vector sets, respectively performing average pooling and maximum pooling on convolution results, and splicing the average pooling and maximum pooling results to obtain multi-group level semantic information; the combination word vectors in the different combination word vector sets have different combination word lengths;
the word level semantic information extraction layer is used for performing attention distribution calculation on the word vectors and the emotion category matrixes of the sentences to be processed to obtain an attention score matrix, performing average pooling and maximum pooling on calculation results between the word vectors and the attention score matrix of the sentences to be processed respectively, and splicing the average pooling and maximum pooling results to obtain word level semantic information;
and the emotion recognition layer splices the sentence-level semantic information, the multi-element group-level semantic information and the word-level semantic information to generate a long sequence matrix, so as to obtain the probability distribution of the to-be-processed sentence on each emotion category.
2. The emotion recognition method of claim 1, wherein after obtaining the probability distribution of the sentence to be processed on each emotion category, the method further comprises:
acquiring the maximum probability value of the sentence to be processed from the probability distribution of the sentence to be processed on each emotion category;
when the maximum probability value of the sentence to be processed is larger than a preset threshold value, taking the emotion category corresponding to the maximum probability value as the emotion category of the sentence to be processed;
and when the maximum probability value of the sentence to be processed is smaller than or equal to a preset threshold value, taking a neutral emotion category as the emotion category of the sentence to be processed.
3. The emotion recognition method of claim 1, wherein, prior to the obtaining of the sentence to be processed, the method further comprises:
acquiring a sample sentence and marking information of the sample sentence;
performing word segmentation on the sample sentence to obtain a word vector of each word of the sample sentence;
and training an emotion classification model based on the word vectors of the sample sentences and the labeling information of the sample sentences.
4. The emotion recognition method of claim 3, wherein the training of the emotion classification model based on the word vector of the sample sentence and the labeling information of the sample sentence comprises:
step S1, extracting semantic information of sentence level, multi-element level and word level from word vectors of sample sentences by using the emotion classification model to be trained;
step S2, extracting a result based on the semantic information to obtain an emotion recognition result of the sample sentence;
step S3, judging whether a model training termination condition is met or not according to the emotion recognition result of the sample sentence and the labeling information of the sample sentence, adjusting the emotion classification model to be trained when the model training termination condition is not met, and executing step S1 again by using the adjusted emotion classification model; and when the model training termination condition is met, obtaining a trained emotion classification model.
5. The emotion recognition method of claim 3, wherein the obtaining of the sample sentence and the annotation information of the sample sentence includes:
obtaining a statement in a first field, wherein the first field is the same field as the field of the statement to be processed;
extracting a first type of sentences with emotion keywords from the sentences in the first field, and labeling information for the first type of sentences;
extracting second sentences without emotion keywords from the sentences in the first field according to a preset proportion, and labeling information for the second sentences; wherein the second type statement marking information is neutral emotion;
and taking the first type of statement and the second type of statement as sample statements, and taking the information labeled for the first type of statement and the information labeled for the second type of statement as labeling information of the sample statements.
6. The emotion recognition method of claim 4, wherein the obtaining of the sample sentence and the annotation information of the sample sentence includes:
acquiring statements in a second field and label information of the statements in the second field, wherein the second field is a field different from the field of the statements to be processed;
taking the sentences in the second field as sample sentences, and taking the labeling information of the sentences in the second field as the labeling information of the sample sentences;
correspondingly, the obtaining of the emotion recognition result of the sample sentence based on the semantic information extraction result includes:
migrating the semantic information extraction result;
and extracting a result based on the migrated semantic information to obtain an emotion recognition result of the sample sentence.
7. An emotion recognition apparatus, comprising:
the acquisition module is used for acquiring the statement to be processed;
the processing module is used for segmenting the statement to be processed and acquiring a word vector of each word of the statement to be processed based on a preset word vector model;
the input module is used for inputting the word vector of each word of the sentence to be processed into a pre-trained emotion classification model to obtain the probability distribution of the sentence to be processed on each emotion category; wherein the emotion categories are a plurality of basic emotion types;
the training module is used for training the emotion classification model based on the sample sentences and the labeling information of the sample sentences;
the extraction module is used for extracting semantic information of sentence level, multi-group level and word level from the word vector of the sentence to be processed, and obtaining probability distribution of the sentence to be processed on each emotion category based on a semantic information extraction result;
the extraction module comprises a sentence level semantic information extraction layer, a multi-group level semantic information extraction layer, a word level semantic information extraction layer and an emotion recognition layer; wherein,
the sentence level semantic information extraction layer is used for respectively carrying out average pooling and maximum pooling on the word vectors of the sentences to be processed, and splicing the pooling results of the average pooling and the maximum pooling to obtain sentence level semantic information;
the multi-group level semantic information extraction layer is used for respectively combining word vectors of the sentences to be processed according to a plurality of preset combined word lengths to obtain a plurality of combined word vector sets, respectively performing convolution operation on each combined word vector in the plurality of combined word vector sets, respectively performing average pooling and maximum pooling on convolution results, and splicing the average pooling and maximum pooling results to obtain multi-group level semantic information; the combination word vectors in the different combination word vector sets have different combination word lengths;
the word level semantic information extraction layer is used for performing attention distribution calculation on the word vectors and the emotion category matrixes of the sentences to be processed to obtain an attention score matrix, performing average pooling and maximum pooling on calculation results between the word vectors and the attention score matrix of the sentences to be processed respectively, and splicing the average pooling and maximum pooling results to obtain word level semantic information;
and the emotion recognition layer splices the sentence-level semantic information, the multi-element group-level semantic information and the word-level semantic information to generate a long sequence matrix, so as to obtain the probability distribution of the to-be-processed sentence on each emotion category.
8. An electronic device, comprising: a processor, a memory, and a bus, wherein,
the processor and the memory are communicated with each other through the bus;
the memory stores program instructions executable by the processor, the processor invoking the program instructions to perform the method of any of claims 1 to 6.
9. A non-transitory computer-readable storage medium storing computer instructions that cause a computer to perform the method of any one of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110317324.8A CN112699679B (en) | 2021-03-25 | 2021-03-25 | Emotion recognition method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110317324.8A CN112699679B (en) | 2021-03-25 | 2021-03-25 | Emotion recognition method and device, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112699679A CN112699679A (en) | 2021-04-23 |
| CN112699679B true CN112699679B (en) | 2021-06-29 |
Family
ID=75515783
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110317324.8A Active CN112699679B (en) | 2021-03-25 | 2021-03-25 | Emotion recognition method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112699679B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113157892A (en) * | 2021-05-24 | 2021-07-23 | 中国平安人寿保险股份有限公司 | User intention processing method and device, computer equipment and storage medium |
| CN113283232A (en) * | 2021-05-31 | 2021-08-20 | 支付宝(杭州)信息技术有限公司 | Method and device for automatically analyzing private information in text |
| CN113729711B (en) * | 2021-09-30 | 2023-10-13 | 深圳航天科技创新研究院 | Electroencephalogram signal analysis method, device, equipment and storage medium |
| CN116150697A (en) * | 2023-04-19 | 2023-05-23 | 上海钐昆网络科技有限公司 | Abnormal application identification method, device, equipment, storage medium and product |
| CN116312486B (en) * | 2023-05-25 | 2023-08-11 | 中国医学科学院阜外医院 | Emotion recognition method, emotion recognition device, computer-readable storage medium and electronic device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108415972A (en) * | 2018-02-08 | 2018-08-17 | 合肥工业大学 | text emotion processing method |
| CN109753566A (en) * | 2019-01-09 | 2019-05-14 | 大连民族大学 | The model training method of cross-cutting sentiment analysis based on convolutional neural networks |
| CN111125354A (en) * | 2018-10-31 | 2020-05-08 | 北京国双科技有限公司 | Text classification method and device |
| CN111368082A (en) * | 2020-03-03 | 2020-07-03 | 南京信息工程大学 | Emotion analysis method for domain adaptive word embedding based on hierarchical network |
| CN111858933A (en) * | 2020-07-10 | 2020-10-30 | 暨南大学 | Character-based hierarchical text emotion analysis method and system |
| CN112287106A (en) * | 2020-10-28 | 2021-01-29 | 中国计量大学 | Online comment emotion classification method based on dual-channel hybrid neural network |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108763204A (en) * | 2018-05-21 | 2018-11-06 | 浙江大学 | A kind of multi-level text emotion feature extracting method and model |
| CN109508377A (en) * | 2018-11-26 | 2019-03-22 | 南京云思创智信息科技有限公司 | Text feature, device, chat robots and storage medium based on Fusion Model |
| CN110298037B (en) * | 2019-06-13 | 2023-08-04 | 同济大学 | Convolutional neural network matching text recognition method based on enhanced attention mechanism |
-
2021
- 2021-03-25 CN CN202110317324.8A patent/CN112699679B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108415972A (en) * | 2018-02-08 | 2018-08-17 | 合肥工业大学 | text emotion processing method |
| CN111125354A (en) * | 2018-10-31 | 2020-05-08 | 北京国双科技有限公司 | Text classification method and device |
| CN109753566A (en) * | 2019-01-09 | 2019-05-14 | 大连民族大学 | The model training method of cross-cutting sentiment analysis based on convolutional neural networks |
| CN111368082A (en) * | 2020-03-03 | 2020-07-03 | 南京信息工程大学 | Emotion analysis method for domain adaptive word embedding based on hierarchical network |
| CN111858933A (en) * | 2020-07-10 | 2020-10-30 | 暨南大学 | Character-based hierarchical text emotion analysis method and system |
| CN112287106A (en) * | 2020-10-28 | 2021-01-29 | 中国计量大学 | Online comment emotion classification method based on dual-channel hybrid neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112699679A (en) | 2021-04-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112699679B (en) | Emotion recognition method and device, electronic equipment and storage medium | |
| CN109145276A (en) | A kind of text correction method after speech-to-text based on phonetic | |
| CN110598203A (en) | Military imagination document entity information extraction method and device combined with dictionary | |
| CN111177366A (en) | Method, device and system for automatically generating extraction type document abstract based on query mechanism | |
| CN113255320A (en) | Entity relation extraction method and device based on syntax tree and graph attention machine mechanism | |
| CN109614620B (en) | HowNet-based graph model word sense disambiguation method and system | |
| CN111858878B (en) | Method, system and storage medium for automatically extracting answer from natural language text | |
| WO2023137911A1 (en) | Intention classification method and apparatus based on small-sample corpus, and computer device | |
| CN110502757B (en) | Natural language emotion analysis method | |
| CN112925904B (en) | Lightweight text classification method based on Tucker decomposition | |
| CN112860871B (en) | Natural language understanding model training method, natural language understanding method and device | |
| CN113360582B (en) | Relation classification method and system based on BERT model fusion multi-entity information | |
| CN111177351A (en) | Method, device and system for acquiring natural language expression intention based on rule | |
| CN114020906A (en) | Chinese medical text information matching method and system based on twin neural network | |
| CN110969023B (en) | Text similarity determination method and device | |
| CN111368544A (en) | Named entity identification method and device | |
| WO2023178802A1 (en) | Named entity recognition method and apparatus, device, and computer readable storage medium | |
| CN115064154A (en) | Method and device for generating mixed language voice recognition model | |
| CN111444720A (en) | Named entity recognition method for English text | |
| CN113065352B (en) | Method for identifying operation content of power grid dispatching work text | |
| CN113486143A (en) | User portrait generation method based on multi-level text representation and model fusion | |
| CN113065350A (en) | Biomedical text word sense disambiguation method based on attention neural network | |
| CN117454898A (en) | Method and device for realizing legal entity standardized output according to input text | |
| CN111177381A (en) | Slot filling and intention detection joint modeling method based on context vector feedback | |
| CN111401069A (en) | Intention recognition method and intention recognition device for conversation text and terminal |
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 |