A financial expert language semantic sentiment analysis system and method

Abstract

The present invention relates to the technical field of financial data processing, and in particular to a financial expert language semantic sentiment analysis system and method; the present invention forms a word embedding matrix by performing word segmentation, part-of-speech tagging, entity recognition and professional terminology extraction on financial expert speech data, performs sentiment polarity analysis on each text vector in the word embedding matrix, and performs category classification on the financial expert speech data, obtains the sentiment intensity of the word embedding matrix by combining the sentiment polarity and category classification, constructs a market sentiment evaluation model based on historical financial expert speech data and its corresponding historical market feedback data, uses the market sentiment evaluation model to evaluate the market sentiment of the word embedding matrix, and performs correlation analysis between the market sentiment and market key indicators to obtain the influence of the financial expert speech data on the market key indicators; the present invention realizes the sentiment analysis of the financial expert speech data and the influence analysis of the financial expert speech data on the market key indicators.

Description

Technical Field

[0001] The present invention relates to the technical field of financial data processing, and in particular to a system and method for analyzing the semantic sentiment of financial expert language. Background Art

[0002] In the current financial field, the analysis of financial experts' comments mainly relies on manual interpretation, which in turn relies on financial professionals, resulting in low efficiency and the conclusions of the interpretation being easily influenced by subjective factors. Existing automated analysis tools often cannot accurately understand financial professional terms and complex contexts, resulting in inaccurate sentiment analysis. In addition, the lack of prediction and risk warning functions for market sentiment trends makes it difficult for investors and decision makers to grasp market dynamics and potential risks. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention aims to provide a financial expert language semantic sentiment analysis system and method to solve the problem that it is difficult to obtain effective and reliable financial information from the financial expert's speech.

[0004] In order to solve the above problems, the present invention adopts the following technical solutions:

[0005] On the one hand, the present invention provides a financial expert language semantic sentiment analysis system, including a natural language processing module, a sentiment analysis module and a market sentiment assessment module;

[0006] The natural language processing module is used to perform word segmentation, part-of-speech tagging, entity recognition and professional terminology extraction on the financial expert speech data to form a word embedding matrix;

[0007] The sentiment analysis module is used to perform sentiment polarity analysis on each text vector in the word embedding matrix, where the sentiment polarity includes positive, negative and neutral, and classify the financial expert speech data into optimistic, cautious or pessimistic categories, and combine the sentiment polarity and category classification to obtain the sentiment intensity of the word embedding matrix;

[0008] The market sentiment assessment module is used to build a market sentiment assessment model based on historical financial expert speech data and its corresponding historical market feedback data, use the market sentiment assessment model to evaluate the market sentiment of the word embedding matrix, and perform correlation analysis between the market sentiment and key market indicators to obtain the impact of the financial expert speech data on the key market indicators.

[0009] As an implementable method, the financial expert speech data is segmented, part-of-speech tagged, entity recognized, and professional terminology extracted to form a word embedding matrix, including:

[0010] The word segmentation model trained with the financial field corpus is used to segment the speech data of financial experts;

[0011] Perform part-of-speech tagging on the financial experts’ speech data after word segmentation;

[0012] Use long short-term memory network or Transformer model to identify financial entities in the speech data of financial experts after part-of-speech tagging;

[0013] A deep learning model combined with a dictionary in the financial field is used to extract professional terms to form a financial professional terminology library. The professional terms of financial entities are extracted by calculating the cosine similarity between the identified financial entities and the professional terms in the financial professional terminology library.

[0014] The word embedding matrix of financial expert speech data is constructed based on the extracted professional terms of financial entities.

[0015] As an implementable method, the sentiment polarity analysis of each text vector in the word embedding matrix includes:

[0016] Use the trained deep neural network model to analyze each text vector in the word embedding matrix and calculate the sentiment polarity score of each word;

[0017] The categories of optimistic, cautious or pessimistic classification of financial expert opinions data include:

[0018] Use TF-IDF or word2vec to convert the context information of the financial expert speech data into a feature vector, and integrate the feature vector with the score of the sentiment polarity to obtain a comprehensive feature vector;

[0019] The financial expert speech data is classified into categories through the trained softmax regression model;

[0020] The sentiment intensity of the word embedding matrix obtained by combining sentiment polarity and category classification includes:

[0021] The sentiment intensity of the word embedding matrix is ​​calculated based on the weights and polarity scores of different sentiment categories.

[0022] As an implementable method, the market sentiment assessment model is constructed based on historical financial expert speech data and its corresponding historical market feedback data, including:

[0023] Collect historical speech data of financial experts and relevant market feedback data, and perform data preprocessing;

[0024] Extract key features from the preprocessed data to construct a key feature vector, wherein the key features include sentiment polarity scores of historical speech data of financial experts, category classifications, and market key indicators of relevant market feedback data;

[0025] Based on the constructed key feature vectors, the sequence data is constructed and trained using the long short-term memory network to obtain the market sentiment assessment model.

[0026] As an implementation method, the market sentiment evaluation model is used to evaluate the market sentiment of the word embedding matrix, and the market sentiment is correlated with the key market indicators to obtain the impact of the financial expert speech data on the key market indicators, including:

[0027] According to the market sentiment of the word embedding matrix and the score of sentiment polarity and the change curve of the market key indicators of the relevant market feedback data matched with the category classification, the influence of the financial expert speech data on the market key indicators is obtained.

[0028] In another aspect, the present invention provides a method for analyzing the semantic sentiment of financial expert language, comprising:

[0029] Perform word segmentation, part-of-speech tagging, entity recognition, and professional terminology extraction on financial expert speech data to form a word embedding matrix;

[0030] Perform sentiment polarity analysis on each text vector in the word embedding matrix. Sentiment polarity includes positive, negative, and neutral. Financial experts’ speech data is classified into optimistic, cautious, or pessimistic categories. The sentiment intensity of the word embedding matrix is ​​obtained by combining sentiment polarity and category classification.

[0031] A market sentiment evaluation model is constructed based on historical financial expert speech data and its corresponding historical market feedback data. The market sentiment evaluation model is used to evaluate the market sentiment of the word embedding matrix, and the correlation analysis between market sentiment and key market indicators is performed to obtain the impact of financial expert speech data on key market indicators.

[0032] As an implementable method, the financial expert speech data is segmented, part-of-speech tagged, entity recognized, and professional terminology extracted to form a word embedding matrix, including:

[0033] The word segmentation model trained with the financial field corpus is used to segment the speech data of financial experts;

[0034] Perform part-of-speech tagging on the financial experts’ speech data after word segmentation;

[0035] Use long short-term memory network or Transformer model to identify financial entities in the speech data of financial experts after part-of-speech tagging;

[0036] A deep learning model combined with a dictionary in the financial field is used to extract professional terms to form a financial professional terminology library. The professional terms of financial entities are extracted by calculating the cosine similarity between the identified financial entities and the professional terms in the financial professional terminology library.

[0037] The word embedding matrix of financial expert speech data is constructed based on the extracted professional terms of financial entities.

[0038] As an implementable method, the sentiment polarity analysis of each text vector in the word embedding matrix includes:

[0039] Use the trained deep neural network model to analyze each text vector in the word embedding matrix and calculate the sentiment polarity score of each word;

[0040] The categories of optimistic, cautious or pessimistic classification of financial expert opinions data include:

[0041] Use TF-IDF or word2vec to convert the context information of the financial expert speech data into a feature vector, and integrate the feature vector with the score of the sentiment polarity to obtain a comprehensive feature vector;

[0042] The financial expert speech data is classified into categories through the trained softmax regression model;

[0043] The sentiment intensity of the word embedding matrix obtained by combining sentiment polarity and category classification includes:

[0044] The sentiment intensity of the word embedding matrix is ​​calculated based on the weights and polarity scores of different sentiment categories.

[0045] As an implementable method, the market sentiment assessment model is constructed based on historical financial expert speech data and its corresponding historical market feedback data, including:

[0046] Collect historical speech data of financial experts and relevant market feedback data, and perform data preprocessing;

[0047] Extract key features from the preprocessed data to construct a key feature vector, wherein the key features include sentiment polarity scores of historical speech data of financial experts, category classifications, and market key indicators of relevant market feedback data;

[0048] Based on the constructed key feature vectors, the sequence data is constructed and trained using the long short-term memory network to obtain the market sentiment assessment model.

[0049] As an implementation method, the market sentiment evaluation model is used to evaluate the market sentiment of the word embedding matrix, and the market sentiment is correlated with the key market indicators to obtain the impact of the financial expert speech data on the key market indicators, including:

[0050] According to the market sentiment of the word embedding matrix and the score of sentiment polarity and the change curve of the market key indicators of the relevant market feedback data matched with the category classification, the influence of the financial expert speech data on the market key indicators is obtained.

[0051] The beneficial effects of the present invention are as follows: the financial expert language semantic sentiment analysis system and method of the present invention realizes sentiment analysis of financial expert speech data by performing sentiment polarity analysis and category classification on financial expert speech data, and associates the data based on market feedback data to obtain the relationship between the financial expert speech data and key market indicators, thereby realizing the impact analysis of the financial expert speech data on key market indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a schematic diagram of a financial expert language semantic sentiment analysis system of the present invention.

[0053] Figure 2 This is a flow chart of a method for analyzing the semantic sentiment of financial expert language according to the present invention. DETAILED DESCRIPTION

[0054] The present invention is further described in detail below in conjunction with specific embodiments.

[0055] It should be noted that these embodiments are only used to illustrate the present invention rather than to limit the present invention. Simple improvements to the method based on the concept of the present invention all fall within the scope of protection claimed by the present invention.

[0056] See also Figure 1 , which is a financial expert language semantic sentiment analysis system, including a natural language processing module 100, a sentiment analysis module 200 and a market sentiment assessment module 300.

[0057] The natural language processing module 100 is used to perform word segmentation, part-of-speech tagging, entity recognition and professional terminology extraction on the financial expert speech data to form a word embedding matrix.

[0058] The word segmentation of financial expert speech data includes: using a word segmentation model trained with a financial field corpus to segment the financial expert speech data. The word segmentation model can combine rule methods and statistical methods to not only identify common words, but also accurately segment financial professional terms.

[0059] Part-of-speech tagging includes: tagging the part-of-speech of financial expert speech data after word segmentation; assigning a part-of-speech label to each word, such as noun, verb, adjective, etc., to provide more accurate semantic information for subsequent sentiment analysis.

[0060] Entity recognition includes: using long short-term memory networks or Transformer models to identify financial entities in the speech data of financial experts after part-of-speech tagging. Identify financial entities in the text, such as names of people, institutions, stock codes, etc., to extract key information.

[0061] Professional term extraction includes: using a deep learning model combined with a financial field dictionary to extract professional terms to form a financial professional term library, and extracting professional terms of financial entities by calculating the cosine similarity between the identified financial entities and the professional terms in the financial professional term library.

[0062] Forming a word embedding matrix includes: constructing a word embedding matrix of financial expert speech data based on the extracted professional terms of the financial entity.

[0063]

[0064] in, represents the word embedding matrix, represents the input vocabulary sequence, It is a word embedding function that maps each word into a dense vector of fixed dimension that can capture the semantic information of the word.

[0065] For example, take a comment from a financial expert: "The market has been volatile recently, but the investment value of high-quality stocks is still obvious." In the implementation process, the sentence is first segmented into vocabulary units such as "recently", "market", "volatility", and "larger" through word segmentation. Then, part-of-speech tagging is performed, such as "recently" is marked as a time noun and "larger" is marked as an adjective. Then, the entity recognition and term extraction steps will identify financial entities and professional terms such as "market" and "high-quality stocks". Finally, through word embedding technology, each word is converted into a vector that can express its deep semantics. These vectors are then input into the deep learning model for sentiment analysis to determine the emotional tendency expressed by the comment.

[0066] The sentiment analysis module 200 is used to perform sentiment polarity analysis on each text vector in the word embedding matrix. The sentiment polarity includes positive, negative and neutral, and classifies the financial expert speech data into optimistic, cautious or pessimistic categories. The sentiment intensity of the word embedding matrix is ​​obtained by combining the sentiment polarity and category classification.

[0067] The specific process can be:

[0068] The trained deep neural network model is used to analyze each text vector in the word embedding matrix and calculate the sentiment polarity score of each word. The deep neural network can be a convolutional neural network (CNN) or a recurrent neural network (RNN).

[0069]

[0070] in, A score representing the sentiment polarity, is the word embedding matrix, It is the sentiment polarity analysis function.

[0071] The categories of optimistic, cautious or pessimistic financial expert opinions data include:

[0072] Use TF-IDF or word2vec to convert the contextual information of financial expert speech data into feature vectors, and integrate the feature vectors with their sentiment polarity scores to obtain a comprehensive feature vector. The contextual information includes:

[0073] Tone and emotion level: Contextual information is not divided into several levels, but is obtained and expressed in a variety of ways. The following are some common ways to obtain it:

[0074] Lexical features: analysis of parts of speech, word frequency, word embeddings, etc.

[0075] Sentence features: sentence length, sentence structure, grammatical analysis, etc.

[0076] Chapter features: chapter structure, paragraph relationships, topic consistency, etc.

[0077] The financial expert speech data is classified into categories through the trained softmax regression model.

[0078] The softmax regression model is trained using a comprehensive feature vector constructed from sentiment features and context features extracted from financial expert speech data, and the output is the classification probability of the financial expert speech data being optimistic, cautious, or pessimistic.

[0079]

[0080] in, represents the probability distribution of text classification, Represents contextual information, A score representing the sentiment polarity, is a text classification function.

[0081] Combining sentiment polarity and category classification to obtain the sentiment strength of the word embedding matrix includes:

[0082] The sentiment intensity of the word embedding matrix is ​​calculated based on the weights and polarity scores of different sentiment categories.

[0083]

[0084] Among them, I represents the emotional intensity value, is the weight of the i-th sentiment category, is the score of sentiment polarity, and N is the total number of words.

[0085] For example, the word "volatility" may be judged as neutral, while "quality stocks" may be judged as positive sentiment. In the text classification stage, the entire comment may be classified as "optimistic". In the sentiment intensity assessment, according to the model calculation, the comment may obtain a higher sentiment intensity value, indicating that the comment has a greater positive impact on market sentiment. Through this series of analyses, the present invention can provide investors with a quantitative assessment of market sentiment, thereby assisting investment decisions.

[0086] Through the above steps, the present invention can not only accurately judge the emotional tendency of financial speech, but also evaluate the emotional intensity in combination with contextual factors, providing an efficient and reliable solution for market sentiment analysis.

[0087] The market sentiment assessment module 300 is used to construct a market sentiment assessment model based on historical financial expert speech data and its corresponding historical market feedback data, use the market sentiment assessment model to evaluate the market sentiment of the word embedding matrix, and perform correlation analysis between the market sentiment and key market indicators to obtain the impact of the financial expert speech data on the key market indicators.

[0088] Among them, building a market sentiment assessment model includes:

[0089] Collect historical speech data of financial experts and related market feedback data, and perform data preprocessing.

[0090] Preprocessing includes:

[0091] Collect and organize historical speech data of financial experts and market feedback data, including but not limited to stock price fluctuations, trading volume changes, social media sentiment, etc. Clean and normalize the data to eliminate noise and outliers in the data.

[0092]

[0093] in, is the normalized data, is the original data, and are the minimum and maximum values ​​in the data set, respectively.

[0094] Key features are extracted from the preprocessed data to construct key feature vectors. The key features include sentiment polarity scores of historical speech data of financial experts, category classifications, and market key indicators of related market feedback data.

[0095]

[0096] in, is the eigenvector, is the i-th feature.

[0097] Based on the constructed key feature vectors, the sequence data is constructed and trained using the long short-term memory network to obtain the market sentiment assessment model.

[0098]

[0099] in, is the market sentiment assessment result, are model parameters.

[0100] The market sentiment evaluation model is used to evaluate the market sentiment of the word embedding matrix, and the correlation analysis between market sentiment and key market indicators is performed to obtain the impact of financial expert speech data on key market indicators, including:

[0101] Based on the market sentiment of the word embedding matrix and the score of sentiment polarity and the change curve of the market key indicators (such as index rise and fall, trading volume change) of the relevant market feedback data and category classification, the impact of financial expert speech data on the market key indicators is obtained.

[0102] For example, first, through the data preprocessing step, the historical speech data of the expert and the relevant market feedback data are collected. In the feature extraction stage, the sentiment polarity score of the comment, the text classification result, etc. are used as features. Then, the feature vector is processed using the constructed LSTM sentiment evaluation model to obtain the market sentiment evaluation result. Through sentiment quantitative analysis, we can find that there is a significant positive correlation between the expert's speech and the market trading volume, indicating that the expert's speech has a driving effect on the market trading activity / or the expert's speech has market correctness. This analysis result can provide investors with in-depth insights into market dynamics and assist them in making more reasonable investment decisions.

[0103] See also Figure 2 A method for analyzing the semantic sentiment of financial expert language, including:

[0104] S100, perform word segmentation, part-of-speech tagging, entity recognition and professional term extraction on the financial expert speech data to form a word embedding matrix;

[0105] S200, performing sentiment polarity analysis on each text vector in the word embedding matrix, where the sentiment polarity includes positive, negative and neutral, and classifying the financial expert speech data into optimistic, cautious or pessimistic categories, and combining the sentiment polarity and category classification to obtain the sentiment intensity of the word embedding matrix;

[0106] S300. Construct a market sentiment assessment model based on historical financial expert speech data and its corresponding historical market feedback data. Use the market sentiment assessment model to assess the market sentiment of the word embedding matrix, and perform correlation analysis between market sentiment and key market indicators to obtain the impact of financial expert speech data on key market indicators.

[0107] The word segmentation, part-of-speech tagging, entity recognition and professional terminology extraction of the financial expert speech data to form a word embedding matrix includes:

[0108] The word segmentation model trained with the financial field corpus is used to segment the speech data of financial experts;

[0109] Perform part-of-speech tagging on the financial experts’ speech data after word segmentation;

[0110] Use long short-term memory network or Transformer model to identify financial entities in the speech data of financial experts after part-of-speech tagging;

[0111] A deep learning model combined with a dictionary in the financial field is used to extract professional terms to form a financial professional terminology library. The professional terms of financial entities are extracted by calculating the cosine similarity between the identified financial entities and the professional terms in the financial professional terminology library.

[0112] The word embedding matrix of financial expert speech data is constructed based on the extracted professional terms of financial entities.

[0113] The performing sentiment polarity analysis on each text vector in the word embedding matrix includes:

[0114] Use the trained deep neural network model to analyze each text vector in the word embedding matrix and calculate the sentiment polarity score of each word;

[0115] The categories of optimistic, cautious or pessimistic classification of financial expert opinions data include:

[0116] Use TF-IDF or word2vec to convert the context information of the financial expert speech data into a feature vector, and integrate the feature vector with the score of the sentiment polarity to obtain a comprehensive feature vector;

[0117] The financial expert speech data is classified into categories through the trained softmax regression model;

[0118] The sentiment intensity of the word embedding matrix obtained by combining sentiment polarity and category classification includes:

[0119] The sentiment intensity of the word embedding matrix is ​​calculated based on the weights and polarity scores of different sentiment categories.

[0120] The method of constructing a market sentiment assessment model based on historical financial expert speech data and its corresponding historical market feedback data includes:

[0121] Collect historical speech data of financial experts and relevant market feedback data, and perform data preprocessing;

[0122] Extract key features from the preprocessed data to construct a key feature vector, wherein the key features include sentiment polarity scores of historical speech data of financial experts, category classifications, and market key indicators of relevant market feedback data;

[0123] Based on the constructed key feature vectors, the sequence data is constructed and trained using the long short-term memory network to obtain the market sentiment assessment model.

[0124] The market sentiment evaluation model is used to evaluate the market sentiment of the word embedding matrix, and the market sentiment is correlated with the key market indicators to obtain the impact of the financial expert speech data on the key market indicators, including:

[0125] According to the market sentiment of the word embedding matrix and the score of sentiment polarity and the change curve of the market key indicators of the relevant market feedback data matched with the category classification, the influence of the financial expert speech data on the market key indicators is obtained.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described with reference to the preferred embodiments of the present invention, it should be understood by those skilled in the art that various changes may be made in form and details without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A financial expert language semantic sentiment analysis system, characterized in that: Includes natural language processing module, sentiment analysis module and market sentiment assessment module; The natural language processing module is used to perform word segmentation, part-of-speech tagging, entity recognition and professional terminology extraction on the financial expert speech data to form a word embedding matrix; The sentiment analysis module is used to perform sentiment polarity analysis on each text vector in the word embedding matrix, where the sentiment polarity includes positive, negative and neutral, and classify the financial expert speech data into optimistic, cautious or pessimistic categories, and combine the sentiment polarity and category classification to obtain the sentiment intensity of the word embedding matrix; The market sentiment evaluation module is used to build a market sentiment evaluation model based on historical financial expert speech data and its corresponding historical market feedback data, use the market sentiment evaluation model to evaluate the market sentiment of the word embedding matrix, and perform correlation analysis between market sentiment and key market indicators to obtain the impact of financial expert speech data on key market indicators; The financial expert speech data is segmented, part-of-speech tagged, entity recognized, and professional terminology extracted to form a word embedding matrix, including: The word segmentation model trained with the financial field corpus is used to segment the speech data of financial experts; Perform part-of-speech tagging on the financial experts’ speech data after word segmentation; Use long short-term memory network or Transformer model to identify financial entities in the speech data of financial experts after part-of-speech tagging; A deep learning model combined with a dictionary in the financial field is used to extract professional terms to form a financial professional terminology library. The professional terms of financial entities are extracted by calculating the cosine similarity between the identified financial entities and the professional terms in the financial professional terminology library. The word embedding matrix of financial expert speech data is constructed based on the extracted professional terms of financial entities.

2. The financial expert language semantic sentiment analysis system according to claim 1 is characterized in that: The sentiment polarity analysis of each text vector in the word embedding matrix includes: Use the trained deep neural network model to analyze each text vector in the word embedding matrix and calculate the sentiment polarity score of each word; The categories of optimistic, cautious or pessimistic classification of financial expert opinions data include: Use TF-IDF or word2vec to convert the context information of the financial expert speech data into a feature vector, and integrate the feature vector with the score of the sentiment polarity to obtain a comprehensive feature vector; The financial expert speech data is classified into categories through the trained softmax regression model; The sentiment intensity of the word embedding matrix obtained by combining sentiment polarity and category classification includes: The sentiment intensity of the word embedding matrix is ​​calculated based on the weights and polarity scores of different sentiment categories.

3. The financial expert language semantic sentiment analysis system according to claim 1, characterized in that: The market sentiment assessment model constructed based on historical financial expert speech data and its corresponding historical market feedback data includes: Collect historical speech data of financial experts and relevant market feedback data, and perform data preprocessing; Extract key features from the preprocessed data to construct a key feature vector, wherein the key features include sentiment polarity scores of historical speech data of financial experts, category classifications, and market key indicators of relevant market feedback data; Based on the constructed key feature vectors, the sequence data is constructed and trained using the long short-term memory network to obtain the market sentiment assessment model.

4. The financial expert language semantic sentiment analysis system according to claim 3 is characterized in that: The market sentiment evaluation model is used to evaluate the market sentiment of the word embedding matrix, and the market sentiment is analyzed for correlation with the key market indicators to obtain the impact of the financial expert speech data on the key market indicators, including: According to the market sentiment of the word embedding matrix and the score of sentiment polarity and the change curve of the market key indicators of the relevant market feedback data matched with the category classification, the influence of the financial expert speech data on the market key indicators is obtained.

5. A method for analyzing the semantic sentiment of financial expert language, characterized in that: include: Perform word segmentation, part-of-speech tagging, entity recognition, and professional terminology extraction on financial expert speech data to form a word embedding matrix; Perform sentiment polarity analysis on each text vector in the word embedding matrix. Sentiment polarity includes positive, negative, and neutral. Financial experts’ speech data is classified into optimistic, cautious, or pessimistic categories. The sentiment intensity of the word embedding matrix is ​​obtained by combining sentiment polarity and category classification. A market sentiment assessment model is built based on historical financial expert speech data and its corresponding historical market feedback data. The market sentiment assessment model is used to evaluate the market sentiment of the word embedding matrix, and the correlation between market sentiment and key market indicators is analyzed to obtain the impact of financial expert speech data on key market indicators. The financial expert speech data is segmented, part-of-speech tagged, entity recognized, and professional terminology extracted to form a word embedding matrix, including: The word segmentation model trained with the financial field corpus is used to segment the speech data of financial experts; Perform part-of-speech tagging on the financial experts’ speech data after word segmentation; Use long short-term memory network or Transformer model to identify financial entities in the speech data of financial experts after part-of-speech tagging; A deep learning model combined with a dictionary in the financial field is used to extract professional terms to form a financial professional terminology library. The professional terms of financial entities are extracted by calculating the cosine similarity between the identified financial entities and the professional terms in the financial professional terminology library. The word embedding matrix of financial expert speech data is constructed based on the extracted professional terms of financial entities.

6. The method for analyzing the semantic sentiment of financial expert language according to claim 5, characterized in that: The sentiment polarity analysis of each text vector in the word embedding matrix includes: Use the trained deep neural network model to analyze each text vector in the word embedding matrix and calculate the sentiment polarity score of each word; The categories of optimistic, cautious or pessimistic classification of financial expert opinions data include: Use TF-IDF or word2vec to convert the context information of the financial expert speech data into a feature vector, and integrate the feature vector with the score of the sentiment polarity to obtain a comprehensive feature vector; The financial expert speech data is classified into categories through the trained softmax regression model; The sentiment intensity of the word embedding matrix obtained by combining sentiment polarity and category classification includes: The sentiment intensity of the word embedding matrix is ​​calculated based on the weights and polarity scores of different sentiment categories.

7. The method for analyzing the semantic sentiment of financial expert language according to claim 5, characterized in that: The market sentiment assessment model constructed based on historical financial expert speech data and its corresponding historical market feedback data includes: Collect historical speech data of financial experts and relevant market feedback data, and perform data preprocessing; Extract key features from the preprocessed data to construct a key feature vector, wherein the key features include sentiment polarity scores of historical speech data of financial experts, category classifications, and market key indicators of relevant market feedback data; Based on the constructed key feature vectors, the sequence data is constructed and trained using the long short-term memory network to obtain the market sentiment assessment model.

8. The method for analyzing the semantic sentiment of financial expert language according to claim 7, characterized in that: The market sentiment evaluation model is used to evaluate the market sentiment of the word embedding matrix, and the market sentiment is analyzed for correlation with the key market indicators to obtain the impact of the financial expert speech data on the key market indicators, including: According to the market sentiment of the word embedding matrix and the score of sentiment polarity and the change curve of the market key indicators of the relevant market feedback data matched with the category classification, the influence of the financial expert speech data on the market key indicators is obtained.

Images