A dictionary-based text sentiment analysis method, system and storage medium in the power field

Abstract

A dictionary-based text sentiment analysis method, system and storage medium in the electric power field, the method comprising the following steps: training a sentiment orientation judgment model based on a neural network and manually annotated data, the input of the sentiment orientation judgment model being user text data, and the output of the sentiment orientation judgment model being sentiment orientation; obtaining user interaction information, and screening text data corresponding to user interaction information whose sentiment polarity is a set target polarity and whose sentiment orientation is a service provider as a target text. In the present invention, the target text for reflecting the emotional fluctuations of the user due to the service of the service provider is extracted by combining the sentiment extreme value and the sentiment orientation, so as to further extract the reasons for the emotional fluctuations of the user in combination with the target text, thereby helping the service provider to improve services according to customer needs.

Description

Technical Field

[0001] The present invention relates to the fields of information processing and sentiment classification, and in particular to a dictionary-based text sentiment analysis method, system and storage medium in the electric power field. Background Art

[0002] With the fierce information storm brought by big data, massive data almost flooded our hard disks. Facing the vast amount of data, huge wealth is hidden in it. The useful information in various data is the most important and valuable gold mine. Among them, user emotional information, as a very important information, has great value in various fields. For example, user purchase feedback on shopping platforms; user comments under various news reports; user ordering opinions on various takeaway platforms are all closely related to user emotions. Merchants can further improve the goods they sell through the emotional information hidden in user comments, and journalists can understand public opinion and user tendencies through the emotional information implied in user comments. Therefore, how to extract emotional information in user discourse is very important and valuable, and various emotional classification methods are generated and developed rapidly under this background.

[0003] The user's emotions may be caused by the user themselves or by the service provider. If the service provider can accurately obtain the factors that cause the user's emotional fluctuations, it can make targeted improvements to improve the user experience. However, the current emotion classification method cannot help the service provider determine which party caused the user's emotional fluctuations. Summary of the invention

[0004] In order to solve the defect in the above-mentioned prior art that there is a lack of an emotion recognition method that can determine whether the user's emotional fluctuations are caused by the service provider, the present invention proposes a dictionary-based text sentiment analysis method, system and storage medium in the power field.

[0005] One of the purposes of the present invention adopts the following technical solution:

[0006] A dictionary-based text sentiment analysis method in the electric power field includes the following steps:

[0007] S1. Train a sentiment orientation judgment model based on a neural network and manually annotated data. The input of the sentiment orientation judgment model is user text data, and the output of the sentiment orientation judgment model is sentiment orientation. The user text data is text data generated by the user in the process of interacting with the service provider. The sentiment orientation is the object that causes the user's emotions. The sentiment orientation includes the user himself and the service provider. The service provider is an organization or an individual. The service provider that interacts with the user belongs to the service provider.

[0008] S2. Obtain user interaction information, obtain text data corresponding to the user interaction information as user text data, input the user text data into a sentiment orientation judgment model, and obtain the sentiment orientation output by the sentiment orientation judgment model as the sentiment orientation of the user interaction information; filter the text data corresponding to the user interaction information whose sentiment polarity is the set target polarity and whose sentiment orientation is the service provider as the target text; the sentiment polarity is divided into positive and negative, and the target polarity includes at least one of positive and negative.

[0009] Preferably, step S1 further includes: constructing a keyword dictionary to collect words used to reflect the user's emotional fluctuations based on factors of the service provider;

[0010] After step S2, the method further includes step S3: selecting the segmentation words existing in the set keyword dictionary from the segmentation words of the target text as the target segmentation words, and outputting the target segmentation words and sentiment polarity of the target text.

[0011] Preferably, step S2 specifically includes the following steps:

[0012] S21 obtains text data corresponding to user interaction information whose sentiment polarity is the target polarity as candidate text, and obtains segmented words of the candidate text;

[0013] S22. Acquire the sentiment orientation of the candidate text through the sentiment orientation judgment model, and acquire the candidate text whose sentiment orientation is the service provider as the target text.

[0014] Preferably, the method for determining the user's emotional polarity in step S2 is: obtaining text data corresponding to the user interaction information, and segmenting the text data, and then filtering the segmentation words existing in the emotional vocabulary dictionary from the segmentation words of the text data based on the set emotional vocabulary dictionary as emotional words, and then combining the emotional words, negative words and degree words to determine the emotional polarity of the text data.

[0015] Preferably, the method of determining the sentiment polarity of text data specifically includes the following steps:

[0016] SA21, segment the text data and extract segmented words;

[0017] SA22, filter the participle words existing in the set emotional vocabulary dictionary as emotional words;

[0018] SA23, combine each emotion word with the adjacent participle words before and after to form a paragraph;

[0019] SA24, obtain the sentiment value of the paragraph according to the sentiment polarity of the sentiment words and the negative words and degree words in the paragraph;

[0020] SA25. Sum the sentiment values ​​of each paragraph in the text data as the total sentiment value of the text data, and determine the sentiment polarity of the text data based on the total sentiment value.

[0021] Preferably, each word in the emotional vocabulary dictionary is provided with a corresponding emotional reference value; in step SA25, the calculation formula of the emotional value K of the paragraph is:

[0022] K=-ak 0 ×(b 1 ×b 2 ×...b n )

[0023] k 0 represents the emotional reference value corresponding to the emotional word in the emotional vocabulary dictionary, a represents the number of negative words in the paragraph, b i It represents the weight value corresponding to the i-th degree word in the paragraph, 1≤i≤n, n is the number of degree words in the paragraph; the weight value of each degree word is the design value.

[0024] Preferably, in step SA25, when the total sentiment value is a positive number, the sentiment polarity of the text data is judged to be positive; when the total sentiment value is a negative number, the sentiment polarity of the text data is judged to be negative.

[0025] Preferably, the user interaction information in step S2 is voice data, and the emotional polarity of the user interaction information is obtained based on the set voice emotion judgment model in step S2; the voice emotion judgment model is obtained based on neural network and manually annotated data training, the input of the voice emotion judgment model is voice data, and the output is the emotional polarity of the input voice data.

[0026] The second object of the present invention adopts the following technical solution:

[0027] A dictionary-based electric power field text sentiment analysis system comprises a memory and a processor, wherein the memory stores a computer program, and the processor is used to implement the above-mentioned dictionary-based electric power field text sentiment analysis method when executing the computer program.

[0028] One of the purposes of the present invention adopts the following technical solution:

[0029] A storage medium stores a computer program, which, when executed, is used to implement the above-mentioned dictionary-based text sentiment analysis method in the electric power field.

[0030] The advantages of the present invention are:

[0031] (1) In the present invention, the emotion extreme value and emotion orientation are combined to extract the target text that reflects the emotional fluctuations of the user due to the service of the service provider, so as to further extract the reasons for the emotional fluctuations of the user in combination with the target text, thereby helping the service provider to improve the service according to customer needs.

[0032] (2) In the present invention, the emotional orientation is processed based on the text data, which enables accurate extraction of rational information output by the user based on the words in the text data, thereby improving the accuracy of judging the emotional orientation.

[0033] (3) In the present invention, the target word segmentation is screened based on the keyword dictionary, which realizes the accurate extraction of the factors that cause the user's emotional fluctuations based on the service provider. Combining the target word segmentation and the emotional polarity is helpful to help the service provider quickly and accurately extract the user's opinions, so as to improve the service according to the user's wishes.

[0034] (4) In the present invention, candidate texts are first screened according to sentiment polarity, and user interaction information with a larger proportion of unsaturated emotions is filtered out, which greatly reduces the subsequent text processing workload and improves work efficiency.

[0035] (5) In the present invention, a method for identifying emotional polarity based on text data and voice data is provided at the same time, so that the method is applicable to any information interaction mode, has a wide range of applications, and is flexible and reliable.

[0036] (6) In the present invention, the sentiment polarity of text data is identified based on text segmentation, the part of speech of the segmented words of the text data is accurately identified, the accurate processing of the text data is achieved, and the credibility of the sentiment judgment is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A flowchart of a dictionary-based text sentiment analysis method in the electric power field proposed by the present invention;

[0038] Figure 2 This is a flow chart of a method for acquiring target text in the present invention;

[0039] Figure 3 This is a flow chart of a method for determining the sentiment polarity of text data proposed by the present invention. DETAILED DESCRIPTION

[0040] Emotional polarity: refers to whether the emotion contained in a sentence or text is positive or negative. Positive emotional polarity, such as happiness and excitement, is defined as positive; negative emotional polarity, such as disappointment, anger, and rage, is defined as negative.

[0041] A method for judging the sentiment polarity of text data

[0042] A method for judging the sentiment polarity of text data provided in this embodiment is to segment the text data, and then select the segmentation words existing in the sentiment vocabulary dictionary as sentiment words from the segmentation words based on the set sentiment vocabulary dictionary, and then judge the sentiment polarity of the text data by combining the sentiment words, negative words and degree words.

[0043] The method for determining the sentiment polarity of text data specifically comprises the following steps:

[0044] SA21, segment the text data and extract segmented words, where the segmented words are the words after the text data is segmented. Text segmentation is a common technical means for text processing, and this step can use any existing segmentation means to segment the text data.

[0045] SA22, filter the participle words in the set emotional vocabulary dictionary as emotional words;

[0046] SA23. Combine each emotional word with the adjacent participle words before and after to form a paragraph.

[0047] Specifically, after the text data is segmented, multiple segmented words are obtained, and the multiple segmented words are present in the text data, and the multiple segmented words are sorted according to their positions in the text data. In this step SA23, the segmented word used as the emotion word is formed into a paragraph according to the sorting, together with the previous segmented word and the next segmented word. Specifically, the front-to-back position relationship of the three segmented words in the paragraph is the same as the front-to-back position relationship of the three segmented words in the text data.

[0048] SA24. Obtain the sentiment value of the paragraph according to the sentiment polarity of the sentiment words and the negative words and degree words in the paragraph. Specifically, the sentiment polarity of the sentiment words can be obtained through a sentiment vocabulary dictionary.

[0049] In this implementation, in order to more accurately quantify the sentiment value of text data, each word in the adopted sentiment vocabulary dictionary is provided with a corresponding sentiment reference value.

[0050] In specific implementation, the sentiment value K of a paragraph can be obtained according to a calculation formula.

[0051] K=-ak 0 ×(b 1 ×b 2 ×...b n )

[0052] k 0 represents the emotional reference value corresponding to the emotional word in the emotional vocabulary dictionary, a represents the number of negative words in the paragraph, b iIt represents the weight value corresponding to the i-th degree word in the paragraph, 1≤i≤n, n is the number of degree words in the paragraph; the weight value of each degree word is the design value.

[0053] SA25, summing up the sentiment values ​​of each paragraph in the text data as the total sentiment value of the text data, and judging the sentiment polarity of the text data according to the total sentiment value. Specifically, in this step, when the total sentiment value is a positive number, the sentiment polarity of the text data is judged to be positive; when the total sentiment value is a negative number, the sentiment polarity of the text data is judged to be negative.

[0054] The method for determining the sentiment polarity of text data is explained below in conjunction with a specific embodiment.

[0055] Step 1: Get text data A = "The hall is very clean today, the business was handled quickly, I was not disappointed, I am very happy."

[0056] Step 2: Segment the above text data, remove punctuation marks, predicates, pronouns, and structural auxiliary words, and obtain segmented words "The lobby is very clean today / / The business is fast / / Not disappointed / / Very happy". " / / " indicates a segmentation. The segmentation word located in front of the segmentation can only form a paragraph with the previous adjacent segmentation word, and the segmentation word located after the segmentation can only form a paragraph with the next adjacent segmentation word.

[0057] Step 3: Combine the emotional vocabulary dictionary to obtain the emotional words "disappointment" and "happy". "Disappointment" and "happy" are both located at the sentence break, so they can only form a paragraph with the previous adjacent participle word.

[0058] Step 4: In this embodiment, two segments “not disappointed” and “very happy” can be obtained.

[0059] Assume that the emotional reference value associated with "disappointed" in the emotional vocabulary dictionary is -1, and the emotional reference value associated with "happy" in the emotional vocabulary dictionary is 2; the weight value associated with "no" in the set degree word dictionary is -1, and the weight value associated with "very" in the degree word dictionary is 2.

[0060] Then, the sentiment value of the sentence “not-disappointed” = -1╳-1=1; the sentiment value of the sentence “very-happy” = 2╳2=4.

[0061] Step 5: Calculate the total sentiment value KA=1+4=5 of the text data A. That is, the sentiment polarity of the text data A is positive.

[0062] A method for judging the emotional polarity of speech data

[0063] In this implementation, for voice data, the emotional polarity can be determined by combining timbre, volume, etc.

[0064] Specifically, the method for determining the emotional polarity of speech data provided in this embodiment obtains the emotional polarity of user interaction information based on a set speech emotion determination model.

[0065] The speech emotion judgment model is obtained based on neural network and manually annotated data training. The input of the speech emotion judgment model is speech data, and the output is the emotional polarity of the input speech data.

[0066] Specifically, obtaining the speech emotion judgment model includes the following steps:

[0067] Step 1: Obtain speech data samples that are manually labeled with emotional polarity;

[0068] Step 2: Train the selected neural network in combination with the speech data samples. After the training is completed, a speech emotion judgment model for identifying the emotional polarity of the speech data is obtained.

[0069] In this way, for the speech data, the emotional polarity of the speech data can be obtained by simply inputting the speech data into the speech emotion judgment model.

[0070] A dictionary-based text sentiment analysis method in the power field

[0071] A dictionary-based text sentiment analysis method in the electric power field provided in this embodiment includes the following steps:

[0072] S1. Train a sentiment orientation judgment model based on neural networks and manually annotated data. The input of the sentiment orientation judgment model is user text data, and the output of the sentiment orientation judgment model is sentiment orientation. The user text data is the text data generated by the user in the process of interaction with the service provider. The sentiment orientation is the object that causes the user's emotions, and the sentiment orientation includes the user himself and the service provider. The service provider is the user's interaction object, which can be a human customer service or a robot customer service. The service provider is an organization or an individual.

[0073] S2. Obtain user interaction information, and select text data corresponding to user interaction information whose sentiment polarity is the target polarity and whose sentiment is directed to the service provider as target text.

[0074] The target polarity includes at least one of positive and negative. Specifically, the target polarity can be set according to the needs of the service provider. For example, when the service provider hopes to improve the shortcomings according to the needs of the customer, the target polarity can be set to negative, so as to filter out the target text that reflects the negative emotional fluctuations of the user due to the service of the service provider.

[0075] In this step, when the user interaction information is text data, the emotional polarity of the user interaction information can be obtained by the above-mentioned method for judging the emotional polarity of text data. When the user interaction information is speech, the emotional polarity of the user interaction information can be obtained by the speech emotion judgment model; or after converting the user interaction information into text data, the emotional polarity of the user interaction information can be obtained by the above-mentioned method for judging the emotional polarity of text data.

[0076] In this step, the emotional orientation of the user interaction information is obtained by inputting the text data corresponding to the user interaction information as the user text data into the emotional orientation judgment model, and obtaining the emotional orientation output by the emotional orientation judgment model as the emotional orientation of the user interaction information. In this way, the emotional orientation is processed based on the text data, and the rational information output by the user is accurately extracted according to the words in the text data, thereby improving the accuracy of judging the emotional orientation.

[0077] In this embodiment, the emotional extreme value and emotional orientation are combined to extract the target text that reflects the emotional fluctuations of the user due to the service of the service provider, so that the reasons for the emotional fluctuations of the user can be further extracted in combination with the target text, thereby helping the service provider to improve the service according to customer needs.

[0078] In specific implementation, the extraction of target text can be divided into the following two steps:

[0079] S21. Obtain text data corresponding to user interaction information with a target sentiment polarity as candidate text, and obtain segmented words of the candidate text.

[0080] S22. Acquire the sentiment orientation of the candidate text through the sentiment orientation judgment model, and acquire the candidate text whose sentiment orientation is the service provider as the target text.

[0081] In this implementation, candidate texts are first screened according to sentiment polarity, and user interaction information with unsaturated emotions (specifically, user interaction information that does not contain sentiment words or has a sentiment value of 0) is filtered out, which greatly reduces the subsequent text processing workload and improves work efficiency.

[0082] When the dictionary-based power field text sentiment analysis method is further implemented, step S1 also includes: constructing a keyword dictionary to collect words that reflect the emotional fluctuations of users based on factors of service providers. Step S2 is followed by step S3: selecting the segmentation words existing in the keyword dictionary from the segmentation words of the target text as the target segmentation words, and outputting the target segmentation words and sentiment polarity of the target text. Specifically, any existing segmentation rule can be used when segmenting the target text.

[0083] Specifically, the keyword dictionary is manually set, and the keywords included in the keyword dictionary include: attitude, hygiene, efficiency, smile, etc.

[0084] In this embodiment, the target word segmentation is filtered based on the keyword dictionary, which realizes the accurate extraction of the factors that cause the user's emotional fluctuations based on the service provider. Combining the target word segmentation and emotional polarity helps the service provider to quickly and accurately extract user opinions, so as to improve the service according to the user's wishes.

[0085] This embodiment also proposes a dictionary-based electric power field text sentiment analysis system, including a memory and a processor, wherein the memory stores a computer program, and the processor is used to implement the above-mentioned dictionary-based electric power field text sentiment analysis method when executing the computer program.

[0086] This embodiment also proposes a storage medium storing a computer program, and when the computer program is executed, it is used to implement the above-mentioned dictionary-based text sentiment analysis method in the power field.

[0087] The above-mentioned dictionary-based text sentiment analysis method in the electric power field is described in detail below in combination with two specific embodiments.

[0088] Example 1

[0089] In this embodiment, the manual customer service communicates with the user by voice. In this embodiment, the target polarity is set to negative.

[0090] Step 1: Use the speech emotion judgment model to judge the emotional extreme value of each speech output by the user. When the emotional extreme value of a certain speech B is a negative value indicating negativity, the text data of the speech B is further obtained as the candidate text B1. Specifically, the speech can be directly converted into text data through the speech-to-text conversion rule, that is, the speech B can be directly converted into text as the candidate text B1.

[0091] Assume B1 = "Your customer service had an argument with my mother today. My mother was very angry. Your service attitude is very bad."

[0092] Step 2: Determine the sentiment orientation of the candidate text B1 through the sentiment orientation judgment model. In this embodiment, the sentiment orientation of the candidate text B1 is the service provider. Therefore, the candidate text B1 is converted into a target text, which is recorded as target text B1 = "Your customer service had a quarrel with my mother today. My mother was very angry. Your service attitude is too bad."

[0093] Step 3: Segment the target text B1 to obtain segmented words “The customer service staff had a quarrel today / / They were all very angry / / The service attitude was too bad”.

[0094] Step 4: Match the obtained segmented words with the keyword dictionary. Assuming that "customer service, quarrel, service, attitude" all exist in the keyword dictionary, output "negative; customer service, quarrel, service, attitude".

[0095] Thus, based on this record, the service provider can quickly learn the reasons for the user's dissatisfaction and make corrections.

[0096] Example 1

[0097] In this embodiment, the robot communicates with the user in text. In this embodiment, the target polarity is set to negative.

[0098] Step 1: Use the method for determining the sentiment polarity of text data provided in this embodiment to perform sentiment polarity analysis on each sentence of text data output by the user.

[0099] Assume that the text data B1 input by the user = "Your customer service had an argument with my mother today. My mother was very angry. Your service attitude is too bad."

[0100] The segmented words corresponding to the text data B1 are "The customer service staff had a quarrel today / / They were all furious / / The service attitude was too bad". Among them, "angry" and "bad" both exist in the sentiment vocabulary dictionary. Taking "angry" and "bad" as emotional words, we get two segments "all-angry-bad" and "too-bad". Assume that the sentiment reference value associated with "angry" in the sentiment vocabulary dictionary is -3, and the sentiment reference value associated with "bad" in the sentiment vocabulary dictionary is -2; "all", "bad" and "too" are all adverbs of degree, and the corresponding weight values ​​are 2, 3, and 4 respectively. Therefore, the total sentiment value corresponding to the text data B1 is (-3×2×3)+(-2×4)=-24, and the total sentiment value of -24 is lower than 0. Therefore, the sentiment polarity of the text data B1 is negative, and the text data B1 is converted into a candidate text, recorded as candidate text B1.

[0101] Step 2: Determine the sentiment orientation of the candidate text B1 through the sentiment orientation judgment model. In this embodiment, the sentiment orientation of the candidate text B1 is the service provider. Therefore, the candidate text B1 is converted into a target text, which is recorded as target text B1 = "Your customer service had a quarrel with my mother today. My mother was very angry. Your service attitude is too bad."

[0102] Step 3: Match the segmented words corresponding to the target text B1 "Customer service had a quarrel today / / Everyone was so angry / / Service attitude was too bad" with the keyword dictionary. Assuming that "customer service, quarrel, service, attitude" all exist in the keyword dictionary, then output "negative; customer service, quarrel, service, attitude".

[0103] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A dictionary-based text sentiment analysis method in the power field, characterized in that: The following steps are involved: S1. Training a sentiment orientation judgment model based on a neural network and manually annotated data, wherein the input of the sentiment orientation judgment model is user text data, and the output of the sentiment orientation judgment model is sentiment orientation; The user text data is text data generated by the user in the process of interacting with the service provider, and the emotional orientation is the object that causes the user's emotions. The emotional orientation includes the user himself and the service provider. The service provider is an organization or an individual, and the service provider who interacts with the user is affiliated with the service provider; S2, obtaining user interaction information, obtaining text data corresponding to the user interaction information as user text data, inputting the user text data into a sentiment orientation judgment model, and obtaining the sentiment orientation output by the sentiment orientation judgment model as the sentiment orientation of the user interaction information; Filter text data corresponding to user interaction information whose sentiment polarity is the set target polarity and whose sentiment is directed to the service provider as target text; The sentiment polarity is divided into positive and negative, and the target polarity contains at least one of positive and negative; Step S1 also includes: constructing a keyword dictionary to collect words that reflect the user's emotional fluctuations based on factors of the service provider; After step S2, the method further includes step S3: selecting the segmentation words existing in the set keyword dictionary from the segmentation words of the target text as the target segmentation words, and outputting the target segmentation words and sentiment polarity of the target text.

2. The dictionary-based text sentiment analysis method in the electric power field according to claim 1, characterized in that: Step S2 specifically includes the following steps: S21, obtaining text data corresponding to user interaction information with a target sentiment polarity as candidate text, and obtaining segmented words of the candidate text; S22. Acquire the sentiment orientation of the candidate text through the sentiment orientation judgment model, and acquire the candidate text whose sentiment orientation is the service provider as the target text.

3. The dictionary-based text sentiment analysis method in the electric power field according to claim 1, characterized in that: The method for determining the user's emotional polarity in step S2 is as follows: obtaining text data corresponding to the user interaction information, and segmenting the text data, and then filtering the segmented words existing in the emotional vocabulary dictionary from the segmented words of the text data based on the set emotional vocabulary dictionary as emotional words, and then combining the emotional words, negative words and degree words to determine the emotional polarity of the text data.

4. The dictionary-based text sentiment analysis method in the electric power field according to claim 3, characterized in that: The method of determining the sentiment polarity of text data specifically includes the following steps: SA21, segment the text data and extract segmented words; SA22, filter the participle words existing in the set emotional vocabulary dictionary as emotional words; SA23, combine each emotion word with the adjacent participle words before and after to form a paragraph; SA24, obtain the sentiment value of the paragraph according to the sentiment polarity of the sentiment words and the negative words and degree words in the paragraph; SA25. Sum the sentiment values ​​of each paragraph in the text data as the total sentiment value of the text data, and determine the sentiment polarity of the text data based on the total sentiment value.

5. The dictionary-based text sentiment analysis method in the electric power field according to claim 4, characterized in that: Each word in the emotional vocabulary dictionary has a corresponding emotional reference value; in step SA25, the calculation formula of the emotional value K of the paragraph is: K=-ak0×(b1×b2×...b n ) k0 represents the emotional reference value corresponding to the emotional word in the emotional vocabulary dictionary, a represents the number of negative words in the paragraph, b i It represents the weight value corresponding to the i-th degree word in the paragraph, 1≤i≤n, n is the number of degree words in the paragraph; the weight value of each degree word is the design value.

6. The method for analyzing text sentiment in the electric power field based on a dictionary according to claim 5, characterized in that: In step SA25, when the total sentiment value is a positive number, the sentiment polarity of the text data is judged to be positive; when the total sentiment value is a negative number, the sentiment polarity of the text data is judged to be negative.

7. The method for analyzing text sentiment in the electric power field based on a dictionary according to claim 1, characterized in that: In step S2, the user interaction information is voice data, and in step S2, the emotional polarity of the user interaction information is obtained based on the set voice emotion judgment model; the voice emotion judgment model is obtained based on neural network and manually labeled data training, the input of the voice emotion judgment model is voice data, and the output is the emotional polarity of the input voice data.

8. A dictionary-based text sentiment analysis system in the power field, characterized by: It comprises a memory and a processor, wherein the memory stores a computer program, and the processor is used to implement the dictionary-based text sentiment analysis method in the electric power field as described in any one of claims 1 to 7 when executing the computer program.

9. A storage medium, characterized in that: A computer program is stored, and when the computer program is executed, it is used to implement the dictionary-based text sentiment analysis method in the power field as described in any one of claims 1 to 7.

Images