CN110781273B - Text data processing method and device, electronic equipment and storage medium - Google Patents
Text data processing method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN110781273B CN110781273B CN201910873641.0A CN201910873641A CN110781273B CN 110781273 B CN110781273 B CN 110781273B CN 201910873641 A CN201910873641 A CN 201910873641A CN 110781273 B CN110781273 B CN 110781273B
- Authority
- CN
- China
- Prior art keywords
- matrix
- word vector
- aspect feature
- text data
- text
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- 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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Machine Translation (AREA)
Abstract
The embodiment of the application discloses a text data processing method, a text data processing device, electronic equipment and a storage medium, wherein the method comprises the following steps: the method comprises the steps of obtaining text data to be classified, extracting aspect feature words from the text data, obtaining text word vector representation of the text data to be classified and aspect feature word vector representation of the aspect feature words, respectively inputting the text word vector representation and the aspect feature word vector representation into a target neural network, obtaining first hidden layer meanings represented by the text word vector and second hidden layer meanings represented by the aspect feature word vector, obtaining target classification features of the aspect feature words based on the first hidden layer meanings and the second hidden layer meanings, obtaining predicted emotion polarities of the aspect feature words according to the target classification features of the aspect feature words, and carrying out emotion classification on the text data to be classified according to the predicted emotion polarities so as to obtain emotion classification results corresponding to the aspect feature words and display the emotion classification results. The emotion polarity of the aspect feature words is accurately judged based on the attention mechanism and the neural network.
Description
Technical Field
The present application relates to the field of natural language processing technologies, and in particular, to a text data processing method and apparatus, an electronic device, and a storage medium.
Background
Emotion analysis is a basic task in natural language processing, and can be used for mining opinions of users, performing tasks such as data analysis and public opinion monitoring. The emotion analysis can be divided into extraction of emotion information, classification of emotion information and retrieval and induction of emotion information. The emotion classification refers to dividing the text into two or more types which are commendable or derogative according to the meaning and emotion information expressed by the text, and is used for dividing tendency, viewpoint and attitude of a text author. However, one sentence may involve a plurality of different emotions for different aspects, and currently, there is a limitation in emotion classification of text containing a plurality of emotions or aspects.
Disclosure of Invention
The application provides a text data processing method, a text data processing device, an electronic device and a storage medium, so as to overcome the defects.
In a first aspect, an embodiment of the present application provides a text data processing method, where the method includes: acquiring text data to be classified; extracting aspect feature words from the text data to be classified; acquiring text word vector representation of the text data to be classified and aspect feature word vector representation of the aspect feature words; inputting the text word vector representation and the aspect characteristic word vector representation into a target neural network respectively to obtain a first hidden layer meaning represented by the text word vector and a second hidden layer meaning represented by the aspect characteristic word vector, wherein the target neural network is trained in advance and used for outputting the hidden layer meaning represented by the word vector according to the input word vector representation; obtaining target classification features of the aspect feature words based on the first hidden layer meaning and the second hidden layer meaning; obtaining the predicted emotion polarity of the aspect feature words according to the target classification features of the aspect feature words; and carrying out emotion classification on the text data to be classified according to the predicted emotion polarity so as to obtain and display an emotion classification result corresponding to the aspect feature words.
In a second aspect, an embodiment of the present application further provides a text data processing apparatus, where the apparatus includes: the text acquisition module is used for acquiring text data to be classified; the text extraction module is used for extracting aspect feature words from the text data to be classified; the representation obtaining module is used for obtaining text word vector representation of the text data to be classified and aspect feature word vector representation of the aspect feature words; the network learning module is used for inputting the text word vector representation and the aspect characteristic word vector representation into a target neural network respectively to obtain a first hidden layer meaning represented by the text word vector and a second hidden layer meaning represented by the aspect characteristic word vector, wherein the target neural network is trained in advance and is used for outputting the hidden layer meaning represented by the word vector according to the input word vector representation; the target classification module is used for obtaining target classification characteristics of the aspect characteristic words based on the first hidden layer meaning and the second hidden layer meaning; the emotion prediction module is used for acquiring the predicted emotion polarity of the aspect feature words according to the target classification features of the aspect feature words; and the emotion classification module is used for carrying out emotion classification on the text data to be classified according to the predicted emotion polarity so as to obtain and display an emotion classification result corresponding to the aspect feature words.
In a third aspect, an embodiment of the present application further provides an electronic device, including: one or more processors; a memory; one or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications configured to perform the method of the first aspect.
In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, where a program code is stored in the computer-readable storage medium, and the program code may be called by a processor to execute the method according to the first aspect.
The text data processing method, the text data processing device, the electronic equipment and the storage medium provided by the application acquire text data to be classified, extract aspect feature words from the text data to be classified, acquire text word vector representation of the text data to be classified and aspect feature word vector representation of the aspect feature words, and input the text word vector representation and the aspect feature word vector representation into a target neural network respectively to obtain a first hidden layer meaning represented by the text word vector and a second hidden layer meaning represented by the aspect feature word vector, wherein the target neural network is trained in advance and used for outputting the hidden layer meaning represented by the word vector according to the input word vector representation, then obtain target classification features of the aspect feature words based on the first hidden layer meaning and the second hidden layer meaning, and then obtain predicted emotion polarity of the aspect feature words according to the target classification features of the aspect feature words, and finally, carrying out emotion classification on the text data to be classified according to the predicted emotion polarity so as to obtain and display an emotion classification result corresponding to the aspect feature words. Therefore, the attention mechanism model based on the neural network of the Aspect Level (Aspect Level) can better acquire the interaction information of the Aspect characteristic words and the text data to be classified and make full use of the interaction information, and therefore the accuracy of the emotion classification of the Aspect Level text is improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a flowchart illustrating a method of processing text data according to an embodiment of the present application;
FIG. 2 is a flow chart of a method of processing text data according to another embodiment of the present application;
fig. 3 is a flowchart illustrating a method of steps S2051 to S2053 in a text data processing method according to another embodiment of the present application;
fig. 4 shows a flowchart of a method from step S2081 to step S2083 in a text data processing method according to another embodiment of the present application;
FIG. 5 is a flow chart illustrating a text data processing method according to another embodiment of the present application;
FIG. 6 shows a block diagram of a text data processing apparatus provided in an embodiment of the present application;
fig. 7 shows a block diagram of an electronic device provided in an embodiment of the present application;
Fig. 8 illustrates a storage unit for storing or carrying program codes for implementing a text data processing method according to an embodiment of the present application.
Detailed Description
In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application.
The emotion classification includes three different levels, which are a document Level, a sentence Level, and an Aspect Level (Aspect Level). Document-level sentiment classification classifies opinion-conscious documents (e.g., product reviews) into overall positive or negative opinions. It treats the entire document as the basic unit of information and assumes that the document is from a point of view, including a point of view of a single entity (e.g., a certain model of cell phone). Sentence-level sentiment classification classifies individual sentences within a document. However, the individual sentences cannot be assumed to be from a clear perspective. Compared with document level and statement level emotion classification, the emotion classification of the aspect level is more fine-grained. Its task is to extract and summarize people's opinions about an entity and the characteristics of the entity (also known as targets, Aspect signatures). For example, a product review, the purpose of the sentiment classification at the aspect level is to summarize the positive and negative opinions of different aspects of a product or event, respectively.
In the case that a plurality of aspect feature words exist in a sentence, in order to analyze the emotion polarities of different aspect feature words, a specific model is required to complete the task, which is different from the traditional emotion classification model. For example, in the comment sentence "the restaurant tastes good in food, but the waiter's service attitude is poor", there are two aspect feature words: "taste" and "service attitude" where the emotional polarity of "taste" is positive and the emotional polarity of "service attitude" is negative. The emotional polarity of the entire sentence of this example consists of both positive and negative. If we do not consider the information of the aspect feature words, it is difficult to judge the emotion polarity of the sentence, and this type of error usually exists in the general emotion classification task.
Currently, emotion classification tasks are generally handled based on an IAN model that uses two long-short-term memory networks (LSTM) to model a facet feature word and its context, respectively. It generates an attention vector for the target using the hidden state of the facet feature words and their context. Based on these two attention vectors, an operation is performed to obtain a final representation vector. The method comprises the following specific steps:
The method comprises the following steps: a word embedding matrix of facet feature words and their contexts is obtained.
Step two: using two LSTMs to learn the hidden semantics of the word embedding matrix of the aspect characteristic words and the contexts thereof to obtain a hidden state matrix
And
where m is the aspect feature word length and n is the context sentence length.
Step three: to hTAnd hCAveraging to obtain tavgAnd cavg. The obtained hidden state matrix average value t of the aspect feature wordsavgWith its context hidden state matrix hCMean value of the context hidden state matrix of the simultaneous aspect feature words cavgAnd aspect feature word hidden state matrix hTThrough calculation, two attention vectors are respectively obtained
And
where gamma is a scoring function used to calculate
And
the importance of the context.
Step four: calculating the expression vector of the aspect characteristic words and the context thereof through the attention vector obtained in the step three, wherein the expression vector comprises the following components:
step five: will trAnd crConcatenation to obtain vector dAt the final classification.
However, the above method ignores the interdependency between sentences and aspect feature words, and still needs to be improved in distinguishing the emotion polarities of different aspects in a sentence and improving the accuracy of emotion classification, and the model cannot encode information sequences from back to front.
Therefore, in order to solve the above-mentioned defects, an embodiment of the present application provides a text data processing method, which is applied to an electronic device, where the electronic device may be a smartphone, a tablet, a computer, a wearable electronic device, a server, or other device capable of running a program. The method can overcome the limitation of document level and sentence level emotion classification when a sentence has a plurality of aspect characteristic words. As shown in fig. 1, specifically, the method includes: s101 to S107.
S101: and acquiring text data to be classified.
And acquiring text data to be classified based on the text to be classified, wherein the text data to be classified is in a data form of inputting the text to be classified into the electronic equipment. The text to be classified can be obtained from the internet or the existing offline text. Specifically, the text to be classified may be a document, a sentence, for example, the text to be classified may be a sentence "the food taste of a restaurant is good but the service attitude of a waiter is poor".
S102: and extracting aspect characteristic words from the text data to be classified.
The aspect feature words may be words whose part of speech is noun in the text, for example, in the sentence "the food of a restaurant tastes good, but the service attitude of a waiter is poor", the aspect feature words include "taste", "service attitude".
As an implementation manner, the characteristic word extraction model may be trained based on a Stanford core Natural Language Processing dataset (Stanford CoreNLP), which is a Natural Language Processing (NLP) tool set integrating multiple tools, where the tools include Part of Speech tags (Part of Speech Tagger) including text data tagged with parts of Speech of each word. And extracting the aspect characteristic words from the text data to be classified by using the trained characteristic word extraction model.
S103: and acquiring text word vector representation of the text data to be classified and aspect feature word vector representation of the aspect feature words.
And converting each Word in the text data to be classified into a Word vector matrix by adopting Word Embedding (WE), namely converting the Word into Word vector representation. The word embedding technique is a numerical expression of a word, and generally a word is mapped into a high-dimensional vector (word vector) to express the word.
In one embodiment, a Global Vectors for Word Representation (GloVe) model may be used to obtain a Word vector matrix, and a text Word vector Representation and an aspect feature Word vector Representation may be obtained therefrom. Specifically, based on a GloVe model, a word vector matrix of text data to be classified is obtained and is used as text word vector representation of the text data to be classified, and a word vector matrix of aspect feature words is obtained and is used as aspect feature word vector representation of the aspect feature words. The GloVe model can simultaneously consider a plurality of windows (co-occurrence matrix), introduces global information, is simpler in calculation, and can improve the speed of obtaining word vector representation.
In particular, given a sentence s of length m, s ═ w1；w2；…；wi；…；wm]And an aspect feature word t, t ═ w of length ni；wi+1；…；wi+n+1]Where w represents the words of the text data to be classified, wiAnd characterizing the ith word in the text data to be classified. Mapping S and t to a word vector matrix generated by a GloVe model to respectively obtain text word vector representation S of text data to be classified, wherein [ v ] is S1；v2；…；vi；…；vm]And an aspect feature word vector representation of the aspect feature word T, T ═ vi；vi+1；…；vi+n+1]. For example, sentence s "restaurant food is nice tasting, but waiter's serving attitude is bad", where sentenceLength m of s is 13, w1"restaurant2In term of ═ w3As food, w4Good taste, w13Or "poor". Length n of the first aspect characterising word "taste 11 is ═ 1; second aspect the term "service attitude" length n2＝2。
S104: and respectively inputting the text word vector representation and the aspect characteristic word vector representation into a target neural network to obtain a first hidden layer meaning represented by the text word vector and a second hidden layer meaning represented by the aspect characteristic word vector.
Wherein the target neural network is pre-trained for outputting hidden layer meanings represented by the word vectors according to the input word vector representations. The target neural network may be a Long Short-Term Memory network (LSTM), a bidirectional Long Short-Term Memory network (Bi-LSTM), a Gated Recurrent Unit (GRU), but is not limited thereto.
Respectively inputting the text word vector representation and the aspect characteristic word vector representation into a target neural network, respectively learning the semantics of the text word vector representation and the aspect characteristic word vector representation through the target neural network, and outputting a first hidden layer meaning h represented by the text word vectorSAnd a second hidden layer meaning h represented by the aspect feature word vectorT。
S105: and obtaining the target classification characteristics of the aspect characteristic words based on the first hidden layer meaning and the second hidden layer meaning.
Based on the first hidden layer meaning and the second hidden layer meaning, initial representations of the text data to be classified and the aspect feature words are obtained, two attention matrixes are obtained through a softmax function based on the two initial representations, more concentrated representations are obtained based on the two attention moment matrixes, and then the initial representations of the text data to be classified are combined to obtain the target classification features.
S106: and acquiring the predicted emotion polarity of the aspect feature words according to the target classification features of the aspect feature words.
And (4) regarding the target classification features as final classification features, and inputting the final classification features into a softmax function to calculate the probability that the aspect feature words have a certain emotion polarity. The emotion polarities can include Positive and Negative, and the emotion classification task of the second classification is realized, and can further include Neutral to realize the emotion classification task of the third classification.
Specifically, in one embodiment, the emotion polarities include three labels, c ∈ { P, N, O }, (P stands for Positive, N stands for Negative, and O stands for Neutral), and according to the target classification features of the aspect feature words, the emotion probabilities that the aspect feature words have each of the three emotion polarities can be obtained, specifically, Positive emotion probabilities P that characterize that the feature words have "Positive" emotion polarityPCharacterization of the negative emotion probability P with the emotion polarity "negativeNCharacterization of neutral emotional probability P with emotional polarity "neutralOAnd determining the emotion polarity with the highest emotion probability as the predicted emotion polarity of the aspect characteristic word. If PN>PP>POThen, the emotional polarity is predicted to be "negative".
In some embodiments, "positive" emotional polarity may further include more subdivided emotional categories such as open heart, excitement, and the like, and similarly "negative" may include agony, hate, jealousy, and the like, without limitation.
S107: and carrying out emotion classification on the text data to be classified according to the predicted emotion polarity so as to obtain and display an emotion classification result corresponding to the aspect feature words.
And acquiring the predicted emotion polarity of the aspect characteristic words in the text data to be classified according to the predicted emotion polarity, classifying the text data to be classified according to the predicted emotion polarity, acquiring emotion classification results corresponding to the aspect characteristic words, and displaying the emotion classification results so that a user can know the emotion classification results.
For example, in the sentence "the restaurant has a good taste of food but the serviceability of the servicer is poor", the above-described method is performed with respect to the aspect feature word "taste", the predicted emotion polarity of "taste" is obtained as "positive", and thus the text data to be classified is subjected to emotion classification, the emotion classification result of "taste" is obtained as positive, and the emotion classification result is displayed on the electronic device, so that the user can know the emotion classification result. Similarly, by executing the method for the aspect feature word "service attitude", the predicted emotion polarity of the service attitude "is" negative ", and thus the emotion classification result of the service attitude" is negative and is displayed. Therefore, the text data to be classified can be subjected to emotion classification according to the aspect feature words, the predicted emotion polarity and the emotion classification result of each aspect feature word are obtained respectively, emotion classification based on aspect levels is achieved, the interdependency between the text data to be classified and the aspect feature words is omitted, and higher accuracy can be achieved when the emotion polarities of different aspects in the text are distinguished.
Further, in some embodiments, the emotion classification results may be accumulated for data analysis, and the data analysis results may be displayed on the electronic device, so that the user may obtain statistics of emotion classifications of a plurality of text data to be classified.
The embodiment obtains text data to be classified, extracts aspect feature words from the text data to be classified, then obtains text word vector representation of the text data to be classified and aspect feature word vector representation of the aspect feature words, and inputs the text word vector representation and the aspect feature word vector representation into a target neural network respectively to obtain a first hidden layer meaning represented by the text word vector and a second hidden layer meaning represented by the aspect feature word vector, wherein the target neural network is trained in advance and is used for outputting the hidden layer meaning represented by the word vector according to the input word vector representation, then obtains a target classification feature of the aspect feature words based on the first hidden layer meaning and the second hidden layer meaning, then obtains a predicted emotion polarity of the aspect feature words according to the target classification feature of the aspect feature words, and finally carries out emotion classification on the text data to be classified according to the predicted emotion polarity, and obtaining and displaying the emotion classification result corresponding to the aspect characteristic words. Therefore, the attention mechanism model based on the neural network of the Aspect Level (Aspect Level) can improve the accuracy of the emotion classification of the Aspect Level text.
Referring to fig. 2, a text data processing method provided in the embodiment of the present application is shown, specifically, the method includes: s201 to S209.
S201: and acquiring text data to be classified.
S202: and extracting aspect characteristic words from the text data to be classified.
S203: and acquiring text word vector representation of the text data to be classified and aspect feature word vector representation of the aspect feature words.
In this embodiment, the detailed descriptions of steps S201 to S203 can refer to steps S101 to S103 in the above embodiment, which are not repeated herein.
In one embodiment, step S203 may be implemented by the following code:
s204: and respectively inputting the text word vector representation and the aspect characteristic word vector representation into a target neural network to obtain a first hidden layer meaning represented by the text word vector and a second hidden layer meaning represented by the aspect characteristic word vector.
The target neural network is a bidirectional long-short term memory network (Bi-LSTM), and comprises a first neural network and a second neural network, wherein the first neural network and the second neural network are constructed on the basis of the Bi-LSTM. In one embodiment, the number of hidden layers of the first neural network and the second neural network is 1.
In one embodiment, the Bi-LSTM introduces both forward LSTM units and backward LSTM units in the hidden layer, the forward LSTM units capturing the above feature information and the backward LSTM units capturing the below feature information. Compared with the unidirectional LSTM, the Bi-LSTM based learning of the hidden layer semantics represented by the word vectors can capture the bidirectional semantic dependence of the text data to be classified and capture more robust feature information.
Using two Bi-LSTM learning text word vector representations and the hidden semantics represented by the aspect characteristic word vector to obtain a first hidden layer meaning h represented by the text word vectorSAnd a second hidden layer meaning h represented by the facet feature word vectorT。
Specifically, the text word vector representation is input into a first neural network to obtain a first hidden layer meaning represented by the text word vector
Representing the aspect characteristic word vector and inputting the aspect characteristic word vector into a second neural network to obtain a second hidden layer meaning represented by the aspect characteristic word vector
Specifically, for example, in the sentence s "the restaurant has a good taste of food but the attendant has a poor service attitude", the sentence length m is 13, and the aspect feature word "taste" is the 4 th word in the sentence s, and the length n is 1, then
In one embodiment, step S204 may be implemented by the following code:
S205: and obtaining a first matrix and a second matrix based on the first hidden layer meaning and the second hidden layer meaning.
The first matrix is an attention matrix of text data to be classified to the aspect characteristic words, and the second matrix is an attention matrix of the aspect characteristic words to the text to be classified.
In one embodiment, the first matrix and the second matrix are derived by a softmax function based on the first hidden layer meaning and the second hidden layer meaning.
Specifically, step S205 includes steps S2051 to S2053, please refer to fig. 3, and fig. 3 shows a flowchart of the method of steps S2051 to S2053, where:
step S2051: and averaging the first hidden layer meaning and the second hidden layer meaning respectively to obtain the initial text representation of the text data to be classified and the initial aspect feature word representation of the aspect feature words.
Meaning h to the first hidden layerSAnd a second hidden layer meaning hTRespectively averaging to obtain initial text representation of the text data to be classified
And aspect feature word initial representation of aspect feature word
In one embodiment, step S2051 may be implemented by:
step S2052: and multiplying the text initial representation and the aspect characteristic word initial representation to obtain an initial matrix.
Text is initially represented
And aspect feature word initial representation
The multiplication results in an initial matrix M.
In one embodiment, step S2052 may be implemented by:
step S2053: and respectively basing the rows and the columns of the initial matrix on a softmax function to obtain a first matrix and a second matrix.
And respectively passing the rows and the columns of the initial matrix M through a softmax function to obtain two attention matrixes which are respectively a first matrix alpha corresponding to the rows and a second matrix beta corresponding to the columns.
In one embodiment, step S2053 may be implemented by:
s206: and obtaining the mutual attention of the text data to be classified and the aspect characteristic words according to the first matrix and the second matrix.
And obtaining more concentrated representation according to the first matrix and the second matrix, and obtaining the mutual attention of the text data to be classified and the aspect characteristic words based on the more concentrated representation.
Specifically, the first matrix and the second matrix are averaged respectively to obtain a first representation corresponding to the first matrix and a second representation corresponding to the second matrix.
The first matrix alpha and the second matrix beta are respectively averaged to obtain more concentrated representations, which are respectively a first representation corresponding to the first matrix alpha and a second representation corresponding to the second matrix beta.
Based on the first representation and the second representation, the mutual attention of the text data to be classified and the aspect feature words is obtained.
And multiplying the first representation and the second representation based on the first representation and the second representation to obtain the mutual attention of the text data to be classified and the aspect characteristic words. Therefore, the mutual attention of the text data to be classified and the aspect characteristic words is obtained, the interactive information between the text data to be classified and the aspect characteristic words can be captured better, so that the emotion classification accuracy of the aspect characteristic words in the text data to be classified is improved, namely the emotion classification accuracy of the aspect level text is improved.
In one embodiment, step S206 may be implemented by the following code:
s207: and obtaining the target classification features of the aspect feature words based on the mutual attention.
And obtaining the target classification characteristics of the aspect characteristic words according to the initial text representation and the mutual attention of the text to be classified. Specifically, text is initially represented
And multiplying the target classification characteristics gamma of the aspect characteristic words by the mutual attention.
In one embodiment, step S207 may be implemented by the following code:
s208: and acquiring the predicted emotion polarity of the aspect feature words according to the target classification features of the aspect feature words.
And acquiring the predicted emotion polarity of the aspect characteristic words through a softmax function according to the target classification characteristics of the aspect characteristic words.
Specifically, step S208 includes step S2081 to step S2083, please refer to fig. 4, and fig. 4 shows a flowchart of the method from step S2081 to step S2083, in which:
step S2081: and inputting the target classification characteristics of the aspect characteristic words into a preset function.
Inputting the target classification characteristic gamma of the aspect characteristic word into a preset function x ═ Wl+γ+blWherein W islIs a weight matrix, blIs the bias term.
Step S2082: and acquiring the emotion probability of each type of emotion polarity corresponding to the aspect feature words based on the softmax function.
And inputting the obtained x into a softmax function to calculate the probability that the aspect feature words have each type of emotion polarity. Where emotional polarity includes "positive", "negative", and in some embodiments, may also include "neutral".
Specifically, if the emotion polarity comprises three labels, c is formed by { P, N, O }, (P represents Positive, N represents Negative, and O represents Neutral), and based on the softmax function, the aspect feature words have emotion polarities of P, N and O in the text data to be classifiedProbability P of ccCan be expressed as
Thus, positive emotion probabilities P having emotion polarities "positive" can be calculated PAnd a negative emotion probability P having an emotion polarity of "negativeNAnd a neutral emotion probability P having an emotion polarity of "neutralO。
Step S2083: and determining the emotion polarity with the highest emotion probability as the predicted emotion polarity of the aspect characteristic word.
And the aspect feature words correspond to the emotion probability of each type of emotion polarity, and the emotion polarity with the highest emotion probability is determined as the predicted emotion polarity of the aspect feature words.
S209: and carrying out emotion classification on the text data to be classified according to the predicted emotion polarity so as to obtain and display an emotion classification result corresponding to the aspect feature words.
In this embodiment, the detailed description of step S209 can refer to step S107 in the foregoing embodiment, and is not repeated herein.
The following takes fig. 5 as an example to explain the text data processing method provided in this embodiment:
as shown in fig. 5, fig. 5 is a schematic flowchart illustrating a text data processing method according to an embodiment of the present application. Specifically, the text data to be classified is exemplified as a sentence.
Given a sentence m in length s, s ═ w1；w2；…；wi；…；wm]And an aspect feature word t, t ═ w of length ni；wi+1；…；wi+n+1]Where w represents the words of a sentence, w iThe ith word in the sentence is characterized. Mapping S and t into a word vector matrix generated by a GloVe model to respectively obtain text word vector representation S of a sentence, wherein S is [ v ═ v [ [ v ]1；v2；…；vi；…；vm]And aspect feature word vector representation of aspect feature words T, T ═ vi；vi+1；…；vi+n+1]。
The text word vector representation S is input into Bi-LSTM to obtain the first hidden layer meaning represented by the text word vector
Inputting the aspect characteristic word vector representation T into a second neural network to obtain a second hidden layer meaning represented by the aspect characteristic word vector
Meaning h to the first hidden layerSAnd a second hidden layer meaning hTRespectively averaging to obtain initial text representation of sentence
And aspect feature word initial representation of aspect feature word
Then, the text is initially represented
And aspect feature word initial representation
Multiplying to obtain an initial matrix M, and respectively passing rows and columns of the initial matrix M through a softmax function to obtain two attention matrixes which are respectively a first matrix alpha corresponding to the rows and a second matrix beta corresponding to the columns. The first matrix alpha and the second matrix beta are respectively averaged to obtain more concentrated representations, which are respectively a first representation corresponding to the first matrix alpha and a second representation corresponding to the second matrix beta. Multiplying the first representation and the second representation based on the first representation and the second representation to obtain mutual attention of sentences and aspect characteristic words, and initially representing the text 
And multiplying the target classification characteristics gamma of the aspect characteristic words by the mutual attention.
Of aspect feature wordsTarget classification characteristic gamma input preset function x ═ Wl+γ+blWherein, WlIs a weight matrix, blIs the bias term. And inputting the obtained x into a softmax function to calculate the probability that the aspect characteristic words have a certain type of emotion polarity. The probability P that the aspect characteristic word has emotion polarity of c in the text data to be classifiedcCan be expressed as
And finally, determining the emotion polarity with the highest emotion probability as the predicted emotion polarity of the aspect feature words, and classifying the emotions of the aspects of the sentences according to the predicted emotion polarity.
In the method, the emotion of one aspect feature word in the sentence is classified, and according to the steps, the emotion classification can be performed on other aspect feature words in the sentence, and the predicted emotion polarity of each aspect feature word in the sentence and the emotion classification result corresponding to the aspect feature word are finally obtained.
Furthermore, the obtained emotion classification result corresponding to each aspect feature word can be displayed, so that a user can obtain the emotion classification result of the sentence based on the aspect level and obtain a more fine-grained and accurate classification result.
Thus, the embodiment of emotion classification based on aspect level can overcome the limitation of emotion classification at document level and sentence level, and capture the bidirectional semantic dependence of sentences by using a bidirectional long-short term memory network (Bi-LSTM); and a new attention mechanism model based on the neural network is provided, so that the interactive information of the aspect feature words and the sentences can be better acquired and fully utilized, the accuracy of the classification model is improved, and the problem that the classification of the emotion polarity of the sentences and the emotion polarity of the feature words in different aspects is inaccurate due to the existence of a plurality of aspect feature words in one sentence is solved.
It should be noted that, for the parts not described in detail in the above steps, reference may be made to the foregoing embodiments, and details are not described herein again.
The text data processing method provided by the embodiment captures the bidirectional semantic dependence of the text data to be classified by using a bidirectional long-short term memory network (Bi-LSTM) on the basis of the previous embodiment; and a new attention mechanism model based on a neural network is provided, the mutual attention of the aspect characteristic words and the text data to be classified is obtained, the interactive information of the aspect characteristic words and the text data to be classified can be better obtained and fully utilized, the text emotion classification accuracy is improved, and the problem that the emotion polarity of a sentence and the emotion polarity of different aspect characteristic words are inaccurate due to the existence of a plurality of aspect characteristic words in the sentence is solved.
Referring to fig. 6, which shows a block diagram of a text data processing apparatus according to an embodiment of the present application, the text data processing apparatus 600 may include: text acquisition module 610, text extraction module 620, representation acquisition module 630, web learning module 640, target classification module 650, emotion prediction module 660, and emotion classification module 670.
A text obtaining module 610, configured to obtain text data to be classified;
a text extraction module 620, configured to extract aspect feature words from the text data to be classified;
a representation obtaining module 630, configured to obtain a text word vector representation of the text data to be classified and an aspect feature word vector representation of the aspect feature word;
a network learning module 640, configured to input the text word vector representation and the aspect feature word vector representation into a target neural network respectively, so as to obtain a first hidden layer meaning represented by the text word vector and a second hidden layer meaning represented by the aspect feature word vector, where the target neural network is pre-trained, and is configured to output a hidden layer meaning represented by the word vector according to the input word vector representation;
a target classification module 650, configured to obtain a target classification feature of the aspect feature word based on the first hidden layer meaning and the second hidden layer meaning;
the emotion prediction module 660 is configured to obtain a predicted emotion polarity of the aspect feature word according to the target classification feature of the aspect feature word;
and the emotion classification module 670 is configured to perform emotion classification on the text data to be classified according to the predicted emotion polarity, so as to obtain and display an emotion classification result corresponding to the aspect feature word.
Further, the target neural network includes a bidirectional long-short term memory network.
Further, the target neural network includes a first neural network and a second neural network, and the network learning module 640 includes: a first learning unit and a second learning unit, wherein:
the first learning unit is used for inputting the text word vector representation into the first neural network to obtain a first hidden layer meaning represented by the text word vector;
and the second learning unit is used for inputting the aspect characteristic word vector representation into the second neural network to obtain a second hidden layer meaning represented by the aspect characteristic word vector.
Further, the object classification module 650 includes: a matrix acquisition unit, an attention acquisition unit, and a feature acquisition unit, wherein:
a matrix obtaining unit, configured to obtain a first matrix and a second matrix based on the first hidden layer meaning and the second hidden layer meaning, where the first matrix is an attention matrix of the text data to be classified to the aspect feature word, and the second matrix is an attention matrix of the aspect feature word to the text data to be classified;
the attention obtaining unit is used for obtaining the mutual attention of the text data to be classified and the aspect feature words according to the first matrix and the second matrix;
And the feature acquisition unit is used for obtaining the target classification features of the aspect feature words based on the mutual attention.
Further, the matrix acquisition unit includes: an initial representation subunit, an initial matrix subunit, and a matrix acquisition subunit, wherein:
the initial representation subunit is configured to respectively average the first hidden layer meaning and the second hidden layer meaning to obtain a text initial representation of the text data to be classified and an aspect feature word initial representation of the aspect feature word;
the initial matrix subunit is used for multiplying the text initial representation and the aspect characteristic word initial representation to obtain an initial matrix;
and the matrix acquisition subunit is used for respectively basing the rows and the columns of the initial matrix on a softmax function to obtain the first matrix and the second matrix.
Further, the attention acquiring unit includes: an averaging subunit and an attention subunit, wherein:
the averaging subunit is configured to average the first matrix and the second matrix respectively to obtain a first representation corresponding to the first matrix and a second representation corresponding to the second matrix;
the attention subunit is used for obtaining the mutual attention of the text data to be classified and the aspect feature words based on the first representation and the second representation;
Further, the feature acquisition unit includes: a target subunit, wherein:
and the target subunit is used for obtaining the target classification characteristics of the aspect characteristic words according to the text initial representation of the text data to be classified and the mutual attention.
Further, the emotion prediction module 660 includes: the device comprises a feature input unit, a probability acquisition unit and a polarity determination unit, wherein:
the characteristic input unit is used for inputting the target classification characteristics of the aspect characteristic words into a preset function;
the probability obtaining unit is used for obtaining the emotion probability of each type of emotion polarity corresponding to the aspect feature words based on a softmax function;
and the polarity determining unit is used for determining the emotion polarity with the highest emotion probability as the predicted emotion polarity of the aspect feature word.
The text data processing device provided in the embodiment of the present application is used to implement the corresponding text data processing method in the foregoing method embodiment, and has the beneficial effects of the corresponding method embodiment, which are not described herein again.
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described apparatuses and modules may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the several embodiments provided in the present application, the coupling between the modules may be electrical, mechanical or other type of coupling.
In addition, functional modules in the embodiments of the present application may be integrated into one processing module, or each of the modules may exist alone physically, or two or more modules are integrated into one module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode.
Referring to fig. 7, a block diagram of an electronic device according to an embodiment of the present application is shown. The electronic device 700 may be a smartphone, a tablet computer, a computer, an electronic book, a wearable electronic device, a server, or other electronic device capable of running an application. The electronic device 700 in the present application may include one or more of the following components: a processor 710, a memory 720, and one or more applications, wherein the one or more applications may be stored in the memory 720 and configured to be executed by the one or more processors 710, the one or more programs configured to perform a method as described in the aforementioned method embodiments.
The Memory 720 may include a Random Access Memory (RAM) or a Read-Only Memory (Read-Only Memory). The memory 720 may be used to store instructions, programs, code sets, or instruction sets. The memory 720 may include a stored program area and a stored data area, wherein the stored program area may store instructions for implementing an operating system, instructions for implementing at least one function (such as a touch function, a sound playing function, an image playing function, etc.), instructions for implementing various method embodiments described below, and the like. The data storage area may also store data created by the electronic device 700 during use (e.g., phone books, audio-visual data, chat log data), and the like.
Then if each unit in the text data processing apparatus shown in fig. 6 is used as a function module such as a package, each unit in the text data processing apparatus is stored in the memory 720, can be called by the processor, and executes the corresponding function.
Referring to fig. 8, a block diagram of a computer-readable storage medium according to an embodiment of the present disclosure is shown. The computer-readable storage medium 800 has stored therein a program code 810, the program code 810 being capable of being invoked by a processor to perform the method described in the above method embodiments.
The computer-readable storage medium 800 may be an electronic memory such as a flash memory, an EEPROM (electrically erasable programmable read only memory), an EPROM, a hard disk, or a ROM. Alternatively, the computer-readable storage medium 800 includes a non-volatile computer-readable medium (non-transitory computer-readable storage medium). The computer readable storage medium 800 has storage space for program code 810 for performing any of the method steps described above. The program code can be read from or written to one or more computer program products. The program code 810 may be compressed, for example, in a suitable form.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the same; although the present application 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; such modifications and substitutions do not necessarily depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.
Claims (9)
1. A method of processing text data, the method comprising:
acquiring text data to be classified;
extracting aspect feature words from the text data to be classified;
acquiring text word vector representation of the text data to be classified and aspect feature word vector representation of the aspect feature words;
inputting the text word vector representation and the aspect feature word vector representation into a target neural network respectively to obtain a first hidden layer meaning represented by the text word vector and a second hidden layer meaning represented by the aspect feature word vector, wherein the target neural network is trained in advance and used for outputting the hidden layer meaning represented by the word vector according to the input word vector representation;
obtaining a first matrix and a second matrix based on the first hidden layer meaning and the second hidden layer meaning, wherein the first matrix is an attention matrix of the text data to be classified to the aspect feature words, and the second matrix is an attention matrix of the aspect feature words to the text data to be classified;
obtaining the mutual attention of the text data to be classified and the aspect feature words according to the first matrix and the second matrix;
Obtaining target classification features of the aspect feature words based on the mutual attention;
obtaining the predicted emotion polarity of the aspect feature words according to the target classification features of the aspect feature words;
and carrying out emotion classification on the text data to be classified according to the predicted emotion polarity so as to obtain and display an emotion classification result corresponding to the aspect feature words.
2. The method of claim 1, wherein the target neural network comprises a bidirectional long-short term memory network.
3. The method of claim 1 or 2, wherein the target neural network comprises a first neural network and a second neural network, and wherein inputting the text word vector representation and the aspect feature word vector representation into the target neural network, respectively, results in a first hidden layer meaning for the text word vector representation and a second hidden layer meaning for the aspect feature word vector representation, comprises:
inputting the text word vector representation into the first neural network to obtain a first hidden layer meaning represented by the text word vector;
and inputting the aspect feature word vector representation into the second neural network to obtain a second hidden layer meaning represented by the aspect feature word vector.
4. The method according to claim 1, wherein the deriving a first matrix and a second matrix based on the first hidden layer meaning and the second hidden layer meaning comprises:
averaging the first hidden layer meaning and the second hidden layer meaning respectively to obtain a text initial representation of the text data to be classified and an aspect feature word initial representation of the aspect feature word;
multiplying the initial text representation and the initial aspect feature word representation to obtain an initial matrix;
and respectively basing the rows and the columns of the initial matrix on a softmax function to obtain the first matrix and the second matrix.
5. The method according to claim 4, wherein the obtaining the mutual attention of the text data to be classified and the aspect feature words according to the first matrix and the second matrix comprises:
respectively averaging the first matrix and the second matrix to obtain a first representation corresponding to the first matrix and a second representation corresponding to the second matrix;
obtaining the mutual attention of the text data to be classified and the aspect feature words based on the first representation and the second representation;
The obtaining of the target classification features of the aspect feature words based on the mutual attention includes:
and obtaining the target classification characteristics of the aspect characteristic words according to the text initial representation of the text data to be classified and the mutual attention.
6. The method of claim 1, wherein the obtaining the predicted emotion polarity of the aspect feature word according to the target classification feature of the aspect feature word comprises:
inputting the target classification characteristics of the aspect characteristic words into a preset function;
acquiring the emotion probability of each type of emotion polarity corresponding to the aspect feature words based on a softmax function;
and determining the emotion polarity with the highest emotion probability as the predicted emotion polarity of the aspect feature words.
7. A text data processing apparatus, characterized in that the apparatus comprises:
the text acquisition module is used for acquiring text data to be classified;
the text extraction module is used for extracting aspect feature words from the text data to be classified;
the representation obtaining module is used for obtaining text word vector representation of the text data to be classified and aspect feature word vector representation of the aspect feature words;
the network learning module is used for inputting the text word vector representation and the aspect characteristic word vector representation into a target neural network respectively to obtain a first hidden layer meaning represented by the text word vector and a second hidden layer meaning represented by the aspect characteristic word vector, wherein the target neural network is trained in advance and is used for outputting the hidden layer meaning represented by the word vector according to the input word vector representation;
A matrix obtaining unit, configured to obtain a first matrix and a second matrix based on the first hidden layer meaning and the second hidden layer meaning, where the first matrix is an attention matrix of the text data to be classified to the aspect feature word, and the second matrix is an attention matrix of the aspect feature word to the text data to be classified;
an attention obtaining unit, configured to obtain mutual attention of the text data to be classified and the aspect feature words according to the first matrix and the second matrix;
the feature acquisition unit is used for obtaining the target classification features of the aspect feature words based on the mutual attention;
the emotion prediction module is used for acquiring the predicted emotion polarity of the aspect feature words according to the target classification features of the aspect feature words;
and the emotion classification module is used for carrying out emotion classification on the text data to be classified according to the predicted emotion polarity so as to obtain and display an emotion classification result corresponding to the aspect feature words.
8. An electronic device, comprising:
one or more processors;
a memory;
one or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications configured to perform the method of any of claims 1-6.
9. A computer-readable storage medium, having stored thereon program code that can be invoked by a processor to perform the method according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910873641.0A CN110781273B (en) | 2019-09-17 | 2019-09-17 | Text data processing method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910873641.0A CN110781273B (en) | 2019-09-17 | 2019-09-17 | Text data processing method and device, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110781273A CN110781273A (en) | 2020-02-11 |
| CN110781273B true CN110781273B (en) | 2022-05-31 |
Family
ID=69384171
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910873641.0A Active CN110781273B (en) | 2019-09-17 | 2019-09-17 | Text data processing method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110781273B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111680159B (en) * | 2020-06-11 | 2023-08-29 | 华东交通大学 | Data processing method and device and electronic equipment |
| CN111813937A (en) * | 2020-07-07 | 2020-10-23 | 新华智云科技有限公司 | Positive energy news classification method based on positive energy dictionary |
| CN112597299A (en) * | 2020-12-07 | 2021-04-02 | 深圳价值在线信息科技股份有限公司 | Text entity classification method and device, terminal equipment and storage medium |
| CN112784583A (en) * | 2021-01-26 | 2021-05-11 | 浙江香侬慧语科技有限责任公司 | Multi-angle emotion analysis method, system, storage medium and equipment |
| CN113361617A (en) * | 2021-06-15 | 2021-09-07 | 西南交通大学 | Aspect level emotion analysis modeling method based on multivariate attention correction |
| CN113627550A (en) * | 2021-08-17 | 2021-11-09 | 北京计算机技术及应用研究所 | Image-text emotion analysis method based on multi-mode fusion |
| CN114817526B (en) * | 2022-02-21 | 2024-03-29 | 华院计算技术(上海)股份有限公司 | Text classification method and device, storage medium and terminal |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108984724A (en) * | 2018-07-10 | 2018-12-11 | 凯尔博特信息科技（昆山）有限公司 | It indicates to improve particular community emotional semantic classification accuracy rate method using higher-dimension |
| CN109597997A (en) * | 2018-12-07 | 2019-04-09 | 上海宏原信息科技有限公司 | Based on comment entity, aspect grade sensibility classification method and device and its model training |
| CN109948165A (en) * | 2019-04-24 | 2019-06-28 | 吉林大学 | Fine granularity feeling polarities prediction technique based on mixing attention network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10019438B2 (en) * | 2016-03-18 | 2018-07-10 | International Business Machines Corporation | External word embedding neural network language models |
-
2019
- 2019-09-17 CN CN201910873641.0A patent/CN110781273B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108984724A (en) * | 2018-07-10 | 2018-12-11 | 凯尔博特信息科技（昆山）有限公司 | It indicates to improve particular community emotional semantic classification accuracy rate method using higher-dimension |
| CN109597997A (en) * | 2018-12-07 | 2019-04-09 | 上海宏原信息科技有限公司 | Based on comment entity, aspect grade sensibility classification method and device and its model training |
| CN109948165A (en) * | 2019-04-24 | 2019-06-28 | 吉林大学 | Fine granularity feeling polarities prediction technique based on mixing attention network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110781273A (en) | 2020-02-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110781273B (en) | Text data processing method and device, electronic equipment and storage medium | |
| CN111680159B (en) | Data processing method and device and electronic equipment | |
| WO2020125445A1 (en) | Classification model training method, classification method, device and medium | |
| US11238310B2 (en) | Training data acquisition method and device, server and storage medium | |
| CN108984530A (en) | A kind of detection method and detection system of network sensitive content | |
| CN110377740B (en) | Emotion polarity analysis method and device, electronic equipment and storage medium | |
| CN107807968B (en) | Question answering device and method based on Bayesian network and storage medium | |
| CN106778878B (en) | Character relation classification method and device | |
| WO2021208727A1 (en) | Text error detection method and apparatus based on artificial intelligence, and computer device | |
| CN112015721A (en) | E-commerce platform storage database optimization method based on big data | |
| WO2023108993A1 (en) | Product recommendation method, apparatus and device based on deep clustering algorithm, and medium | |
| CN113407677B (en) | Method, apparatus, device and storage medium for evaluating consultation dialogue quality | |
| CN113094478B (en) | Expression reply method, device, equipment and storage medium | |
| CN114676704A (en) | Sentence emotion analysis method, device and equipment and storage medium | |
| CN111753082A (en) | Text classification method and device based on comment data, equipment and medium | |
| CN111930792A (en) | Data resource labeling method and device, storage medium and electronic equipment | |
| CN110955750A (en) | Combined identification method and device for comment area and emotion polarity, and electronic equipment | |
| CN111538841B (en) | Comment emotion analysis method, device and system based on knowledge mutual distillation | |
| CN115099239B (en) | Resource identification method, device, equipment and storage medium | |
| CN112667816A (en) | Deep learning-based aspect level emotion analysis method and system | |
| CN107797981B (en) | Target text recognition method and device | |
| CN115309864A (en) | Intelligent sentiment classification method and device for comment text, electronic equipment and medium | |
| CN115269768A (en) | Element text processing method and device, electronic equipment and storage medium | |
| CN113626576A (en) | Method and device for extracting relational characteristics in remote supervision, terminal and storage medium | |
| CN115906861B (en) | Sentence emotion analysis method and device based on interaction aspect information fusion |
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 |