Method for sentiment analysis of network public opinion of platform enterprise based on semantic enhancement network

Abstract

The application discloses a platform enterprise network public opinion sentiment analysis method based on a semantic enhancement network, and comprises the following steps: forming a platform enterprise network public opinion sentiment dataset based on platform enterprise public opinion related texts and a platform enterprise network public opinion sentiment dictionary; performing multi-level vector representation on the words of the sentences in the platform enterprise network public opinion sentiment dataset to obtain semantic vectors of the words in the sentences; performing syntactic analysis on the platform enterprise public opinion related texts to calculate probability values of the syntactic analysis trees; performing semantic enhancement matching based on the semantic vectors of the platform enterprise network public opinion sentiment dataset to obtain semantic vectors enhanced in semantics; performing sentence embedding representation based on the semantic vectors enhanced in semantics; and performing sentiment presetting based on the result of the sentence embedding representation to obtain a prediction result. The method disclosed by the application realizes fast and efficient sentiment analysis by using deep learning technology, and is helpful for understanding the emotional state of the public and obtaining public opinions and attitudes in a timely manner.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data processing, and particularly relates to a platform enterprise network public opinion sentiment analysis method based on a semantic enhancement network. BACKGROUND

[0002] The new media, such as microblogging, WeChat public account, Zhihu, and self-media based on the Internet greatly improve the speed and range of platform enterprise public opinion dissemination. The generation, expansion and dissemination of platform enterprise public opinion have important influences on investors, platform enterprises, market supervision and even macroeconomic operation. For platform enterprises, some small credit crises may lead to platform enterprise crisis events, so it is necessary to grasp the early public opinion monitoring to reduce and avoid the outbreak of platform enterprise public opinion crisis. The relevant departments of the state or market supervision can avoid the fermentation of various market-related problems in the network and disturb the market order by monitoring platform enterprise public opinion, so as to grasp the platform enterprise network public opinion risk points and do the risk investigation and evaluation work.

[0003] Positive public opinion can improve the market share and social reputation of the relevant platform enterprise, further enhance its influence and competitiveness, and stabilize the operation of the platform enterprise. Negative public opinion not only damages the interests and social reputation of the platform enterprise, but also destroys the credit foundation on which the platform enterprise relies for survival and development, and may disturb the market order. SUMMARY

[0004] The embodiment of the present application provides a platform enterprise network public opinion sentiment analysis method based on a semantic enhancement network, which realizes rapid and efficient sentiment analysis and judges whether the current platform enterprise public opinion is positive or negative.

[0005] The embodiment of the present application provides a platform enterprise network public opinion sentiment analysis method based on a semantic enhancement network, which includes:

[0006] Based on the platform enterprise public opinion related text monitored in the Internet and the platform enterprise public opinion sentiment dictionary constructed, a platform enterprise public opinion sentiment data set is formed;

[0007] The words in the sentences in the platform enterprise public opinion sentiment data set are represented by multi-level vectors to obtain the semantic vectors of the words in the sentences;

[0008] The platform enterprise public opinion related text is subjected to syntactic analysis, the probability value of the syntactic analysis tree is calculated, and denoising processing is performed;

[0009] Based on the platform enterprise public opinion related text after denoising and the semantic vectors of the platform enterprise public opinion sentiment data set, semantic enhancement matching is performed to obtain the semantic vectors after semantic enhancement;

[0010] performing sentence embedding representation based on the semantic enhanced semantic vectors;

[0011] performing sentiment presetting based on the result of the sentence embedding representation to obtain a prediction result.

[0012] Optionally, based on the platform enterprise public opinion related text monitored in the Internet and the constructed platform enterprise public opinion sentiment dictionary, a platform enterprise public opinion sentiment dataset is formed, which includes:

[0013] The platform enterprise public opinion related text monitored in the Internet and the constructed platform enterprise public opinion sentiment dictionary are taken as text input data, and a TF / IDF algorithm is used for screening to form a platform enterprise public opinion sentiment dataset composed of positive sentiment dictionary and negative sentiment dictionary.

[0014] Optionally, the words in the sentences in the platform enterprise public opinion sentiment dataset are represented by multi-level vectors to obtain semantic vectors of the words in the sentences, which includes:

[0015] Based on the words in the sentences in the platform enterprise public opinion sentiment dataset:

[0016] using a convolution network to perform character level vector representation;

[0017] based on a GloVe word embedding tool to obtain word level vector representation;

[0018] based on a Bert embedding tool to obtain sentence level vector representation;

[0019] concatenating the character level vector representation, the word level vector representation and the sentence level vector representation, and encoding the concatenated information through an Albert model to obtain semantic vectors of the words in the sentences, which includes:

[0020]

[0021] wherein, represents the semantic vector of the word, represents the character level vector representation of the word w, is the word level vector representation of the word w, represents the sentence level vector representation of the word w.

[0022] Optionally, performing syntactic analysis on the platform enterprise public opinion related text, calculating the probability value of the syntactic analysis tree, and performing denoising processing, which includes:

[0023] using a PCFG syntactic analysis model to perform syntactic analysis on the platform enterprise public opinion related text, and calculating the probability value of the syntactic analysis tree;

[0024] All parse trees are sorted according to their probability values, and the parse tree with the highest probability value is selected as the analysis result; and

[0025] Based on the probability value of each word in the sentence obtained from PCFG, the syntax tree is pruned to remove words with low probabilities.

[0026] Optionally, semantic enhancement matching is performed based on the denoised text related to the platform enterprise's public opinion and the semantic vector of the platform enterprise's public opinion sentiment dataset to obtain semantically enhanced semantic vectors, including:

[0027] Based on the denoised text related to public opinion of the platform enterprises and the platform enterprise public opinion sentiment dataset, semantic enhancement matching is performed to construct a semantic enhancement matrix. .

[0028] Optionally, sentence embedding representation based on semantically enhanced semantic vectors includes:

[0029] The semantic enhancement matrix is ​​used as input for convolution operation:

[0030]

[0031] in It is the ReLU function. Indicates the first l Layer f The characteristic mapping of the type output port, representing, , × For window size, and For parameters, Indicates the first Layer feature number;

[0032] use and The width and height are used as the pooling operations:

[0033]

[0034] The matching vector output by pooling is mapped to a low-dimensional space to obtain the matching vector[ ];

[0035] The matching vector is used as input to the sentence embedding module, and a GRU unit is used to obtain the output hidden vector. [] is used as a sentence embedding representation.

[0036] Optionally, sentiment pre-setting based on the sentence embedding representation to obtain prediction results includes:

[0037] Hidden vector [ ] as input to the sentiment prediction module, defining a matching score :

[0038]

[0039] wherein, 、 、 、 、 are parameters, is a virtual text vector learned in training, is the i-th hidden state of the hidden vector ], i is the final hidden state of the word w.

[0040] The application also proposes a platform enterprise network public opinion sentiment analysis device based on a semantic enhancement network, comprising:

[0041] A text input module: used to form a platform enterprise network public opinion sentiment dataset based on platform enterprise public opinion related texts monitored in the Internet and a platform enterprise network public opinion sentiment dictionary constructed;

[0042] A word embedding module: used to perform multi-level vector representation on the words of the sentences in the platform enterprise network public opinion sentiment dataset to obtain semantic vectors of the words in the sentences;

[0043] A semantic enhancement module: used to perform syntactic analysis on the platform enterprise public opinion related texts, calculate the probability values of the syntactic analysis trees, and perform denoising processing; based on the denoised platform enterprise public opinion related texts and the semantic vectors of the platform enterprise network public opinion sentiment dataset, perform semantic enhancement matching to obtain semantic vectors enhanced in semantics;

[0044] A sentence embedding module: used to perform sentence embedding representation based on the semantic vectors enhanced in semantics;

[0045] A sentiment prediction module: used to perform sentiment prediction based on the results of the sentence embedding representation to obtain a prediction result.

[0046] The application also proposes a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the platform enterprise network public opinion sentiment analysis method based on the semantic enhancement network as described above.

[0047] ​The platform enterprise network public opinion sentiment analysis method based on the semantic enhancement network provided in the embodiment of the application utilizes deep learning technology to realize rapid and efficient sentiment analysis, is directed to the emotion analysis of network financial texts, is helpful to understand the emotional state of the public, timely obtain public opinion and attitude, and quickly reach information subjects.

[0048] The above description is only a summary of the technical scheme of the application, in order to more clearly understand the technical means of the application, the content of the specification can be implemented, and in order to make the above and other purposes, characteristics and advantages of the application more obvious and easy to understand, the following specific embodiments of the application are described. BRIEF DESCRIPTION OF DRAWINGS

[0049] Various other advantages and benefits will become apparent to those of ordinary skill in the art, upon reading the following detailed description of the preferred embodiments. The accompanying drawings are intended to depict only preferred embodiments of the application, and therefore should not be considered to narrow the scope of the present application in any way. Instead, they are included to provide illustration of the embodiments of the application in conformity with the principles of the present disclosure. In the drawings:

[0050] Figure 1 The basic flowchart of the network public opinion sentiment analysis method of the embodiment of the application is shown in the figure. DETAILED DESCRIPTION

[0051] Exemplary embodiments of the present disclosure will be described more fully hereinafter with reference to the accompanying drawings, in which exemplary embodiments of the present disclosure are shown. It is to be understood that the present disclosure can be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the present disclosure to those skilled in the art.

[0052] The embodiment of the application provides a platform enterprise network public opinion sentiment analysis method based on a semantic enhancement network, as shown in the figure, comprising: Figure 1

[0053] ​In step S101, based on the platform enterprise public opinion related text monitored in the Internet and the constructed platform enterprise public opinion sentiment dictionary, a platform enterprise public opinion sentiment dataset is formed. In some specific examples, based on the platform enterprise public opinion related text monitored in the Internet and the constructed platform enterprise public opinion sentiment dictionary, a platform enterprise public opinion sentiment dataset is formed, including: taking the platform enterprise public opinion related text monitored in the Internet and the constructed platform enterprise public opinion sentiment dictionary as text input data, using TF / IDF algorithm for screening, to form a platform enterprise public opinion sentiment dataset composed of positive sentiment dictionary and negative sentiment dictionary. For example, sentiment analysis can be carried out on platform enterprise related microblog information on the Internet and user comment text information on major financial websites. Public opinion text comes from major social media platforms that platform enterprises pay close attention to; the platform enterprise public opinion sentiment dataset combines the current Chinese sentiment dictionary (such as HOWNET, DLUTSD), and combines the relevant literature and reports in the platform enterprise field, and carries out screening through word segmentation, TF / IDF algorithm and manual means, to form a platform enterprise public opinion sentiment dataset composed of positive sentiment dictionary and negative sentiment dictionary.

[0054] In step S102, the words in the sentences in the platform enterprise public opinion sentiment dataset are represented by multi-level vectors to obtain the semantic vectors of the words in the sentences. In some embodiments, the words in the sentences in the platform enterprise public opinion sentiment dataset are represented by multi-level vectors to obtain the semantic vectors of the words in the sentences, including:

[0055] Based on the words in the sentences in the platform enterprise public opinion sentiment dataset:

[0056] Using a convolution network for character-level vector representation;

[0057] Based on the GloVe word embedding tool to obtain word-level vector representation;

[0058] Based on the Bert embedding tool to obtain sentence-level vector representation;

[0059] Concatenate the character-level vector representation, the word-level vector representation and the sentence-level vector representation, and encode the concatenated information through the Albert model to obtain the semantic vectors of the words in the sentences:

[0060]

[0061] wherein, represents the semantic vector of the word, represents the character-level vector representation of the word w, is the word-level vector representation of the word w, represents the sentence-level vector representation of the word w.

[0062] In step S103, the platform enterprise public opinion related text is subjected to syntactic analysis, the probability value of the syntactic analysis tree is calculated, and denoising processing is performed. In some embodiments, the platform enterprise public opinion related text is subjected to syntactic analysis, the probability value of the syntactic analysis tree is calculated, and denoising processing is performed, including:

[0063] The platform enterprise public opinion related text is subjected to syntactic analysis by using a PCFG (Probabilistic Context Free Grammar) syntactic analysis model, the probability value of the syntactic analysis tree is calculated, when there are multiple probability trees, all analysis trees are sorted according to the probability value, and the syntactic analysis tree with a higher probability value is selected as the analysis result. And

[0064] According to the probability value of each word in the sentence obtained by PCFG, the syntactic tree is pruned, the words with low probability are removed, the noise of long sentences is removed, and the process of identifying the original short sentences is performed.

[0065] In step S104, based on the denoised platform enterprise public opinion related text and the semantic vector of the platform enterprise public opinion sentiment data set, semantic enhancement matching is performed to obtain a semantic vector enhanced by semantics. In some embodiments, based on the denoised platform enterprise public opinion related text and the semantic vector of the platform enterprise public opinion sentiment data set, semantic enhancement matching is performed to obtain a semantic vector enhanced by semantics, including: based on the denoised platform enterprise public opinion related text and the platform enterprise public opinion sentiment data set, semantic enhancement matching is performed to construct a semantic enhancement matrix .

[0066] Specifically, after obtaining the denoised public opinion text, a semantic enhancement matrix is constructed The words in the denoised public opinion text and the platform enterprise public opinion sentiment data set are subjected to semantic enhancement matching, represents and the similarity between them, wherein , dimension The matrix is defined as follows:

[0067]

[0068] Thus, the semantics of the sentiment words in the platform enterprise dictionary is integrated, the semantics of the word vectors in the public opinion text is strengthened, and the semantics of the public opinion text is further enhanced.

[0069] In step S105, sentence embedding representation is performed based on the semantic vector enhanced by semantics. In some embodiments, sentence embedding representation is performed based on the semantic vector enhanced by semantics, including:

[0070] The semantic enhancement matrix is taken as input, and a convolution operation is performed:

[0071]

[0072] wherein is a ReLU function, represents the feature map of the output port of the l type of the f layer, represents , x is a window size, and are parameters, represents the number of features of the layer;

[0073] A pooling operation is performed using and as the width and height of the pooling operation:

[0074]

[0075] The matching vector of the pooling output is mapped into a low-dimensional space to obtain a matching vector .

[0076] The matching vector is taken as input of a sentence embedding module, and a GRU unit is used to obtain an output hidden vector as a sentence embedding representation.

[0077] In step S106, sentiment prediction is performed based on the result of the sentence embedding representation to obtain a prediction result. In some embodiments, performing sentiment prediction based on the result of the sentence embedding representation to obtain a prediction result comprises:

[0078] The hidden vector is taken as input of a sentiment prediction module, and a matching score is defined:

[0079]

[0080] wherein , , , , are parameters, is a virtual text vector learned in training, is the th hidden state of the hidden vector i , is the final hidden state of the word w.

[0081] The internet platform enterprise has the characteristics of low cost, high efficiency, wide coverage and high risk. The emerging social media represented by microblog provides relevant platform enterprise information. Compared with traditional platform enterprises, the platform enterprise network public opinion sentiment analysis method based on semantic enhancement network provided in the application uses deep learning technology to realize rapid and efficient sentiment analysis, and is helpful to understand the emotional state of the public, obtain public opinion and attitude in time, and quickly reach the information subject.

[0082] In the representation process of words in platform enterprise public opinion text, the application provides a multi-granularity word embedding method, which combines character-level, word-level and sentence-level word embedding, and multi-level fusion can obtain semantic-rich word vector representation to capture deeper local semantics of platform enterprise public opinion text.

[0083] The platform enterprise public opinion sentiment dataset is constructed, and the weight of the sentiment words in the platform enterprise public opinion text is strengthened based on the platform enterprise public opinion sentiment dataset, and the semantic enhancement neural network is used to realize the semantic enhancement of the platform enterprise public opinion sentiment, so as to obtain the embedded representation of the platform enterprise public opinion text with more comprehensive semantics.

[0084] The semantic enhancement process of the application prunes the syntax tree of the platform enterprise public opinion text through the PCFG syntax analysis model to reduce the noise in the sentence, and then uses the denoised public opinion text and the platform enterprise sentiment dictionary to carry out semantic enhancement matching to strengthen the semantic of the text vector of the public opinion text.

[0085] Based on the rich platform enterprise sentiment semantic sentence vector, the application carries out sentiment prediction, uses the softmax function to carry out judgment, and outputs the current platform enterprise public opinion state, whether it is positive public opinion, negative public opinion or no emotion, and finally realizes the rapid discovery of platform enterprise public opinion, and further guides the platform enterprise to carry out risk prevention and response.

[0086] The application also provides a platform enterprise network public opinion sentiment analysis device based on a semantic enhancement network, comprising:

[0087] A text input module is used to form a platform enterprise public opinion sentiment dataset based on the platform enterprise public opinion related text monitored in the Internet and the platform enterprise public opinion sentiment dictionary constructed;

[0088] A word embedding module is used to perform multi-level vector representation on the words of the sentences in the platform enterprise public opinion sentiment dataset to obtain the semantic vector of the words in the sentences;

[0089] The semantic enhancement module is configured to perform syntax analysis on the platform enterprise public opinion related text, calculate a probability value of a syntax analysis tree, and perform denoising processing; perform semantic enhancement matching based on the denoised platform enterprise public opinion related text and a semantic vector of a platform enterprise public opinion sentiment data set, to obtain a semantic vector enhanced in semantics.

[0090] The sentence embedding module is configured to perform sentence embedding representation based on the semantic vector enhanced in semantics.

[0091] The sentiment prediction module is configured to perform sentiment prediction based on a result of the sentence embedding representation, to obtain a prediction result.

[0092] The present application also provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the platform enterprise network public opinion sentiment analysis method based on the semantic enhancement network.

[0093] It should be noted that, in this document, the terms "comprising", "including", or any other variant thereof are intended to cover non-exclusive inclusions, so that a process, method, article, or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such a process, method, article, or device. Without more limitations, the element defined by the phrase "including a" does not exclude the presence of additional identical elements in the process, method, article, or device including the element.

[0094] The above-mentioned embodiment numbers of the present application are only for description, and do not represent the advantages or disadvantages of the embodiments.

[0095] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be realized by means of software and necessary general hardware platforms, of course, they can also be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes a plurality of instructions for making a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) execute the methods described in various embodiments of the present application.

[0096] The embodiments of the present application are described above with reference to the accompanying drawings, but the present application is not limited to the above-described specific embodiments, and the above-described specific embodiments are merely illustrative, but not restrictive, and a person of ordinary skill in the art can make many forms under the inspiration of the present application without departing from the purpose of the present application and the scope protected by the claims, and these all belong to the protection of the present application.

Claims

1. A platform enterprise network public opinion sentiment analysis method based on a semantic enhancement network, characterized in that, The method comprises the following steps: Based on the platform enterprise public opinion related text monitored in the Internet and the constructed platform enterprise public opinion sentiment dictionary, a platform enterprise public opinion sentiment data set is formed; The words in the sentences in the platform enterprise public opinion sentiment data set are represented by multi-level vectors to obtain the semantic vectors of the words in the sentences; The platform enterprise public opinion related text is subjected to syntactic analysis, the probability value of the syntactic analysis tree is calculated, and denoising processing is performed; Based on the denoised platform enterprise public opinion related text and the semantic vectors of the platform enterprise public opinion sentiment data set, semantic enhancement matching is performed to obtain semantic enhancement semantic vectors; Based on the semantic enhancement semantic vectors, sentence embedding representation is performed; Based on the results of the sentence embedding representation, sentiment prediction is performed to obtain a prediction result; The platform enterprise public opinion related text is subjected to syntactic analysis, the probability value of the syntactic analysis tree is calculated, and denoising processing is performed, which comprises the following steps: The platform enterprise public opinion related text is subjected to syntactic analysis by using a PCFG syntactic analysis model, and the probability value of the syntactic analysis tree is calculated; All analysis trees are sorted according to the probability value, and the syntactic analysis tree with a high probability value is selected as the analysis result; and According to the probability value of each word in the sentence obtained by the PCFG, the syntactic tree is pruned, and the words with low probability are removed; The sentence embedding representation based on the semantic enhancement semantic vectors comprises the following steps: The semantic enhancement matrix is taken as input and subjected to convolution operation: wherein is a ReLU function, denotes the feature map of the output port of the l layer of type f , , x is a window size, and are parameters, denotes the feature number of the layer; Using and pooling operation as the width and height of the pooling operation: The matching vectors of the pooling outputs are mapped into a low-dimensional space to obtain matching vectors [f (x) ]k ] The matching vector is used as input to the sentence embedding module, and a GRU unit is used to obtain the output hidden vector. ] as a sentence embedding representation; The sentiment prediction based on the results of the sentence embedding representation to obtain a prediction result comprises the following steps: The hidden vector [h] is taken as an input to the sentiment prediction module, which defines a matching score S: :​ wherein , , , , is a parameter, is a virtual text vector learned in training, is the th hidden state of a hidden vector i , is the final hidden state of the word w.

2. The platform enterprise network public opinion sentiment analysis method based on semantic enhancement network according to claim 1, wherein, Based on the platform enterprise public opinion related text monitored in the Internet and the constructed platform enterprise public opinion sentiment dictionary, a platform enterprise public opinion sentiment data set is formed, which comprises the following steps: The platform enterprise public opinion related text monitored in the Internet and the constructed platform enterprise public opinion sentiment dictionary are taken as text input data, and a TF / IDF algorithm is used for screening to form a platform enterprise public opinion sentiment data set composed of a positive sentiment dictionary and a negative sentiment dictionary.

3. The platform enterprise network public opinion sentiment analysis method based on semantic enhancement network according to claim 2, wherein, The words in the sentences in the platform enterprise public opinion sentiment data set are represented by multi-level vectors to obtain the semantic vectors of the words in the sentences, which comprises the following steps: Based on the words in the sentences in the platform enterprise public opinion sentiment data set: Character-level vector representation is performed by using a convolutional network; Word-level vector representation is obtained based on a GloVe word embedding tool; Sentence-level vector representation is obtained based on a Bert embedding tool; The character-level vector representation, the word-level vector representation and the sentence-level vector representation are concatenated, and the concatenated information is encoded by an Albert model to obtain the semantic vectors of the words in the sentences: wherein, represents a semantic vector for a word, represents a character-level vector representation for a word w, is a word-level vector representation for a word w, represents a sentence-level vector representation for a word w.

4. The platform enterprise network public opinion sentiment analysis method based on semantic enhancement network according to claim 1, wherein, The semantic enhancement matching based on the denoised platform enterprise public opinion related text and the semantic vectors of the platform enterprise public opinion sentiment data set to obtain semantic enhancement semantic vectors comprises the following steps: Based on the denoised platform enterprise public opinion related text, and, platform enterprise public opinion sentiment dataset is matched for semantic enhancement to construct a semantic enhancement matrix .

5. A platform enterprise network public opinion sentiment analysis device based on a semantic enhancement network, characterized in that, It comprises: A text input module is used to form a platform enterprise public opinion sentiment data set based on the platform enterprise public opinion related text monitored in the Internet and the constructed platform enterprise public opinion sentiment dictionary; A word embedding module is used to represent the words in the sentences in the platform enterprise public opinion sentiment data set by multi-level vectors to obtain the semantic vectors of the words in the sentences; The semantic enhancement module is configured to perform syntactic analysis on the platform enterprise public opinion related text, calculate probability values of syntactic analysis trees, and perform noise removal processing. The semantic enhancement module is configured to perform syntactic analysis on the platform enterprise public opinion related text, calculate probability values of syntactic analysis trees, and perform noise removal processing. The semantic enhancement module is configured to perform syntactic analysis on the platform enterprise public opinion related text, calculate probability values of syntactic analysis trees, and perform noise removal processing. The semantic enhancement module is configured to perform syntactic analysis on the platform enterprise public opinion related text, calculate probability values of syntactic analysis trees, and perform noise removal processing. The semantic enhancement module is configured to perform syntactic analysis on the platform enterprise public opinion related text, calculate probability values of syntactic analysis trees, and perform noise removal processing. The semantic enhancement module is configured to perform syntactic analysis on the platform enterprise public opinion related text, calculate probability values of syntactic analysis trees, and perform noise removal processing. The semantic enhancement module is configured to perform syntactic analysis on the platform enterprise public opinion related text, calculate probability values of syntactic analysis trees, and perform noise removal processing. The semantic enhancement module is configured to perform syntactic analysis on the platform enterprise public opinion related text, calculate probability values of syntactic analysis trees, and perform noise removal processing. The semantic enhancement module is configured to perform syntactic analysis on the platform enterprise public opinion related text, calculate probability values of syntactic analysis trees, and perform noise removal processing. The semantic enhancement module is configured to perform syntactic analysis on the platform enterprise public opinion related text, calculate probability values of syntactic analysis trees, and perform noise removal processing. wherein is a ReLU function, denotes the l layer's f type output port's feature map, , × is a window size, and are parameters, denotes the layer's feature number; Using and pooling operation as the width and height of the pooling operation: The matching vectors of the pooling outputs are mapped into a low-dimensional space to obtain matching vectors [f (x) ]k ] The matching vector is used as input to the sentence embedding module, and a GRU unit is used to obtain the output hidden vector. ] as a sentence embedding representation; The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the platform enterprise network public opinion sentiment analysis method based on the semantic enhancement network according to any one of claims 1 to 4. The hidden vector [h] is taken as an input to the sentiment prediction module, which defines a matching score S: :​ wherein , , , , is a parameter, is a virtual text vector learned in training, is the th hidden state of the hidden vector i , is the final hidden state of the word w.

6. A computer-readable storage medium, characterized in that, ​

Images