CN114580430B - Method for extracting fish disease description emotion words based on neural network - Google Patents
Method for extracting fish disease description emotion words based on neural network Download PDFInfo
- Publication number
- CN114580430B CN114580430B CN202210172472.XA CN202210172472A CN114580430B CN 114580430 B CN114580430 B CN 114580430B CN 202210172472 A CN202210172472 A CN 202210172472A CN 114580430 B CN114580430 B CN 114580430B
- Authority
- CN
- China
- Prior art keywords
- emotion
- text
- model
- fish
- data
- 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
- 230000008451 emotion Effects 0.000 title claims abstract description 62
- 208000010824 fish disease Diseases 0.000 title claims abstract description 45
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 18
- 239000013598 vector Substances 0.000 claims abstract description 31
- 238000004458 analytical method Methods 0.000 claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 17
- 238000003745 diagnosis Methods 0.000 claims abstract description 16
- 238000000605 extraction Methods 0.000 claims abstract description 6
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims abstract description 5
- 201000010099 disease Diseases 0.000 claims abstract 4
- 241000251468 Actinopterygii Species 0.000 claims description 15
- 238000009826 distribution Methods 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 8
- 230000002996 emotional effect Effects 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 230000007935 neutral effect Effects 0.000 claims description 6
- 238000012795 verification Methods 0.000 claims description 5
- 238000012360 testing method Methods 0.000 claims description 4
- 239000003086 colorant Substances 0.000 claims description 3
- 238000007405 data analysis Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims 1
- 238000013145 classification model Methods 0.000 abstract description 2
- 238000002372 labelling Methods 0.000 description 6
- 238000013527 convolutional neural network Methods 0.000 description 5
- 238000003058 natural language processing Methods 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 238000005266 casting Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 238000001727 in vivo Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 210000000936 intestine Anatomy 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 210000002784 stomach Anatomy 0.000 description 1
- 238000009827 uniform distribution Methods 0.000 description 1
- 210000001835 viscera Anatomy 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
- Y02A40/80—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in fisheries management
- Y02A40/81—Aquaculture, e.g. of fish
Abstract
A method for extracting a fish disease description emotion word based on a neural network belongs to the technical field of emotion word analysis. Based on the prior knowledge, the method learns the emotion knowledge in the text semantic information through a neural network, thereby assisting in remote disease diagnosis. The method comprises the steps of providing a series of fish disease descriptions at an input end by a user, adding manually marked fish disease aspects and emotion polarities to form a data set, transmitting the data set into a pre-training model, converting the data set into word vectors, and transmitting the word vectors into a sequence model to process time sequence relations in sentences. Finally, the processed semantic information is transmitted into a classification model to finish extraction and analysis of emotion words in fish disease description. Compared with the existing fish disease diagnosis method based on the expert system, the method is used for reducing dependence on priori knowledge and rules and extracting emotion word parts in semantic information.
Description
Technical Field
The invention relates to the technical field of emotion word analysis based on aspects, in particular to a method for extracting a fish disease description emotion word based on a neural network.
Background
With the development of computer technology, the number of people using the internet is rapidly increasing, so far, the number of global mobile phone users is more than 50 hundred million, and the number of internet users is 45 hundred million. Of these, there are 42 billions of social media users. These numbers account for the majority of the world's general population. It is conceivable that the internet will produce an inconceivable number every day. This level of data gives the opportunity for artificial intelligence to evolve at a high rate, and artificial intelligence also deeply alters our lifestyle.
Natural language processing (Natural Language Processing, NLP) is an important direction in the field of artificial intelligence, a subject of research on how to let machines understand human language, and a method for realizing effective interaction between human and computer in natural language. Meanwhile, as the number of network users increases, more and more people can publish own ideas on the social platform and share own ideas. Thus, on many open platforms, there are numerous utterances with emotional colors and tendencies. Analysis of these utterances is of great importance to reality. The method not only can predict the preference of clients and the emotion of people, but also can predict risks. So the emotion analysis task at the present stage is very critical.
In recent years, emotion analysis has become one of the most active research directions in NLP, and has been widely used in information retrieval and text mining. Because the Internet serves as an important social platform for expression and sharing, the Internet brings rich topics containing emotional tendency to users. Text emotion analysis is the process of analyzing, processing, generalizing and reasoning the text with emotion tendencies. Wherein the aspect-based emotion analysis subtask helps merchants and businesses obtain valuable feedback information, thereby improving their products. To date, emotion analysis based on aspects is mainly a text dataset in the conventional field. Few people pay attention to texts containing human emotional tendencies in professional fields such as fish diseases, but more and more adjectives to human emotional tendencies are used in the text descriptions, and the degree words express the ideas and ideas of professionals. These fish disease descriptive text with a large emotional tendency enables us to complete emotion extraction and analysis of different aspects.
Disclosure of Invention
The invention aims at overcoming the defects of the prior art and providing semantic analysis for fish disease description by means of a neural network, and particularly aims at emotion words in texts. According to the invention, text vectorization is carried out based on a pre-training model, a sequence model processes a text time sequence relation, and classification model completion aspect and emotion polarity prediction are carried out. In the data processing, a special data iterator is arranged, and the input format of the model is met. At the output end, the output result is a predefined aspect set and emotion polarity set. In this process, except for the data preparation and labeling process, which is performed manually, the rest is performed automatically.
The invention provides a method for extracting fish disease description emotion words based on a neural network, which comprises the following specific implementation processes:
the first step: based on the diagnosis process of the fish diseases under the line, aspect and emotion polarity division is carried out. Firstly, according to the practical basis, clinical manifestations are mainly in the diagnosis process, and space-time factors are auxiliary. We therefore divide the description of fish diseases into two main categories: 1. clinical factors, 2. Space-time factors. The collected text data features are then specifically analyzed and subdivided, wherein clinical factors consist of five aspects of body surface, body, posture, physique and fish gill, and space-time factors consist of two parts of environment and time. Then, after referring to text data, the emotion polarity classification is performed, and the real problem is fish disease diagnosis, so that the text is neutral or negative description, and the emotion polarity is classified into four types of positive, neutral, negative and negative according to specific problems. In summary, the fish disease description consists of 7 aspects, with emotional polarity consisting of four degrees.
Environment: refers to the surrounding environment, the water body environment, the geographic environment and the like of the fish.
Time section: refers to the season of the fish every year, the month and the morning and evening of the day, etc.
Body surface: refers to the manifestations of disorders on the skin and the oral characteristics.
In vivo: refers to the internal features of fish, such as intestines and stomach, viscera.
Posture is as follows: refers to external manifestations of fish, such as: eating manifestations, active manifestations, etc
Physical constitution: refers to the body length, weight, fat and thin of fish
Fish gill: refers to the state of the part of the gill of the fish.
And a second step of: the method is to process the data set, preprocess the collected fish disease description, and remove blank spaces and non-Chinese characters. Then, manually labeling the category and emotion polarity, wherein the method adopts three persons to label the same data set, and the labeling result is determined by more ticket obtaining numbers. After that, the data set is subjected to three aspects of data distribution, data annotation distribution and correlation coefficient within the data set. Finally, the whole data set is processed according to the following steps of 6:2:2, dividing the training set, the verification set and the test set, and finishing the process.
And a third step of: and combining a neural network to carry out text vectorization, text time sequence processing and classification prediction.
A. Firstly, three data sets are integrated into three data iterators, the configuration of the iterators is transmitted according to the specification of the BERT model (the length of sentences required by the Chinese BERT model is the longest of 32), and meanwhile, the calculation power provided by equipment is considered, the size of Batch is set to be 8, namely, only 8 sentences are transmitted at one time. After passing the BERT model, the text becomes transformed into text vectors.
B. Because of the text vector, which contains a lot of timing information, we need to process the transformed text vector into the sequence model (bilstm+attention), while the parameters of the model are continuously optimized. In the device effort considered, our bar BiLSTM hidden layer size was set to 512.
C. Finally, since the problem itself is a task of emotion analysis, which is a category of classification tasks, we choose the normal layer as the classification layer after the sequence model.
Fourth step: in the specific application process, the whole model reaches an optimal parameter state after training. At this time, the specific fish description can be transferred into the model according to sentences, and the model can output aspects and emotion colors of the fish description, so that semantic information of the fish description is obtained and is used for assisting in fish disease diagnosis.
The beneficial effects of the invention are as follows: the existing remote fish disease diagnosis methods all depend on prior knowledge of experts and system rule formulation. This process ignores the data information, and the present invention is to extract the affective information part of the data information. The neural network is combined on the basis of the existing system, so that aspects and emotion polarities in the description of the fish diseases can be automatically identified. Compared with the existing method, the method optimizes the existing fish disease diagnosis expert system and fills the blank of the fish disease diagnosis based on semantic information. Meanwhile, the invention reduces the manual participation and improves the efficiency. We found that we can reach 81% -84% effect by training the classifier using a simple bert+ feature extraction approach, which demonstrates that our proposed dataset is of very high quality.
Drawings
Fig. 1 is a flowchart of a method for extracting a fish disease description emotion word based on a neural network.
Fig. 2 is a model structure diagram of a method for extracting a fish disease description emotion word based on a neural network.
Detailed Description
The invention will be described in further detail with reference to the drawings and the specific examples.
As shown in FIG. 1, the invention provides a method for describing emotion words of fish diseases based on a neural network, which comprises the following steps:
the first step: based on the diagnosis process of the fish diseases under the line, aspect and emotion polarity division is carried out. Firstly, according to the practical basis, clinical manifestations are mainly in the diagnosis process, and space-time factors are auxiliary. We therefore divide the description of fish diseases into two main categories: 1. clinical factors, 2. Space-time factors. The collected text data features are then specifically analyzed and subdivided, wherein clinical factors consist of five aspects of body surface, body, posture, physique and fish gill, and space-time factors consist of two parts of environment and time. Then, after referring to text data, the emotion polarity classification is performed, and the real problem is fish disease diagnosis, so that the text is neutral or negative description, and the emotion polarity is classified into four types of positive, neutral, negative and negative according to specific problems. In summary, the fish disease description consists of 7 aspects, the emotion polarity consists of four degrees, and specific data are shown in reference to Table three.
And a second step of: the method is to process the data set, preprocess the collected fish disease description, and remove blank spaces and non-Chinese characters. Then, manually labeling the category and emotion polarity, wherein the method adopts three persons to label the same data set, and the labeling result is determined by more ticket obtaining numbers. After that, the data set is subjected to data analysis from three aspects of the data distribution, the data annotation distribution and the correlation coefficient within the data set. Finally, the whole data set is processed according to the following steps of 6:2: and 2, dividing the training set into a verification set and a test set, and completing all manual labeling work.
And a third step of: the fish disease description emotion word method model based on the neural network consists of three parts: firstly, a semantic embedding layer obtains vectorized text representation; secondly, a semantic decision layer obtains deep semantic information through a sequence model; and thirdly, a classification layer for predicting emotion types and emotion polarities. The method specifically comprises the following steps:
A. firstly, three data sets are integrated into three data iterators, the configuration of the iterators is transmitted according to the specification of the BERT model (the length of sentences required by the Chinese BERT model is the longest of 32), and meanwhile, the calculation power provided by equipment is considered, the size of Batch is set to be 8, namely, only 8 sentences are transmitted at one time. The data in the iterator is then transferred into the BERT model to a text vector. The specific operation steps are as follows: the first choice is to enter a text sequence T= { T with the size of Batch 0 ,t 1 ,t 2 ,t 3 ,t 4 ,t 5 ,t 6 ,t 7 Each t= { w } 1 ,w 2 ,w3,w 4 ,., wn consists of n words. Each T in the text sequence T will then get its original word casting by being mapped to a vector space of fixed dimensions using BERT (Bidirectional Encoder Representation from Transformers), and then each w in T will depend on which sentence and the way it belongs toThe position in the sentence results in one segment embedding and position embedding, and finally the three vectors are added to obtain the input vector x for each w. There are a total of 12 layers of transformers in Bert, each of which has an Encoder and a Decoder that encode and decode incoming word vectors, respectively. The feedforward neural network and normalization are continuously performed in the encoding and decoding processes. Finally, a 768-dimensional word vector is obtained at the output end.
hi＝Bert(xi) i∈(1,N) (1)
Wherein h is i Is a word vector representation at the output, hi= { h i1 ,h i2 ,...,h in And the feature vector of the sentence. N represents the sentence length.
B. The layer is a semantic decision layer, and is a text vector, which contains a large amount of time sequence information, so that the converted text vector needs to be transmitted into a sequence model (BiLSTM+attribute) to be processed, and meanwhile, parameters of the model are continuously optimized. In the device effort considered, we set the hidden layer size of BiLSTM to 512. It contains a forward LSTM and a reverse LSTM, which separate text from w 1 To w n And w n To w 1 Obtaining two LSTM feature representations
Forward Li=Forward LSTM (Hi) i ε (1, N) (2)
Reverse li=reverse LSTM (Hi) i e (N, 1) (3)
Finally we obtain at the output a BiLSTM feature representation for each w:
bi= [ Forward LSTM, reverse LSTM ] i ε (1, N) (4)
Where Bi is the characteristic representation of text after BiLSTM, and N is the sentence length.
And then the whole sentence is transmitted into an Attention module, and weight distribution is automatically carried out.
A＝∑Bi*Hi i∈(1,N) (5)
Wherein Bi represents a characteristic representation of text, H i A representation of the intent of each word in the text is calculated.
C. Finally, since the problem itself is a task of emotion analysis, which is a category of classification tasks, we choose the normal layer as the classification layer after the sequence model. This layer predicts the likelihood of aspect category and emotion polarity through the fully connected layer.
out1，out2＝Linear(A) out1∈U，out2∈Q (6)
Where out1 and out2 are aspect and emotion outputs, respectively, U is the aspect category, and Q is the emotion category.
Fourth step: training a model, namely calculating cross entropy of yR and T, and then completing model training by optimizing the cross entropy. Such that the BERT parameter matrix W bert BiLSTM+attention parameter matrix W bi And W is at And a parameter matrix W of Linear L Optimally. After that, the whole model can automatically label aspects and emotion polarities in the description of the fish diseases and is used for assisting in diagnosis of the fish diseases.
Fifth step, experiment verification:
1) Experimental environment and data
TABLE 1 extraction method data set of fish disease description emotion words based on neural network
Table 2 aspect-based emotion analysis dataset contrast
| Year of year | Data volume | Aspects of the invention | FIELD | Professional field | |
| SemEval-2014 | 2014 | 7686 | 5 | 2 | × |
| SemEval-2015 | 2015 | 4766 | 26 | 3 | × |
| SemEval-2016 | 2016 | 5984 | 28 | 2 | × |
| SentiHood | 2016 | 5215 | 2 | 1 | × |
| Cricket ballComment on | 2018 | 3034 | 5 | 1 | × |
| Our | 2021 | 2300 | 5 | 1 | √ |
We constructed a new set of aspect-level emotion analysis data for fish disease descriptions. Table 2 shows a comparison of the disclosed aspect level emotion analysis with our proposed dataset. One problem with these existing datasets is that they all fall within the general domain without data information within the professional domain. Another problem is that existing domains do not guarantee the generalization capability required in fish disease descriptions. To address the previous drawbacks, we propose this dataset consisting of 1 field (fish disease) and 7 aspects. Our dataset is the only dataset that contains knowledge of the fish disease domain.
To evaluate the model and experimental results, we used two indices, accuracy (Acc) and F1.
Where Rec is the recall, all is All data, and correct is the predicted correct data.
As for the modelThe super parameters of (a) are as follows: the position embedding and weight matrix is initialized with a uniform distribution U (-0.1,0.1) and the bias is initialized to zero. drop_out is set to 0.3, and learning rate is 10 -5 The Bert hidden layer size is 768, the bilstm hidden layer is 256, and the layer number is 1. The model adopts a Bertadam optimizer, and the learning rate is 10 -5 . And stopping the model when the maximum training times are 1000 times, namely lifting.
2) Contrast model
To evaluate the performance of our proposed model. The experimental results of the three baseline models are compared, and each baseline model is briefly introduced.
First three baseline models:
bert+linear: the sentence is encoded by the Bert only to obtain a word vector and the word vector is used as a characteristic representation of the sentence to be transmitted into the full-connection layer for training. The advantage is that the model is simple, and the disadvantage is that the model is difficult to fit to an ideal state.
Bert+BiLSTM, a word vector is obtained by encoding a sentence by Bert, then a characteristic representation of the sentence is obtained by entering BiLSTM, and then the full concatenation layer is entered. The model can be optimized and simple later, and meanwhile, the BiLSTM processing sequence task is more reasonable. The disadvantage is that feature representation does not solve the problem of dependency between sentence words.
And (4) Bert+CNN, namely, coding sentences through Bert to obtain word vectors, then, transmitting the word vectors into CNN to obtain a characteristic representation of the sentences, and then, transmitting the characteristic representation into a full-connection layer for training. The model has the advantages that internal sharing parameters can be realized by means of rolling and pooling, and training is simplified. The disadvantage is that it is difficult to capture long-range features, limited by the convolution kernel size. Experimental results:
table 3 AE task ACC and F1 for each model
| Model | ACC | F1 |
| Bert+Linear | 81.96％ | 82.08％ |
| Bert+BiLSTM | 82.49％ | 82.55％ |
| Bert+CNN | 83.00％ | 83.04％ |
| Bert+BiLSTM+ATT | 83.55％ | 83.64％ |
3) Analysis of results
It is first evident that feature representation using either BiLSTM or CNN is better than directly training the model using the Bert-encoded word vector.
(2) Even when processing sequential tasks, CNN performs better on certain specific texts than BiLSTM.
(3) The effect of the model we propose is entirely better than the other three, which illustrates the effectiveness we introduce.
(4) From table 3, we found that we could reach 81% -84% effect when we used simple bert+ feature extraction to train the classifier, demonstrating that our proposed dataset is of very high quality.
The foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (1)
1. The extraction method of the fish disease description emotion words based on the neural network is characterized by mainly comprising the following steps of:
s1, classifying aspect categories and emotion polarities based on an offline fish disease diagnosis process;
s11, aspect classification
According to the premise that clinical manifestation is dominant and space-time factors are auxiliary in the diagnosis process, fish disease description is divided into two main categories: clinical factors and space-time factors;
analyzing the collected text data characteristics in detail to subdivide the text data characteristics;
wherein clinical factors include five aspects of body surface, body, posture, physique and fish gill, and space-time factors include two parts of environment and time and section;
s13, emotion polarity division
Combining the reference text with specific problems, dividing emotion polarities into positive, neutral, negative and negative; thus, aspects described by fish disease include 7 aspects of body surface, body, posture, physique, gill, environment and time, emotional polarity includes four positive, neutral, negative and negative;
s2, processing a data set:
s21, preprocessing the collected fish disease description, and removing blank spaces and non-Chinese characters;
s22, manually marking the category and emotion polarity, marking the same data set by three persons, and determining that the marking result is more in number of votes;
s23, carrying out data analysis on the data set from three aspects of data distribution, data annotation distribution and correlation coefficients in the data set, and carrying out data analysis on the data set according to 6:2:2, dividing the training set, the verification set and the test set;
s3, a fish disease description emotion word method model based on a neural network comprises three main parts: firstly, a semantic embedding layer obtains vectorized text representation; secondly, a semantic decision layer obtains deep semantic information through a sequence model; the classification layer is used for predicting emotion types and emotion polarities; the method specifically comprises the following steps:
s31, a semantic embedding layer: integrating three data sets of a training set, a verification set and a test set into three data iterators, wherein the configuration of the iterators is transmitted according to the specification of a BERT model, and Batch is set to be 8, namely only 8 sentences are transmitted at one time; the data in the iterator is transferred into the BERT model to be converted into text vectors, and the specific operation steps are as follows:
(1) A Batch-sized text sequence T= { T is imported 0 ，t 1 ，t 2 ，t 3 ，t 4 ，t 5 ，t 6 ，t 7 Each t= { w } 1 ，w 2 ，w 3 ，w 4 ,., wn consists of n words;
(2) Each text sequence T in the text sequence T can be mapped to a vector space with a fixed dimension by using BERT to obtain an initial word filling, and each phrase w in the text sequence T can obtain segment embedding and position embedding according to the sentence to which the phrase w belongs and the position of the sentence;
(3) Adding the three vectors to obtain an input vector x of each phrase w; a total of 12 layers of transformers in the Bert, each transformer having an Encoder and a Decoder for encoding and decoding the incoming word vector, respectively; continuously passing through a feedforward neural network and normalizing in the encoding and decoding processes; finally obtaining 768-dimension word vectors at the output end;
hi＝Bert(xi)i∈(1，N) (1)
wherein h is i Is word vector representation of the output end, and N represents sentence length;
s32, a semantic decision layer: because the text vector contains time sequence information, the converted text vector is transmitted into a sequence model BiLSTM+attribute for processing, and meanwhile, parameters of the model are continuously optimized;
the hidden layer size of BiLSTM is set to 512, which contains a forward LSTM and a reverse LSTM, respectively, to separate text from w 1 To w n And w n To w 1 Obtaining two LSTM feature representations
Forward Li=Forward LSTM (Hi) i ε (1, N) (2)
Reverse li=reverse LSTM (Hi) i e (N, 1) (3)
Finally, the BiLSTM characteristic representation of each w is obtained at the output:
bi= [ Forward LSTM, reverse LSTM ] i ε (1, N) (4)
Wherein Bi is the characteristic representation of the text after BiLSTM, and N represents the sentence length;
then the whole sentence is transmitted into an Attention module, and weight distribution is automatically carried out;
A＝∑Bi*Hii∈(1，N) (5)
wherein Bi represents a characteristic representation of text, H i Calculating a representation of the intent of each word in the text, hi= { Hi1, hi2,.. hin } representing a feature vector of the sentence;
s33, classifying layer: as the problem is an emotion analysis task and belongs to the category of classification tasks, a layer is selected as a classification layer after a sequence model; the layer predicts the possibility of aspect category and emotion polarity through the full connection layer;
out1，out2＝Linear(A) out1∈U，out2∈Q (6)
wherein, linear (A) is a layer, out1 and out2 are aspect and emotion output respectively, U is aspect category, Q is emotion category;
s4, training a model: calculating y R And the cross entropy of T, and completing model training by optimizing the cross entropy; such that the BERT parameter matrix W bert BiLSTM+attention parameter matrix W bi And W is at Parameter matrix W of Linear L Optimizing;
wherein y is R T is respectively expressed as a prediction label and a real label, and L is a parameter matrix of the whole model;
s5, transmitting the specific disease fish description into a trained model according to sentences, wherein the model can output aspect categories and emotion colors of the disease fish description, and semantic information of the disease fish description is obtained and used for assisting in fish disease diagnosis.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210172472.XA CN114580430B (en) | 2022-02-24 | 2022-02-24 | Method for extracting fish disease description emotion words based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210172472.XA CN114580430B (en) | 2022-02-24 | 2022-02-24 | Method for extracting fish disease description emotion words based on neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114580430A CN114580430A (en) | 2022-06-03 |
| CN114580430B true CN114580430B (en) | 2024-04-05 |
Family
ID=81770517
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210172472.XA Active CN114580430B (en) | 2022-02-24 | 2022-02-24 | Method for extracting fish disease description emotion words based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114580430B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110245229A (en) * | 2019-04-30 | 2019-09-17 | 中山大学 | A kind of deep learning theme sensibility classification method based on data enhancing |
| KR20210004057A (en) * | 2019-07-03 | 2021-01-13 | 인하대학교 산학협력단 | Machine Learning and Semantic Knowledge-based Big Data Analysis: A Novel Healthcare Monitoring Method and Apparatus Using Wearable Sensors and Social Networking Data |
| US11194972B1 (en) * | 2021-02-19 | 2021-12-07 | Institute Of Automation, Chinese Academy Of Sciences | Semantic sentiment analysis method fusing in-depth features and time sequence models |
| CN114049926A (en) * | 2021-10-27 | 2022-02-15 | 徐州医科大学 | Electronic medical record text classification method |
-
2022
- 2022-02-24 CN CN202210172472.XA patent/CN114580430B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110245229A (en) * | 2019-04-30 | 2019-09-17 | 中山大学 | A kind of deep learning theme sensibility classification method based on data enhancing |
| KR20210004057A (en) * | 2019-07-03 | 2021-01-13 | 인하대학교 산학협력단 | Machine Learning and Semantic Knowledge-based Big Data Analysis: A Novel Healthcare Monitoring Method and Apparatus Using Wearable Sensors and Social Networking Data |
| US11194972B1 (en) * | 2021-02-19 | 2021-12-07 | Institute Of Automation, Chinese Academy Of Sciences | Semantic sentiment analysis method fusing in-depth features and time sequence models |
| CN114049926A (en) * | 2021-10-27 | 2022-02-15 | 徐州医科大学 | Electronic medical record text classification method |
Non-Patent Citations (2)
| Title |
|---|
| 基于BERT的文本情感分析;刘思琴;冯胥睿瑞;;信息安全研究;20200305(第03期);全文 * |
| 基于复杂句式短文本情感分类研究;李毅捷;段利国;李爱萍;;现代电子技术;20181112(第22期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114580430A (en) | 2022-06-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111444709B (en) | Text classification method, device, storage medium and equipment | |
| Zhang et al. | Textual sentiment analysis via three different attention convolutional neural networks and cross-modality consistent regression | |
| CN110032635B (en) | Problem pair matching method and device based on depth feature fusion neural network | |
| CN111581401B (en) | Local citation recommendation system and method based on depth correlation matching | |
| Alwehaibi et al. | Comparison of pre-trained word vectors for arabic text classification using deep learning approach | |
| CN111414461B (en) | Intelligent question-answering method and system fusing knowledge base and user modeling | |
| CN108345585A (en) | A kind of automatic question-answering method based on deep learning | |
| CN105279495A (en) | Video description method based on deep learning and text summarization | |
| CN107832326A (en) | A kind of natural language question-answering method based on deep layer convolutional neural networks | |
| CN109145085A (en) | The calculation method and system of semantic similarity | |
| CN113724882A (en) | Method, apparatus, device and medium for constructing user portrait based on inquiry session | |
| CN109783825A (en) | A kind of ancient Chinese prose interpretation method neural network based | |
| CN115393933A (en) | Video face emotion recognition method based on frame attention mechanism | |
| Zouzou et al. | Text sentiment analysis with CNN & GRU model using GloVe | |
| CN114462385A (en) | Text segmentation method and device | |
| CN110874392B (en) | Text network information fusion embedding method based on depth bidirectional attention mechanism | |
| CN116467443A (en) | Topic identification-based online public opinion text classification method | |
| CN111581364A (en) | Chinese intelligent question-answer short text similarity calculation method oriented to medical field | |
| Sun et al. | Text sentiment analysis based on CNN-BiLSTM-attention model | |
| CN112950414A (en) | Legal text representation method based on decoupling legal elements | |
| CN114580430B (en) | Method for extracting fish disease description emotion words based on neural network | |
| Yang et al. | Multi-intent text classification using dual channel convolutional neural network | |
| CN114417863A (en) | Word weight generation model training method and device and word weight generation method and device | |
| Shan et al. | A novel semantic matching method for chatbots based on convolutional neural network and attention mechanism | |
| CN113792121A (en) | Reading understanding model training method and device and reading understanding method and device |
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 |