WO2021169364A1

WO2021169364A1 - Semantic emotion analysis method and apparatus, device, and storage medium

Info

Publication number: WO2021169364A1
Application number: PCT/CN2020/125154
Authority: WO
Inventors: 邓悦; 郑立颖; 徐亮
Original assignee: 平安科技（深圳）有限公司
Priority date: 2020-09-23
Filing date: 2020-10-30
Publication date: 2021-09-02
Also published as: CN112131888A; CN112131888B

Abstract

Provided are a semantic emotion analysis method and apparatus, a computer device, and a computer-readable storage medium, relating to the field of intelligent decision making in artificial intelligence. The semantic emotion analysis method comprises: acquiring an importance metric value corresponding to each word in a sentence to be analyzed (S1); according to the importance metric value corresponding to each word in the sentence to be analyzed, obtaining, by means of two recurrent neural network models which run in parallel, an implicit expression corresponding to the sentence to be analyzed (S2); inputting the implicit expression corresponding to the sentence to be analyzed and a sentence label corresponding to the sentence to be analyzed into a semantic sentiment analysis classifier (S3); and receiving an emotion analysis classification result of the semantic emotion analysis classifier for the sentence to be analyzed (S4). A self-attention mechanism is introduced, the importance of each word in a sentence is quantified by means of an importance metric value, and the meaning of the current word in the whole sentence is then acquired according to the position of an important word, so that each word in the sentence and the importance metric value corresponding thereto are fused together in the hidden state of the finally output whole sentence.

Description

Method, device, equipment and storage medium for analyzing semantic emotion

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on September 23, 2020, the application number is 2020110099004, and the invention title is "Methods, Apparatus, Equipment, and Storage Media for Analyzing Semantic Emotions", the entire content of which is incorporated by reference Incorporated in this application.

Technical field

This application relates to the field of intelligent decision-making in artificial intelligence, in particular to methods, devices, equipment and storage media for analyzing semantic emotions.

Background technique

The earliest word-building model adopted is the bag-of-words model. The bag-of-words model treats a sentence as a simple collection of words and combines them into a complete sentence through simple vector operations. With the development of deep learning, the application of neural networks is becoming more and more popular. As a sequential model, the cyclic neural network RNN treats text as a sequence of words, which can effectively capture the relationship between time series variables, but the inventor is aware of the sequential model The grammatical relationship in the sentence structure cannot be distinguished, and the importance of each word in the sentence cannot be determined. It is not conducive to understanding the key points of the entire sentence. Therefore, it cannot recognize the difference in sentence meaning caused by the change of the semantic role of the word or phrase, resulting in failure The non-linear dependence between common words in natural language is captured, and the emotional characteristics of sentences are obtained.

technical problem

The main purpose of this application is to provide analysis of semantic emotions, aiming to solve the technical problem of not being able to capture the non-linear dependence between common words in natural language and obtaining the emotional features of sentences.

Technical solutions

This application proposes a method for analyzing semantic emotions, including:

Obtain the importance metric value corresponding to each word in the sentence to be analyzed;

According to the importance metric value corresponding to each word in the sentence to be analyzed, the implicit expression corresponding to the sentence to be analyzed is obtained through two cyclic neural network models running in parallel, wherein the implicit expression Incorporating the semantic dependency of context;

Input the implicit expression corresponding to the sentence to be analyzed and the preset sentence label corresponding to the sentence to be analyzed into the semantic sentiment analysis classifier;

Receiving the sentiment analysis classification result of the sentence to be analyzed by the semantic sentiment analysis classifier.

This application also provides a device for analyzing semantic emotions, including:

The acquiring module is used to acquire the importance metric value corresponding to each word in the sentence to be analyzed;

The obtaining module is used to obtain the implicit expression corresponding to the sentence to be analyzed through two parallel running cyclic neural network models according to the importance metric value corresponding to each word in the sentence to be analyzed. Implicit expressions incorporate the semantic dependency of context;

An input module for inputting the implicit expression corresponding to the sentence to be analyzed and the preset sentence label corresponding to the sentence to be analyzed into the semantic sentiment analysis classifier;

The receiving module is configured to receive the sentiment analysis classification result of the sentence to be analyzed by the semantic sentiment analysis classifier.

The present application also provides a computer device, including a memory and a processor, the memory stores a computer program, and the processor implements the steps of the foregoing method when the computer program is executed.

The present application also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned method are realized.

Beneficial effect

This application introduces a self-attention mechanism to quantify the importance of each word in a sentence through the importance measurement value, and then according to the position of the important word, through the RNN improved by the hierarchical traversal of the tree, it can not only obtain The word meaning corresponding to the current word, and according to two parallel running cyclic neural network models, the implicit expression of the sentence to be analyzed based on the current word is obtained, and the semantic dependency of the context is merged to get the meaning of the current word in the entire sentence. As a result, the hidden state of the entire sentence in the final output is integrated with each word in the sentence and its corresponding importance measurement, so that the emotional tendency of the sentence meaning is more obvious, and the sentence expression is more accurate.

Description of the drawings

Fig. 1 is a schematic flowchart of a method for analyzing semantic emotions according to an embodiment of the present application;

Fig. 2 is a schematic diagram of convolution calculation using two RNNs according to an embodiment of the present application;

Fig. 3 is a schematic diagram of convolution calculation using two RNNs according to an embodiment of the present application;

FIG. 4 is a schematic diagram of a system flow diagram for analyzing semantic emotions according to an embodiment of the present application;

Fig. 5 is a schematic diagram of the internal structure of a computer device according to an embodiment of the present application.

The best embodiment of the present invention

In order to make the purpose, technical solutions, and advantages of this application clearer and clearer, the following further describes the application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

1, a method for analyzing semantic emotions according to an embodiment of the present application includes:

S1: Obtain the importance metric value corresponding to each word in the sentence to be analyzed;

S2: According to the importance metric value corresponding to each word in the sentence to be analyzed, the implicit expression corresponding to the sentence to be analyzed is obtained through two parallel running cyclic neural network models, where the implicit The expression incorporates the semantic dependency of the context;

S3: Input the implicit expression corresponding to the sentence to be analyzed and the preset sentence label corresponding to the sentence to be analyzed into the semantic sentiment analysis classifier;

S4: Receive the sentiment analysis classification result of the sentence to be analyzed by the semantic sentiment analysis classifier.

The importance metric of this application is calculated by introducing a self-attention mechanism, by quantifying the importance of each word in the sentence to be analyzed, and iteratively calculated by two cyclic neural network models running in parallel, so that the final output is The hidden state of the sentence to be analyzed combines the semantics of each word in the sentence to be analyzed and its corresponding importance measure, the semantic dependency of the context, and each word in the sentence and its corresponding importance measure. The emotional tendency of the sentence meaning is more obvious, the sentence expression is more precise, and the expression ability is greatly improved. The above sentence tag distinguishes the positional relationship of each sentence in the text, including but not limited to the first sentence, or the first sentence of the paragraph, and so on. Then, by inputting the implicit expression representing the hidden state of the entire sentence into a semantic sentiment analysis classifier with pre-trained parameters, sentiment classification analysis is performed. Emotion classification includes positive emotion and negative emotion. Through the emotional analysis of the sentence, the understanding of the speaker’s mentality is improved, and the purpose of identifying the personality characteristics of the speaker more accurately is achieved.

Further, according to the importance metric value corresponding to each word in the sentence to be analyzed, the step S2 of obtaining the implicit expression corresponding to the sentence to be analyzed through two parallel running cyclic neural network models includes:

S21: Split the sentence to be analyzed into a tree structure according to the importance metric corresponding to each word in the sentence to be analyzed, where the tree structure includes leaf nodes, child nodes, and root nodes;

S22: Input the clauses contained in the first leaf node into the first recurrent neural network according to the forward order of the sentence to be analyzed, and input the clauses contained in the second leaf node into the first recurrent neural network according to the reverse order of the sentence to be analysed. A two-loop neural network, wherein the first leaf node and the second leaf node are a pair of leaf nodes that belong to any designated child node;

S24: Multiply the forward hidden vector output by the first recurrent neural network by the reverse hidden vector output by the second recurrent neural network to obtain the vector expression of the designated child node;

S25: According to the vector expression of the designated child node, according to the tree structure, sequentially recursively calculate the vector expression to the root node corresponding to the first word, wherein the first word is the corresponding one when the importance metric value is the largest word;

S26: Use the vector expression of the root node corresponding to the first word as an implicit expression of the sentence to be analyzed.

This application realizes sentence segmentation and splitting of the sentence to be analyzed through the importance measurement value, and splits the sentence to be analyzed into an inverted tree structure. Then through the two recurrent neural networks to perform convolution operations from the forward and reverse directions, and then multiply the output results of the two recurrent neural networks to obtain the vector expression of the specified root node, and then follow the inverted tree structure in turn , All child nodes and leaf nodes recursively to the tree structure participate in the operation until the vector expression of the first word with the largest importance metric value is obtained, that is, the vector expression corresponding to the total root node of the inverted tree structure is obtained as the waiting Analyze the implicit expression of the statement.

The root node of the inverted tree structure can also be called a parent node, and the parent node is defined as the original vector corresponding to the entire sentence. The parent node is down to the left and right child nodes. Set the two clauses obtained by dividing the whole sentence according to the specified words as the left and right subtrees of the parent node. Then regard the left and right subtrees as a sequence, and use RNN to encode the sequence. Separate the child nodes of the left subtree from the child nodes of the right subtree, and use two RNNs for convolution calculation: the first RNN encodes the sequence of left child nodes from front to back according to the order of the entire sentence, and the second Each RNN encodes the sequence of right child nodes from back to front according to the order of the whole sentence. The final output of each RNN is the vector representation corresponding to the specified word of the split left and right subtrees. The specified word is the current child node. The vector representation of the current child node is composed of the hidden state of the RNN model on the left and the hidden state of the RNN model on the right. decided together. _{Take the word corresponding to the vector v i that} ranks first in the importance metric as the designated word, _{and divide the original sentence [v 1} ,v ₂ ,...,v _n ], and the sentence on the left after the division is the root node The left subtree child node, the sentence on the right is the right subtree child node of the root node. Thus the left subtree of child nodes comprises a _{_{[v 1, v 2, ...}} , v i], the right subtree of child nodes comprises a _{_{[v i, v i + 1}} , ..., v n], such as the schematic figures 2 Shown. Thus, for non-leaf nodes, use the following formula to calculate the child node re-expression of the vector v: i.e. first calculate _{_{_{f i = RNN F (v 1}}} , v 2, ..., v i); and b _i = RNN _B ( v _i ,v _i+1 ,...,v _n ); then the vector expression v is obtained _{by v=f i} ·b _i. The above RNN _F represents the forward-propagating RNN, RNN _B represents the reverse-propagating RNN, f _i is the hidden vector expression obtained by the forward-propagating RNN, and b _i is the hidden vector expression obtained by the reverse-propagating RNN, and finally f _i and b _{Do the} dot product of i to get a word expression v that incorporates context. Then, the left and right child nodes of the child node are respectively regarded as the next-level child nodes, and the recursive loop is performed in the above-mentioned manner until the sentence is segmented and split to the leaf node to stop. The above recursive calculation recursively from the leaf node to the root node corresponding to the first word, and output the vector expression corresponding to the first word as the implicit expression of the hidden state of the entire sentence.

Further, the step S21 of splitting the sentence to be analyzed into a tree structure according to the importance metric value corresponding to each word in the sentence to be analyzed includes:

S211: Determine the first word with the largest importance measurement value in the sentence to be analyzed according to the importance measurement value corresponding to each word in the sentence to be analyzed;

S212: Using the first word as a dividing point, split the sentence to be analyzed into a first clause and a second clause, wherein the first word is used as the root node of the tree structure;

S213: Using the second word with the largest importance measure in the first clause as a demarcation point, split the first clause into a third clause and a fourth clause, and use the second clause The third word with the largest importance measure is the demarcation point. The second clause is split into a fifth clause and a sixth clause, where the second word and the third word are all The child nodes of the root node;

S214: According to the splitting process of the first clause and the second clause, split the sentence to be analyzed into leaf nodes to form a tree structure composed of multiple layers of nodes, where the leaf nodes are nodes without child nodes.

In this application, the importance metric value corresponding to each word in the sentence to be analyzed is obtained at one time through the calculation method of the importance metric value. Then the first word with the largest importance measure is taken as the parent node of the tree structure, and then the word with the largest importance measure in the two clauses of the sentence to be analyzed is taken as the child node of the parent node, and then continue according to the importance The metric value, the word with the largest importance metric value in the sub-clause corresponding to the clause is used as the next-level child node of the above-mentioned child node until it is split to the leaf node.

Further, the step S1 of obtaining the importance metric value corresponding to each word in the sentence to be analyzed includes:

S11: Perform word embedding and position coding on the sentence to be analyzed to obtain the vector expression corresponding to each word in the sentence to be analyzed;

S12: The vector expression corresponding to each word in the sentence to be analyzed is sequentially input into the self-attention network according to the order in the sentence to be analyzed, and the important corresponding to each word in the sentence to be analyzed is obtained. Metric.

This application introduces a self-attention mechanism through the vector of word embedding and position encoding, and quantifies the importance of each word in the sentence, so that the hidden state of the final output is integrated with each word in the sentence and its corresponding importance The measurement value greatly improves the expressive ability of the model, so the subsequent sentiment classification results of the candidate's answer will be more accurate.

Further, the step S11 of performing word embedding and position coding on the sentence to be analyzed to obtain the vector expression corresponding to each word in the sentence to be analyzed includes:

S111: Calculate the first vector of the specified word after word embedding according to the first calculation formula, and calculate the second vector of the position code corresponding to the specified word according to the second formula;

S112: According to the first vector and the second vector, a vector expression corresponding to the designated word is calculated by a third calculation formula;

S113: According to the calculation process of the vector expression corresponding to the designated word, calculate the vector expression corresponding to each word in the sentence to be analyzed.

The first calculation formula of this application is: w _i = Embedding(x), and the second calculation formula is:

The third formula is _{^{v i = g (W v ·}} (w i · p i) + b v). For example, the sentence to be analyzed is a sentence containing n words, expressed as X=[x ₁ ,x ₂ ,...,x _n ], assuming that the vector of the i-th word is denoted as v _i , and v _i is defined as It combines the word embedding vector of the i-th word and the position-encoded vector. W _i is the i th word in the word vectors after embedding, said embedding word sentence word for one hot encoding may be set in advance as a vector dimension 512 dimension. p _i is the vector of the position code of the i-th word, W is the weight matrix, b is the bias parameter, and g is the activation function. The vectors w _i and p _i are multiplied by the linear transformation and nonlinear activation function g, and restored to 512 dimensions, and the vector expressions corresponding to the i-th word are obtained [v ₁ ,v ₂ ,...,v _n ].

Further, the vector expression corresponding to each word in the sentence to be analyzed is sequentially input into the self-attention network according to the order in the sentence to be analyzed, and the corresponding vector expression of each word in the sentence to be analyzed is obtained. Step S12 of the importance measure includes:

S121: The vector expression corresponding to each word in the sentence to be analyzed is sequentially input into the self-attention network according to the order in the sentence to be analyzed;

S122: By calling the fourth calculation formula in the self-attention network, respectively calculate the importance metric value corresponding to each word in the sentence to be analyzed.

The fourth calculation formula of this application is as follows: soft max(v ^T *v/d _k )*v, the importance metric vector is calculated as the importance metric, so that each word in the sentence presents a different importance metric, Can carry out gradient split sentence.

Further, before step S1 of obtaining the importance metric value corresponding to each word in the sentence to be analyzed, the method includes:

S101: Load the preset classification function to the classifier, and initialize the assignment;

S102: Input the vector expression and sentence label of the training sentence into the classifier loaded with the preset classification function for classification training;

S103: Determine whether the loss function converges, where the loss function is the cross entropy of the predicted classification result and the true classification result;

S104: If yes, determine that the semantic sentiment analysis classifier has been obtained through training.

In this application, the obtained v can be used as the vector representation of the entire sentence, and it is _{input into the classifier together with the label yi} for prediction. The function of the classifier is as follows:

Among them, W and b are the parameters of the classifier. After initializing the assignment randomly, according to the predicted result

The cross entropy of the real label y is used as a loss function to continuously modify the parameters W and b of the classifier.

After the training is completed, for the newly input unlabeled candidate answer text X`, the hidden state v′ can be obtained after the previous processing and calculation, and the prediction using the classifier is as follows:

The schematic flow chart of the method for analyzing semantic emotions of the present application is shown in FIG. 3. The semantic sentiment analysis of this application can quickly judge certain personality characteristics of the speaker based on the speaker's answer, and give necessary and reasonable follow-up questions. By capturing the expression of the context, the understanding of the entire sentence answered by the speaker is improved, and the interview experience for both the interviewer and the candidate is improved. At the same time, the response speed of the hardware has also been improved, so not only the storage space of the computer is saved, but also the running speed of the software is improved.

Referring to Fig. 4, an apparatus for analyzing semantic emotions according to an embodiment of the present application includes:

The obtaining module 1 is used to obtain the importance metric value corresponding to each word in the sentence to be analyzed;

The obtaining module 2 is used to obtain the implicit expression corresponding to the sentence to be analyzed through two parallel running cyclic neural network models according to the importance metric value corresponding to each word in the sentence to be analyzed, wherein, The implicit expression integrates the semantic dependency of the context;

The input module 3 is used to input the implicit expression corresponding to the sentence to be analyzed and the preset sentence label corresponding to the sentence to be analyzed into the semantic sentiment analysis classifier;

The receiving module 4 is configured to receive the sentiment analysis classification result of the sentence to be analyzed by the semantic sentiment analysis classifier.

Further, module 2 is obtained, including:

The splitting unit is configured to split the sentence to be analyzed into a tree structure according to the importance metric value corresponding to each word in the sentence to be analyzed, wherein the tree structure includes leaf nodes and child nodes And the root node;

The input unit is used to input the clauses contained in the first leaf node into the first recurrent neural network according to the forward ordering of the sentences to be analyzed, and the clauses contained in the second leaf node according to the reverse of the sentences to be analyzed Sort and input the second recurrent neural network, wherein the first leaf node and the second leaf node are a pair of leaf nodes that belong to any designated child node;

The first obtaining unit is configured to multiply the forward hidden vector output by the first recurrent neural network by the reverse hidden vector output by the second recurrent neural network to obtain the vector expression of the designated child node;

The calculation unit is configured to recursively calculate the vector expression to the root node corresponding to the first word according to the vector expression process of the designated root node and according to the tree structure, wherein the first word is the importance metric value The word corresponding to the maximum time;

As a unit, it is used to express the vector of the root node corresponding to the first word as an implicit expression of the sentence to be analyzed.

The root node of the inverted tree structure is also called the parent node, and the parent node is defined as the original vector corresponding to the entire sentence. The parent node is down to the left and right child nodes. Set the two clauses obtained by dividing the whole sentence according to the specified words as the left and right subtrees of the parent node. Then regard the left and right subtrees as a sequence, and use RNN to encode the sequence. Separate the child nodes of the left subtree from the child nodes of the right subtree, and use two RNNs for convolution calculation: the first RNN encodes the sequence of left child nodes from front to back according to the order of the entire sentence, and the second Each RNN encodes the sequence of right child nodes from back to front according to the order of the whole sentence. The final output of each RNN is the vector representation corresponding to the specified word of the split left and right subtrees. The specified word is the current child node. The vector representation of the current child node is the hidden state of the RNN model on the left and the hidden state of the RNN model on the right. decided together. Take the importance ranking metric vectors v _i corresponding to the word, as the designated word of the original sentence _{_{[v i, v i + 1}} , ..., v n] is divided, the division of the sentence as the root of the left The left subtree child node of the node, and the sentence on the right as the right subtree child node of the root node. Thus the left subtree of child nodes comprises a _{_{[v 1, v 2, ...}} , v i], the right subtree of child nodes comprises a _{_{[v i, v i + 1}} , ..., v n], such as the schematic figures 2 Shown. Thus, for non-leaf nodes, use the following formula to calculate the child node re-expression of the vector v: i.e. first calculate _{_{_{f i = RNN F (v 1}}} , v 2, ..., v i); and b _i = RNN _B ( v _i ,v _i+1 ,...,v _n ); then the vector expression v is obtained _{by v=f i} ·b _i. The above-mentioned RNN _F represents the forward-propagating RNN, RNN _B represents the reverse-propagating RNN, f _i is the hidden vector expression obtained by the forward-propagating RNN, and b _i is the hidden vector expression obtained by the reverse-propagating RNN, and finally f _i and b _{Do the} dot product of i to get a word expression v that incorporates context. Then, the left and right child nodes of the child node are respectively regarded as the next-level child nodes, and the recursive loop is performed in the above-mentioned manner until the sentence is segmented and split to the leaf node to stop. The above recursive calculation recursively from the leaf node to the root node corresponding to the first word, and output the vector expression corresponding to the first word as the implicit expression of the hidden state of the entire sentence.

Further, the split unit includes:

The determining subunit is used to determine the first word with the largest importance measurement value in the sentence to be analyzed according to the importance measurement value corresponding to each word in the sentence to be analyzed;

The first splitting subunit is used to split the sentence to be analyzed into a first clause and a second clause using the first word as a dividing point, wherein the first word serves as the tree structure The root node;

The second splitting subunit is used for splitting the first clause into a third clause and a fourth clause using the second word with the largest importance measure in the first clause as a demarcation point. The third word with the largest importance measure in the second clause is the demarcation point, and the second clause is split into a fifth clause and a sixth clause, wherein the second word and the The third word is a child node of the root node;

The third splitting subunit is used to split the sentence to be analyzed into leaf nodes according to the splitting process of the first clause and the second clause to form a tree structure composed of multiple layers of nodes, wherein the Leaf nodes are nodes that have no child nodes.

Further, obtaining module 1 includes:

The coding unit is used for word embedding and position coding of the sentence to be analyzed to obtain the vector expression corresponding to each word in the sentence to be analyzed;

The second obtaining unit is used to express the vector corresponding to each word in the sentence to be analyzed, according to the order in the sentence to be analyzed, and sequentially input into the self-attention network to obtain each of the sentences to be analyzed. The importance metric value corresponding to each word.

Further, the coding unit includes:

The first calculation subunit is configured to calculate a first vector of a specified word after word embedding according to a first calculation formula, and calculate a second vector of a position code corresponding to the specified word according to a second formula;

The second calculation subunit is configured to calculate the vector expression corresponding to the designated word through a third calculation formula according to the first vector and the second vector;

The third calculation subunit is used to calculate the vector expression corresponding to each word in the sentence to be analyzed according to the calculation process of the vector expression corresponding to the designated word.

The third formula is _{^{v i = g (W v ·}} (w i · p i) + b v). For example, the sentence to be analyzed is a sentence containing n words, expressed as X=[x ₁ ,x ₂ ,...,x _n ], assuming that the vector of the i-th word is denoted as v _i , and v _i is defined as It combines the word embedding vector of the i-th word and the position-encoded vector. W _i is the i th word in the word vectors after embedding, said embedding word sentence word for one hot encoding may be set in advance as a vector dimension 512 dimension. p _i is the vector of the position code of the i-th word, W is the weight matrix, b is the bias parameter, and g is the activation function. Multiply the vectors w _i and p _i through linear transformation and nonlinear activation function g, and restore them to 512 dimensions to obtain the vector expressions corresponding to the i-th word [v _i ,v _i+1 ,... ,v _n ].

Further, the second obtaining unit includes:

The input subunit is used to express the vector corresponding to each word in the sentence to be analyzed, and sequentially input into the self-attention network according to the order in the sentence to be analyzed;

The fourth calculation subunit is configured to calculate the importance metric value corresponding to each word in the sentence to be analyzed by calling the fourth calculation formula in the self-attention network.

The fourth calculation formula of this application is as follows: soft max(v ^T *v/d _k )*v, the importance metric value vector is calculated as the importance metric value, so that each word in the sentence presents a different importance metric , You can perform gradient split sentences.

Further, devices for analyzing semantic emotions include:

The assignment module is used to load the preset classification function to the classifier and initialize the assignment;

The training module is used to input the vector expression and sentence label of the training sentence into the classifier loaded with the preset classification function for classification training;

A judging module for judging whether the loss function has converged, where the loss function is the cross entropy of the predicted classification result and the true classification result;

The judging module is used for judging that the semantic sentiment analysis classifier is obtained by training if it converges.

The data for analyzing semantic emotions of this application is stored in the blockchain, and based on the advantages of the blockchain, data storage and sharing are realized. Blockchain is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. Blockchain, essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block. The blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.

The underlying platform of the blockchain can include processing modules such as user management, basic services, smart contracts, and operation monitoring. Among them, the user management module is responsible for the identity information management of all blockchain participants, including the maintenance of public and private key generation (account management), key management, and maintenance of the correspondence between the user’s real identity and the blockchain address (authority management), etc. In the case of authorization, supervise and audit certain real-identity transactions, and provide risk control rule configuration (risk control audit); basic service modules are deployed on all blockchain node devices to verify the validity of business requests, After completing the consensus on the valid request, it is recorded on the storage. For a new business request, the basic service first performs interface adaptation analysis and authentication processing (interface adaptation), and then encrypts the business information through the consensus algorithm (consensus management), After encryption, it is completely and consistently transmitted to the shared ledger (network communication), and recorded and stored; the smart contract module is responsible for contract registration and issuance, contract triggering and contract execution. Developers can define the contract logic through a certain programming language and publish it to On the blockchain (contract registration), according to the logic of the contract terms, call keys or other events to trigger execution, complete the contract logic, and also provide the function of contract upgrade and cancellation; the operation monitoring module is mainly responsible for the deployment of the product release process , Configuration modification, contract settings, cloud adaptation, and visual output of real-time status during product operation, such as: alarms, monitoring network conditions, monitoring node equipment health status, etc.

Referring to FIG. 5, an embodiment of the present application also provides a computer device. The computer device may be a server, and its internal structure may be as shown in FIG. 5. The computer equipment includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, the processor designed by the computer is used to provide calculation and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer equipment is used to store all the data needed for the process of analyzing semantic emotions. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program is executed by the processor to realize the method of analyzing semantic emotion.

The processor executes the above method for analyzing semantic emotions, including: obtaining the importance metric value corresponding to each word in the sentence to be analyzed; according to the importance metric value corresponding to each word in the sentence to be analyzed, passing two The cyclic neural network model runs in parallel to obtain the implicit expression corresponding to the sentence to be analyzed, wherein the implicit expression incorporates the semantic dependency of the context; the implicit expression corresponding to the sentence to be analyzed is combined with The preset sentence label corresponding to the sentence to be analyzed is input into the semantic sentiment analysis classifier; and the result of sentiment analysis of the sentence to be analyzed by the semantic sentiment analysis classifier is received.

The above-mentioned computer equipment introduces a self-attention mechanism to quantify the importance of each word in the sentence through the importance measurement value, and then according to the position of the important word, through the RNN improved by the hierarchical traversal of the tree, the current The meaning of words in the entire sentence, so that each word in the sentence and its corresponding importance measure are integrated in the hidden state of the entire sentence in the final output.

In one embodiment, the above-mentioned processor obtains the implicit expression corresponding to the sentence to be analyzed through two parallel running cyclic neural network models according to the importance metric value corresponding to each word in the sentence to be analyzed. The step includes: splitting the sentence to be analyzed into a tree structure according to the importance metric value corresponding to each word in the sentence to be analyzed, wherein the tree structure includes leaf nodes, child nodes, and Root node; the clauses contained in the first leaf node are input into the first recurrent neural network according to the forward order of the sentence to be analyzed, and the clauses contained in the second leaf node are input according to the reverse order of the sentence to be analyzed The second recurrent neural network, wherein the first leaf node and the second leaf node are a pair of leaf nodes that belong to any designated child node; the forward hidden vector output by the first recurrent neural network, Multiply the inverse hidden vector output by the second recurrent neural network to obtain the vector expression of the designated child node; according to the vector expression of the designated child node, according to the tree structure, sequentially recursively calculate to the first word corresponding The vector expression of the root node of the first word is the word corresponding to the maximum importance metric value; the vector expression of the root node corresponding to the first word is used as the implicit expression of the sentence to be analyzed.

In one embodiment, the above-mentioned processor splits the sentence to be analyzed into a tree structure according to the importance metric value corresponding to each word in the sentence to be analyzed, including: according to the sentence to be analyzed The importance metric value corresponding to each word in each word in the sentence to be analyzed is determined, and the first word with the largest importance metric value in the sentence to be analyzed is determined; the first word is used as the demarcation point, and the sentence to be analyzed is split into the first sentence Clause and the second clause, wherein the first word is the root node of the tree structure; the second word with the largest importance metric in the first clause is used as the demarcation point, and the first word One clause is split into a third clause and a fourth clause, and the second clause is split into a fifth clause with the third word with the largest importance measure in the second clause as the demarcation point. Sentence and the sixth clause, wherein the second word and the third word are both child nodes of the root node; according to the splitting process of the first clause and the second clause, all From the sentence to be analyzed to the leaf node, a tree structure composed of multiple layers of nodes is formed, wherein the leaf node is a node without child nodes.

In one embodiment, the step of obtaining the importance metric value corresponding to each word in the sentence to be analyzed by the above-mentioned processor includes: performing word embedding and position coding on the sentence to be analyzed to obtain each word in the sentence to be analyzed. Corresponding vector expression; the vector expression corresponding to each word in the sentence to be analyzed is sequentially input into the self-attention network according to the order in the sentence to be analyzed, and each word in the sentence to be analyzed is obtained. The corresponding importance measure.

In one embodiment, the above-mentioned processor performs word embedding and position encoding on the sentence to be analyzed to obtain the vector expression corresponding to each word in the sentence to be analyzed, including: calculating the specified word by word embedding according to the first calculation formula After the first vector, calculate the second vector corresponding to the position code of the specified word according to the second formula; calculate the vector expression corresponding to the specified word by the third calculation formula according to the first vector and the second vector; According to the calculation process of the vector expression corresponding to the specified word, the vector expression corresponding to each word in the sentence to be analyzed is calculated.

In one embodiment, the above-mentioned processor expresses the vector corresponding to each word in the sentence to be analyzed, and sequentially inputs the vector expression in the sentence to be analyzed into the self-attention network to obtain the sentence to be analyzed. The step of the importance metric value corresponding to each word includes: expressing the vector corresponding to each word in the sentence to be analyzed, and sequentially inputting it into the self-attention network according to the order in the sentence to be analyzed; The fourth calculation formula in the self-attention network is called to respectively calculate the importance metric value corresponding to each word in the sentence to be analyzed.

In one embodiment, before the step of obtaining the importance metric value corresponding to each word in the sentence to be analyzed by the above-mentioned processor, it includes: loading a preset classification function into the classifier, and initializing the assignment; expressing the vector of the training sentence And sentence labels, input into the classifier loaded with the preset classification function for classification training; determine whether the loss function is convergent, where the loss function is the cross entropy of the predicted classification result and the real classification result; if it is, it is determined that the training has obtained the semantic emotion Analyze the classifier.

Those skilled in the art can understand that the structure shown in FIG. 5 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.

An embodiment of the present application also provides a computer-readable storage medium. The computer-readable storage medium may be non-volatile or volatile, and has a computer program stored thereon, which is realized when the computer program is executed by a processor. The method for analyzing semantic sentiment includes: obtaining the importance metric value corresponding to each word in the sentence to be analyzed; according to the importance metric value corresponding to each word in the sentence to be analyzed, through two parallel running loops The network model obtains the implicit expression corresponding to the sentence to be analyzed, where the implicit expression incorporates the semantic dependency of the context; and the implicit expression corresponding to the sentence to be analyzed and the sentence to be analyzed The corresponding sentence label is input to the semantic sentiment analysis classifier; the sentiment analysis classification result of the sentence to be analyzed by the semantic sentiment analysis classifier is received.

The above-mentioned computer-readable storage medium introduces a self-attention mechanism to quantify the importance of each word in the sentence through the importance measurement value, and then according to the position of the important word, the RNN is improved by the idea of traversing the hierarchy of the tree. , Get the meaning of the current word in the entire sentence, so that each word in the sentence and its corresponding importance measurement value are integrated in the hidden state of the entire sentence in the final output.

In one embodiment, the above-mentioned processor performs word embedding and position coding on the sentence to be analyzed to obtain the vector expression corresponding to each word in the sentence to be analyzed, including: calculating the specified word by word embedding according to the first calculation formula After the first vector, calculate the second vector corresponding to the position code of the specified word according to the second formula; calculate the vector expression corresponding to the specified word by the third calculation formula according to the first vector and the second vector; According to the calculation process of the vector expression corresponding to the specified word, the vector expression corresponding to each word in the sentence to be analyzed is calculated.

A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by computer programs instructing relevant hardware. The above-mentioned computer programs can be stored in a non-volatile computer readable storage medium. Here, when the computer program is executed, it may include the processes of the above-mentioned method embodiments. Wherein, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and/or volatile memory. Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not a limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Claims

A method of analyzing semantic sentiment, including:

Obtain the importance metric value corresponding to each word in the sentence to be analyzed;

According to the importance metric value corresponding to each word in the sentence to be analyzed, the implicit expression corresponding to the sentence to be analyzed is obtained through two cyclic neural network models running in parallel, wherein the implicit expression Incorporating the semantic dependency of context;

Input the implicit expression corresponding to the sentence to be analyzed and the preset sentence label corresponding to the sentence to be analyzed into the semantic sentiment analysis classifier;

Receiving the sentiment analysis classification result of the sentence to be analyzed by the semantic sentiment analysis classifier.
The method for analyzing semantic sentiment according to claim 1, wherein, according to the importance metric value corresponding to each word in the sentence to be analyzed, through two parallel running cyclic neural network models, the to-be-analyzed sentence is obtained. The steps to analyze the implicit expression corresponding to the sentence include:

Splitting the sentence to be analyzed into a tree structure according to the importance metric value corresponding to each word in the sentence to be analyzed, wherein the tree structure includes a leaf node, a child node, and a root node;

The clauses contained in the first leaf node are input into the first loop neural network according to the forward order of the sentence to be analyzed, and the clauses contained in the second leaf node are input into the second loop according to the reverse order of the sentence to be analyzed A neural network, wherein the first leaf node and the second leaf node are a pair of leaf nodes that belong to any designated child node;

Multiply the forward hidden vector output by the first recurrent neural network by the reverse hidden vector output by the second recurrent neural network to obtain the vector expression of the designated child node;

According to the vector expression of the designated child node, according to the tree structure, sequentially recursively calculate the vector expression to the root node corresponding to the first word, where the first word is the word corresponding to the maximum importance metric;

The vector expression of the root node corresponding to the first word is used as the implicit expression of the sentence to be analyzed.
The method for analyzing semantic emotion according to claim 2, wherein the step of splitting the sentence to be analyzed into a tree structure according to the importance metric value corresponding to each word in the sentence to be analyzed, include:

Determine the first word with the largest importance metric in the sentence to be analyzed according to the importance metric value corresponding to each word in the sentence to be analyzed;

Using the first word as a dividing point, split the sentence to be analyzed into a first clause and a second clause, wherein the first word is used as the root node of the tree structure;

Taking the second word with the largest importance measure in the first clause as the demarcation point, the first clause is divided into a third clause and a fourth clause, and the second clause is important in the second clause. The third word with the largest sexual metric is the demarcation point. The second clause is split into a fifth clause and a sixth clause, where the second word and the third word are both the root Child nodes of the node;

According to the splitting process of the first clause and the second clause, the sentence to be analyzed is split into leaf nodes to form a tree structure composed of multi-layer nodes, wherein the leaf nodes are those without child nodes node.
The method for analyzing semantic sentiment according to claim 1, wherein the step of obtaining the importance metric value corresponding to each word in the sentence to be analyzed comprises:

Performing word embedding and position coding on the sentence to be analyzed to obtain a vector expression corresponding to each word in the sentence to be analyzed;

The vector expression corresponding to each word in the sentence to be analyzed is sequentially input into the self-attention network according to the order in the sentence to be analyzed, and the importance metric corresponding to each word in the sentence to be analyzed is obtained. value.
The method for analyzing semantic emotions according to claim 4, wherein the step of performing word embedding and position coding on the sentence to be analyzed to obtain the vector expression corresponding to each word in the sentence to be analyzed comprises:

Calculate the first vector of the specified word after the word is embedded according to the first calculation formula, and calculate the second vector of the position code corresponding to the specified word according to the second formula;

According to the first vector and the second vector, the vector expression corresponding to the designated word is calculated by a third calculation formula;

According to the calculation process of the vector expression corresponding to the specified word, the vector expression corresponding to each word in the sentence to be analyzed is calculated.
The method for analyzing semantic emotions according to claim 4, wherein the vector expression corresponding to each word in the sentence to be analyzed is sequentially input into the self-attention network according to the order in the sentence to be analyzed , The step of obtaining the importance metric value corresponding to each word in the sentence to be analyzed includes:

The vector expression corresponding to each word in the sentence to be analyzed is sequentially input into the self-attention network according to the order in the sentence to be analyzed;

By calling the fourth calculation formula in the self-attention network, the importance metric value corresponding to each word in the sentence to be analyzed is calculated respectively.
The method for analyzing semantic sentiment according to claim 1, wherein before the step of obtaining the importance metric value corresponding to each word in the sentence to be analyzed, the method comprises:

Load the preset classification function to the classifier and initialize the assignment;

Input the vector expression and sentence label of the training sentence into the classifier loaded with the preset classification function for classification training;

Judging whether the loss function converges, where the loss function is the cross entropy of the predicted classification result and the true classification result;

If it is, it is determined that the semantic sentiment analysis classifier is obtained by training.
A device for analyzing semantic emotions, which includes:

The acquiring module is used to acquire the importance metric value corresponding to each word in the sentence to be analyzed;

The obtaining module is used to obtain the implicit expression corresponding to the sentence to be analyzed through two parallel running cyclic neural network models according to the importance metric value corresponding to each word in the sentence to be analyzed. Implicit expressions incorporate the semantic dependency of context;

An input module for inputting the implicit expression corresponding to the sentence to be analyzed and the preset sentence label corresponding to the sentence to be analyzed into the semantic sentiment analysis classifier;

The receiving module is configured to receive the sentiment analysis classification result of the sentence to be analyzed by the semantic sentiment analysis classifier.
A computer device includes a memory and a processor, the memory stores a computer program, and a method for analyzing semantic emotions when the processor executes the computer program, which includes:

Obtain the importance metric value corresponding to each word in the sentence to be analyzed;

According to the importance metric value corresponding to each word in the sentence to be analyzed, the implicit expression corresponding to the sentence to be analyzed is obtained through two cyclic neural network models running in parallel, wherein the implicit expression Incorporating the semantic dependency of context;

Input the implicit expression corresponding to the sentence to be analyzed and the preset sentence label corresponding to the sentence to be analyzed into the semantic sentiment analysis classifier;

Receiving the sentiment analysis classification result of the sentence to be analyzed by the semantic sentiment analysis classifier.
8. The computer device according to claim 9, wherein the corresponding importance metric value of each word in the sentence to be analyzed is obtained through two cyclic neural network models running in parallel to obtain the sentence corresponding to the sentence to be analyzed. The steps of the implicit expression include:

Splitting the sentence to be analyzed into a tree structure according to the importance metric value corresponding to each word in the sentence to be analyzed, wherein the tree structure includes a leaf node, a child node, and a root node;

The clauses contained in the first leaf node are input into the first loop neural network according to the forward order of the sentence to be analyzed, and the clauses contained in the second leaf node are input into the second loop according to the reverse order of the sentence to be analyzed A neural network, wherein the first leaf node and the second leaf node are a pair of leaf nodes that belong to any designated child node;

Multiply the forward hidden vector output by the first recurrent neural network by the reverse hidden vector output by the second recurrent neural network to obtain the vector expression of the designated child node;

According to the vector expression of the designated child node, according to the tree structure, sequentially recursively calculate the vector expression to the root node corresponding to the first word, where the first word is the word corresponding to the maximum importance metric;

The vector expression of the root node corresponding to the first word is used as the implicit expression of the sentence to be analyzed.
10. The computer device according to claim 10, wherein the step of splitting the sentence to be analyzed into a tree structure according to the importance metric value corresponding to each word in the sentence to be analyzed comprises:

Determine the first word with the largest importance metric in the sentence to be analyzed according to the importance metric value corresponding to each word in the sentence to be analyzed;

Using the first word as a dividing point, split the sentence to be analyzed into a first clause and a second clause, wherein the first word is used as the root node of the tree structure;

Taking the second word with the largest importance measure in the first clause as the demarcation point, the first clause is divided into a third clause and a fourth clause, and the second clause is important in the second clause. The third word with the largest sexual metric is the demarcation point. The second clause is split into a fifth clause and a sixth clause, where the second word and the third word are both the root Child nodes of the node;

According to the splitting process of the first clause and the second clause, the sentence to be analyzed is split into leaf nodes to form a tree structure composed of multi-layer nodes, wherein the leaf nodes are those without child nodes node.
The computer device according to claim 9, wherein the step of obtaining the importance metric value corresponding to each word in the sentence to be analyzed comprises:

Performing word embedding and position coding on the sentence to be analyzed to obtain a vector expression corresponding to each word in the sentence to be analyzed;

The vector expression corresponding to each word in the sentence to be analyzed is sequentially input into the self-attention network according to the order in the sentence to be analyzed, and the importance metric corresponding to each word in the sentence to be analyzed is obtained. value.
The computer device according to claim 12, wherein the step of performing word embedding and position coding on the sentence to be analyzed to obtain the vector expression corresponding to each word in the sentence to be analyzed comprises:

Calculate the first vector of the specified word after the word is embedded according to the first calculation formula, and calculate the second vector of the position code corresponding to the specified word according to the second formula;

According to the first vector and the second vector, the vector expression corresponding to the designated word is calculated by a third calculation formula;

According to the calculation process of the vector expression corresponding to the specified word, the vector expression corresponding to each word in the sentence to be analyzed is calculated.
The computer device according to claim 13, wherein the vector expression corresponding to each word in the sentence to be analyzed is sequentially input into the self-attention network according to the order in the sentence to be analyzed to obtain the The steps to describe the importance metric value corresponding to each word in the sentence to be analyzed include:

The vector expression corresponding to each word in the sentence to be analyzed is sequentially input into the self-attention network according to the order in the sentence to be analyzed;

By calling the fourth calculation formula in the self-attention network, the importance metric value corresponding to each word in the sentence to be analyzed is calculated respectively.
A computer-readable storage medium has a computer program stored thereon, and when the computer program is executed by a processor, a method for analyzing semantic emotions is realized, which includes:

Obtain the importance metric value corresponding to each word in the sentence to be analyzed;

According to the importance metric value corresponding to each word in the sentence to be analyzed, the implicit expression corresponding to the sentence to be analyzed is obtained through two cyclic neural network models running in parallel, wherein the implicit expression Incorporating the semantic dependency of context;

Input the implicit expression corresponding to the sentence to be analyzed and the preset sentence label corresponding to the sentence to be analyzed into the semantic sentiment analysis classifier;

Receiving the sentiment analysis classification result of the sentence to be analyzed by the semantic sentiment analysis classifier.
The computer-readable storage medium according to claim 15, wherein said to-be-analyzed sentence obtains said to-be-analyzed cyclic neural network model through two parallel running recurrent neural network models according to the importance metric value corresponding to each word in said sentence to be analyzed. The steps to analyze the implicit expression corresponding to the sentence include:

Splitting the sentence to be analyzed into a tree structure according to the importance metric value corresponding to each word in the sentence to be analyzed, wherein the tree structure includes a leaf node, a child node, and a root node;

The clauses contained in the first leaf node are input into the first loop neural network according to the forward order of the sentence to be analyzed, and the clauses contained in the second leaf node are input into the second loop according to the reverse order of the sentence to be analyzed A neural network, wherein the first leaf node and the second leaf node are a pair of leaf nodes that belong to any designated child node;

Multiply the forward hidden vector output by the first recurrent neural network by the reverse hidden vector output by the second recurrent neural network to obtain the vector expression of the designated child node;

According to the vector expression of the designated child node, according to the tree structure, sequentially recursively calculate the vector expression to the root node corresponding to the first word, where the first word is the word corresponding to the maximum importance metric;

The vector expression of the root node corresponding to the first word is used as the implicit expression of the sentence to be analyzed.
16. The computer-readable storage medium according to claim 16, wherein the step of splitting the sentence to be analyzed into a tree structure according to the importance metric value corresponding to each word in the sentence to be analyzed, include:

Determine the first word with the largest importance metric in the sentence to be analyzed according to the importance metric value corresponding to each word in the sentence to be analyzed;

Using the first word as a dividing point, split the sentence to be analyzed into a first clause and a second clause, wherein the first word is used as the root node of the tree structure;

Taking the second word with the largest importance measure in the first clause as the demarcation point, the first clause is divided into a third clause and a fourth clause, and the second clause is important in the second clause. The third word with the largest sexual metric is the demarcation point. The second clause is split into a fifth clause and a sixth clause, where the second word and the third word are both the root Child nodes of the node;

According to the splitting process of the first clause and the second clause, the sentence to be analyzed is split into leaf nodes to form a tree structure composed of multiple layers of nodes, wherein the leaf nodes are those without child nodes node.
15. The computer-readable storage medium according to claim 15, wherein the step of obtaining the importance metric value corresponding to each word in the sentence to be analyzed comprises:

Performing word embedding and position coding on the sentence to be analyzed to obtain a vector expression corresponding to each word in the sentence to be analyzed;

The vector expression corresponding to each word in the sentence to be analyzed is sequentially input into the self-attention network according to the order in the sentence to be analyzed, and the importance metric corresponding to each word in the sentence to be analyzed is obtained. value.
18. The computer-readable storage medium according to claim 18, wherein the step of performing word embedding and position coding on the sentence to be analyzed to obtain the vector expression corresponding to each word in the sentence to be analyzed comprises:

Calculate the first vector of the specified word after the word is embedded according to the first calculation formula, and calculate the second vector of the position code corresponding to the specified word according to the second formula;

According to the first vector and the second vector, the vector expression corresponding to the designated word is calculated by a third calculation formula;

According to the calculation process of the vector expression corresponding to the specified word, the vector expression corresponding to each word in the sentence to be analyzed is calculated.
The computer-readable storage medium according to claim 19, wherein the vector expression corresponding to each word in the sentence to be analyzed is sequentially input into the self-attention network according to the order in the sentence to be analyzed , The step of obtaining the importance metric value corresponding to each word in the sentence to be analyzed includes:

The vector expression corresponding to each word in the sentence to be analyzed is sequentially input into the self-attention network according to the order in the sentence to be analyzed;

By calling the fourth calculation formula in the self-attention network, the importance metric value corresponding to each word in the sentence to be analyzed is calculated respectively.