CN106446147A - Emotion analysis method based on structuring features - Google Patents

Abstract

The invention discloses an emotion analysis method based on structuring features. The emotion analysis method includes the steps of collecting Twitter text data; building a Twitter text database; collecting existing emotion polarity value dictionaries; manually establishing related auxiliary dictionaries; preprocessing the Twitter text database; defining an emotion score influence factor, extracting language features of information, and updating the value of the emotion score influence factor every time one language feature is extracted; calculating the emotion polarity values of the Twitter text data through the emotion polarity value dictionaries and the emotion score influence factor. According to the emotion analysis method based on the structuring features, it is avoided that in supervision methods, a large amount of marked data is required to train a classifier, and analysis and generalization are difficult; the CPU processing requirement, the internal storage requirement and the overhead for calculating training time are reduced.

Description

A Sentiment Analysis Method Based on Structural Features

technical field

The invention relates to a sentiment analysis method. In particular, it concerns an unsupervised method for sentiment analysis based on structured features.

Background technique

With the emergence and popularity of social media, more and more users tend to share their unique insights or simply express their emotions and emotions through different social media platforms. Among these social platforms, Twitter has become one of the most popular websites. According to statistics in 2016, it currently has more than 645,000,000 registered users, and the average number of tweets per day exceeds 190,000,000. Through Twitter's API, we can obtain a large amount of rich data, so that we can fully detect and mine these data, which is a good opportunity for sentiment analysis. So as to help us infer the public's views on various things, and use these conclusions to make more informed predictions and choices. Sentiment analysis based on Twitter text data has naturally become a current research hotspot.

Sentiment analysis for Twitter text data mainly involves technologies such as natural language processing, opinion mining and sentiment classification. At present, there are two main methods for implementing sentiment analysis: one is the unsupervised method based on the dictionary, which mainly relies on a sentiment dictionary containing a large number of sentiment polarity information, such as LIWC ^[1] , ANEW ^[2] , AFINN ^[3] , VADER ^[4] , SentiWordNet ^[5] , etc.; the second method is the supervised method, which uses machine learning algorithms to extract features from a large number of labeled data to train classifiers, such as SVM (Support Vector Machine), Bayes, DecisionTree, etc. The most commonly used features are the presence or frequency of n-grams (consecutive 1, 2, 3 or more independent text units in a text). However, this approach requires a large amount of labeled data during the training phase, and thus is computationally expensive in terms of CPU processing, memory requirements, and training time. In addition, for a considerable portion of the data, the decision scores predicted by the supervised classifier are very close to the decision boundary, which implies that the classifier is very uncertain about which class the text belongs to. Therefore, assigning to such data The labels are either completely wrong or right by chance ^[6] . Therefore, here we tend to choose dictionary-based unsupervised methods for sentiment analysis.

The main challenges currently facing the field of sentiment analysis based on Twitter text data are mainly brought about by the characteristics of Twitter text itself: for example, the length of a tweet is limited to 140 characters, which provides us with relatively limited information; Regular language structure and grammatical expression, a tweet may also contain many acronyms, emoticons, hashtags, slang, link addresses, etc., which makes emotion extraction and opinion mining difficult. Existing commonly used traditional natural language processing techniques (Natural Language Preprocessing, NLP) such as word segmentation, standardization, part-of-speech tagging, etc. can be effectively applied to normal written normative texts, but they are no longer applicable to Twitter data.

Contents of the invention

The technical problem to be solved by the present invention is to provide a structural feature-based sentiment analysis method that avoids the need for a large amount of labeled data to train classifiers in supervised methods.

The technical scheme adopted in the present invention is: a kind of sentiment analysis method based on structural feature, comprises the following steps:

1) Collect Twitter text data and build a Twitter text database;

2) Collect existing sentiment polarity value dictionaries, and preferentially select sentiment dictionaries manually generated;

3) Manually establish relevant auxiliary dictionaries, including: standard word dictionary, negative word dictionary, enhanced modifier dictionary, weakened modifier dictionary and Internet slang dictionary;

4) Preprocess the Twitter text database, including:

(1) First, word segmentation is performed on the data in the Twitter text database;

(2) Standardize;

(3) Part-of-Speech Tagging (POS Tagging) on the text;

5) define affective score influence factor, carry out language feature extraction to the information that step 4) preprocessing obtains, described language feature comprises the language feature of word level, the language feature of phrase level and the language feature of sentence level, extracts each Language features will update the value of the affective score impact factor once;

6) Use the emotional polarity value dictionary obtained in step 2) and the emotional score impact factor obtained in step 5) to calculate the emotional polarity value for each Twitter text data.

The sentiment polarity value dictionary in step 2) includes: 3 manually generated sentiment dictionaries AFINN, SentiStrength and VADER, and an automatically generated sentiment dictionary Opinion Observer.

Step 4) The word segmentation described in step (1) is to divide the Twitter text data into the smallest meaningful independent text units, and mark the type of each independent text unit respectively.

Step 4) The standardization described in the step (2) is to utilize the standard English dictionary to change the independent text unit using repeated letters into a standard form, to recognize symbol expressions, picture expressions and text expressions in the Twitter text data, and to judge and label Corresponding emotional polarity.

Step 4) The part-of-speech tagging of the text described in step (3) is to mark the part-of-speech category of each independent text unit.

The definition of affective score impact factor described in step 5) is to introduce an affective score impact factor IF _t for each independent text unit t, wherein IF _t ≥ 0, the initial value is 1, in order to reflect the language features For the extent to which the emotional strength of an independent text unit is enhanced or weakened, the formula of the emotional score influencing factor is as follows:

in Refers to the influence factor of the sentiment score of the independent text unit t after updating, Refers to the sentiment score influencing factor of the independent text unit t before updating, p refers to a certain feature, and P refers to all feature sets that can affect the sentiment score influencing factor.

In step 5), the language feature extraction of word level includes:

If the letters in an independent text unit are all capitalized, then the all capitalized flag s ^AllCaps = 1, otherwise s ^AllCaps = 0, and update the emotional score influencing factor formula IF _t :

in Refers to the influence factor of the sentiment score of the independent text unit t after updating, Refers to the influence factor of the sentiment score of the independent text unit t before updating;

If an individual text unit uses repeated letters, assign an elongation factor to each individual text unit middle t _orig represents the original independent text unit, t _norm represents the normalized independent text unit, and updates the emotional score impact factor formula IF _t :

In step 5), the language feature extraction of phrase level includes:

Using the negative word dictionary created manually in step 3), determine the beginning of phrases containing negative content, determine period, question mark, exclamation mark and non-standard independent text units as the end symbols of phrases containing negative content, and update the sentiment score impact factor formula IF _t :

where t is an independent text unit within the negated range.

Use the enhanced modifier dictionary and weakened modifier dictionary manually established in step 3) to find out all the modifiers in the Twitter text data The stretch factor for modifiers is calculated according to the following formula:

Among them, m represents a modifier, M ^DM represents a set of modifiers, if the letters of a modifier m are all capitalized, the modifiers are all capitalized otherwise If a modifier m uses repeated letters, use the repeated letter flag otherwise

And update the emotional score impact factor formula IF _t :

In step 5), the language feature extraction of sentence level includes:

Through the sentence structure of X but Y, determine the Twitter text data using turning conjunctions (such as but, yet, etc.), mark the part X before the conjunction and the part Y after the conjunction, and update the emotional score impact factor formula IF _t :

where, if the independent text unit t is in X, then If in Y, then

Through the following three sentence structures: If X, Y, If X(,) then Y, and Y, if X., determine the Twitter text data using conditional sentences, where X is a conditional sentence and Y is a result sentence in the sentence structure, and update Emotional score impact factor formula IF _t :

where the independent text unit t is in X.

Step 6) described computing emotion polarity value, comprises:

(1) Calculate the basic sentiment polarity value of each independent text unit t, let L be the dictionary set of sentiment polarity value used, L _t = {l∈L _t |t∈l} means the sentiment including the independent text unit t A subset of the dictionary, the basic emotional polarity value s _t of each independent text unit t is obtained by the following formula:

Where score(l, t) is the basic emotional polarity value of each independent text unit t given in the dictionary l, |L _t | indicates the number of emotional polarity value dictionaries containing independent text unit t;

(2) For each independent text unit t, update the basic emotional polarity value s _t of each independent text unit t by using the emotional score influence factor IF _t obtained in step 5):

(3) Calculate the overall emotional score S _{T for each Twitter text data T} :

A sentiment analysis method based on structured features of the present invention avoids the need for a large amount of labeled data to train classifiers in supervised methods, which is difficult to analyze and generalize, and reduces CPU processing, memory requirements and training time calculations overhead. The beneficial effects of the present invention are specifically:

1. Avoid the use of supervised methods based on machine learning, and do not need to rely on a large amount of labeled data to train classifiers to achieve sentiment analysis;

2. Adopting a fine emotional perception preprocessor, which can effectively process informal social media text information, and improve the efficiency of subsequent processing and classification accuracy;

3. A structured feature extraction mode is proposed, which can easily update the impact factors of the emotional score we defined, and then improve the calculation process of the emotional score.

Description of drawings

Fig. 1 is a flow chart of the sentiment analysis method based on structured features of the present invention.

detailed description

A structural feature-based sentiment analysis method of the present invention will be described in detail below with reference to the embodiments and the accompanying drawings.

As shown in Figure 1, a kind of sentiment analysis method based on structured feature of the present invention, comprises the following steps:

1) Collect Twitter text data and build a Twitter text database;

2) Collect existing emotional polarity value dictionaries, and preferentially select emotional dictionaries manually generated; the emotional polarity value dictionaries include: 3 manually generated emotional dictionaries AFINN, SentiStrength and VADER, and an automatically generated The Sentiment Dictionary Opinion Observer. Table 1 gives an overview of the sentiment polarity value dictionary and its characteristics.

Table 1 Overview of sentiment lexicon

3) Manually build relevant auxiliary dictionaries, including: standard word dictionary, negative word dictionary, enhanced modifier dictionary, weakened modifier dictionary and Internet slang dictionary; Table 2 gives a summary of the dictionaries we use.

Table 2 Auxiliary dictionary overview

4) Preprocess the Twitter text database, including:

(1) First, segment the data in the Twitter text database. The word segmentation is to divide the Twitter text data into the smallest meaningful independent text units, and at the same time mark the type of each independent text unit, such as words, hashtags, emoticons, link addresses, etc. Use regular expressions to match different types of independent text units and label them accordingly.

(2) Standardize. The standardization is to use a standard English dictionary to change independent text units that use repeated letters into a standard form, to identify symbolic expressions, picture expressions and text expressions in Twitter text data, and to judge and mark the corresponding emotional polarity. details as follows:

a. Letter elongation. Letter elongation refers to the use of repeated letters to increase the expression of words. First, the index of the word is found from D ^en and D ^slang based on the phonetic code. If our normalization processor encounters a word that does not exist in this Words in the two dictionaries are then phonetically encoded to identify matching options, then the Levenshtein distance between the input and each matching option is calculated to measure their similarity, and the best match is returned.

b. Symbolic expressions. Symbolic expressions are graphical representations of facial expressions composed of punctuation marks or letters such as :-), :), :o), etc. We normalize positive and negative symbolic expressions as [EMOTICON+] and [ EMOTICON-].

c. Picture expression (emoji). Since 2010, more and more picture emoticons have been added to the standard Unicode (UNICODE)-Unicode 8.0, such as Similar to symbolic emoticons, we normalize all image emoticons into predefined independent text units s, such as [EMOJI+], [EMOJI0], [EMOJI-].

d. Text expressions (emotext). Finally, we standardize text expressions such as haha, hehe, xixi, and we find these text expressions by matching regular expressions that contain at least k repeated letters (currently set k=2), and then Normalizes each text expression to its core form, such as changing hhahahah to haha.

(3) Perform Part-of-Speech Tagging (POS Tagging) on the text. The part-of-speech tagging of the text is to tag the part-of-speech category of each independent text unit, such as nouns, adjectives, and verbs.

5) define affective score influence factor, carry out language feature extraction to the information that step 4) preprocessing obtains, described language feature comprises the language feature of word level, the language feature of phrase level and the language feature of sentence level, extracts each Language features just update the numerical value of the emotional score impact factor once; the definition of the emotional score impact factor is to introduce an emotional score impact factor IF _t for each independent text unit t, where IF _t ≥ 0, the initial value is 1. To reflect the extent to which the language features enhance or weaken the emotional strength of an independent text unit, the formula for the emotional score influencing factor is as follows:

in Refers to the influence factor of the sentiment score of the independent text unit t after updating, Refers to the sentiment score influencing factor of the independent text unit t before updating, p refers to a certain feature, and P refers to all feature sets that can affect the sentiment score influencing factor.

Language feature extraction at the word level includes:

If the letters in an independent text unit are all capitalized, then the all capitalized flag s ^AllCaps = 1, otherwise s ^AllCaps = 0, and update the emotional score influencing factor formula IF _t :

in Refers to the influence factor of the sentiment score of the independent text unit t after updating, Refers to the influence factor of the sentiment score of the independent text unit t before updating.

If an individual text unit uses repeated letters, assign an elongation factor to each individual text unit middle t _orig represents the original independent text unit, t _norm represents the normalized independent text unit, and updates the emotional score impact factor formula IF _t :

in Refers to the influence factor of the sentiment score of the independent text unit t after updating, Refers to the influence factor of the sentiment score of the independent text unit t before updating.

Phrase-level language feature extraction includes:

Using the negative word dictionary created manually in step 3), determine the beginning of phrases containing negative content, determine period, question mark, exclamation mark and non-standard independent text units as the end symbols of phrases containing negative content, and update the sentiment score impact factor formula IF _t :

where t is an independent text unit within the negated range, Refers to the influence factor of the sentiment score of the independent text unit t after updating, Refers to the influence factor of the sentiment score of the independent text unit t before updating.

Use the enhanced modifier dictionary and weakened modifier dictionary manually established in step 3) to find out all the modifiers in the Twitter text data The stretch factor for modifiers is calculated according to the following formula:

Among them, m represents a modifier, M ^DM represents a set of modifiers, if the letters of a modifier m are all capitalized, the modifiers are all capitalized otherwise If a modifier m uses repeated letters, use the repeated letter flag otherwise

And update the emotional score impact factor formula IF _t :

in Refers to the influence factor of the sentiment score of the independent text unit t after updating, Refers to the influence factor of the sentiment score of the independent text unit t before updating.

The language feature extraction at the sentence level includes:

Through the sentence structure of X but Y, determine the Twitter text data using turning conjunctions (such as but, yet, etc.), mark the part X before the conjunction and the part Y after the conjunction, and update the emotional score impact factor formula IF _t :

where, if the independent text unit t is in X, then If in Y, then Refers to the influence factor of the sentiment score of the independent text unit t after updating, Refers to the influence factor of the sentiment score of the independent text unit t before updating.

Through the following three sentence structures: If X, Y, If X(,) then Y, and Y, if X., determine the Twitter text data using conditional sentences, where X is a conditional sentence and Y is a result sentence in the sentence structure, and update Emotional score impact factor formula IF _t :

where the independent text unit t is in X.

6) Use the emotional polarity value dictionary obtained in step 2) and the emotional score impact factor obtained in step 5) to calculate the emotional polarity value for each Twitter text data. The calculation of emotional polarity value includes:

(1) Calculate the basic sentiment polarity value of each independent text unit t, let L be the dictionary set of sentiment polarity value used, L _t = {l∈L _t |t∈l} means the sentiment including the independent text unit t A subset of the dictionary, the basic emotional polarity value s _t of each independent text unit t is obtained by the following formula:

Where score(l, t) is the basic sentiment polarity value of each independent text unit t given in dictionary l, |L _t | represents the number of sentiment polarity value dictionaries containing independent text unit t.

(2) For each independent text unit t, update the basic emotional polarity value s _t of each independent text unit t by using the emotional score influence factor IF _t obtained in step 5):

(3) Calculate the overall emotional score S _{T for each Twitter text data T} :

Those skilled in the art can understand that the accompanying drawing is only a schematic diagram of a preferred embodiment, and the serial numbers of the above-mentioned embodiments of the present invention are for description only, and do not represent the advantages and disadvantages of the embodiments.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within range.

References in the background technology are as follows:

[1] Pennebaker J W, Francis M E, Booth R J. Linguistic inquiry and wordcount: LIWC 2001 [J]. Mahway: Lawrence Erlbaum Associates, 2001, 71: 2001..

[2] Bradley M M, Lang P J. Affective norms for English words (ANEW): Instruction manual and effective ratings [R]. Technical report C-1, the center for research in psychophysiology, University of Florida, 1999.

[3]Nielsen F A new ANEW:Evaluation of a word list for sentiment analysis in microblogs[J].arXiv preprint arXiv:1103.2903,2011.

[4]Hutto C J, Gilbert E.Vader: A parsimonious rule-based model forsentiment analysis of social media text[C]//Eighth International AAAIConference on Weblogs and Social Media.2014.

[5] Baccianella S, Esuli A, Sebastiani F. SentiWordNet 3.0: An Enhanced Lexical Resource for Sentiment Analysis and Opinion Mining [C]//LREC.2010,10:2200-2204.

[6] Chikersal P, Poria S, Cambria E. SeNTU: sentiment analysis of tweets by combining a rule-based classifier with supervised learning [J]. SemEval-2015, 2015: 647.

Claims (10)

1. A sentiment analysis method based on structured features, characterized in that, comprising the steps: 1) Collect Twitter text data and build a Twitter text database; 2) Collect existing sentiment polarity value dictionaries, and preferentially select sentiment dictionaries manually generated; 3) Manually establish relevant auxiliary dictionaries, including: standard word dictionary, negative word dictionary, enhanced modifier dictionary, weakened modifier dictionary and Internet slang dictionary; 4) Preprocess the Twitter text database, including: (1) First, word segmentation is performed on the data in the Twitter text database; (2) Standardize; (3) Part-of-Speech Tagging (POS Tagging) on the text; 5) define affective score influence factor, carry out language feature extraction to the information that step 4) preprocessing obtains, described language feature comprises the language feature of word level, the language feature of phrase level and the language feature of sentence level, extracts each Language features will update the value of the affective score impact factor once; 6) Use the emotional polarity value dictionary obtained in step 2) and the emotional score impact factor obtained in step 5) to calculate the emotional polarity value for each Twitter text data. 2. a kind of sentiment analysis method based on structured feature according to claim 1, it is characterized in that, step 2) described emotion polarity value dictionary comprises: the emotion dictionary AFINN, SentiStrength and VADER that 3 artificial manually generate , and an automatically generated sentiment dictionary Opinion Observer. 3. a kind of sentiment analysis method based on structured feature according to claim 1, is characterized in that, the participle described in step 4) (1) step is that Twitter text data is segmented into minimum meaningful independent text unit, while marking the type of each individual text unit separately. 4. a kind of sentiment analysis method based on structured feature according to claim 1, it is characterized in that, the standardization described in the step 4) (2) step is to utilize the independent text unit of repeated letter by utilizing standard English dictionary Change to a standard form to recognize symbolic emoticons, picture emoticons, and text emoticons in Twitter text data, and judge and label the corresponding emotional polarity. 5. a kind of sentiment analysis method based on structured feature according to claim 1, is characterized in that, step 4) described in (3) step is carried out part-of-speech tagging to text, is to mark the part-of-speech of each independent text unit category. 6. a kind of sentiment analysis method based on structured feature according to claim 1, it is characterized in that, step 5) described definition emotion score influence factor is to introduce an emotion score for each independent text unit t The impact factor IF _t , where IF _t ≥ 0, with an initial value of 1, is used to reflect the degree to which the language features enhance or weaken the emotional strength of an independent text unit. The formula for the emotional score impact factor is as follows: ${\mathrm{<span\; class="notranslate">\; IF</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; IF</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 10</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$ in Refers to the influence factor of the sentiment score of the independent text unit t after updating, Refers to the sentiment score influencing factor of the independent text unit t before updating, p refers to a certain feature, and P refers to all feature sets that can affect the sentiment score influencing factor. 7. a kind of sentiment analysis method based on structured feature according to claim 1, is characterized in that, step 5) in the language feature extraction of word level comprises: If the letters in an independent text unit are all capitalized, then the all capitalized flag s ^AllCaps = 1, otherwise s ^AllCaps = 0, and update the emotional score influencing factor formula IF _t : ${\mathrm{<span\; class="notranslate">\; IF</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; IF</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 10</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; C</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; p</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$ in Refers to the influence factor of the sentiment score of the independent text unit t after updating, Refers to the influence factor of the sentiment score of the independent text unit t before updating; If an individual text unit uses repeated letters, assign an elongation factor to each individual text unit middle t _orig represents the original independent text unit, t _norm represents the normalized independent text unit, and updates the emotional score impact factor formula IF _t : ${\mathrm{<span\; class="notranslate">\; IF</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; IF</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 10</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; E.</span>}\mathrm{<span\; class="notranslate">\; x</span>}\mathrm{<span\; class="notranslate">\; L</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; .</span>$ 8. a kind of sentiment analysis method based on structured feature according to claim 1, is characterized in that, step 5) in the language feature extraction of phrase level comprises: Using the negative word dictionary created manually in step 3), determine the beginning of phrases containing negative content, determine period, question mark, exclamation mark and non-standard independent text units as the end symbols of phrases containing negative content, and update the sentiment score impact factor formula IF _t : ${\mathrm{<span\; class="notranslate">\; IF</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; IF</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$ where t is an independent text unit within the negated range; Use the enhanced modifier dictionary and weakened modifier dictionary manually established in step 3) to find out all the modifiers in the Twitter text data The stretch factor for modifiers is calculated according to the following formula: ${\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; m</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{\mathrm{<span\; class="notranslate">\; A</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; C</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; p</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{\mathrm{<span\; class="notranslate">\; E.</span>}\mathrm{<span\; class="notranslate">\; x</span>}\mathrm{<span\; class="notranslate">\; L</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$ Among them, m represents a modifier, M ^DM represents a set of modifiers, if the letters of a modifier m are all capitalized, the modifiers are all capitalized otherwise If a modifier m uses repeated letters, use the repeated letter flag otherwise And update the emotional score impact factor formula IF _t : ${\mathrm{<span\; class="notranslate">\; IF</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; IF</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 10</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; D.</span>}\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; .</span>$ 9. a kind of sentiment analysis method based on structured feature according to claim 1, is characterized in that, step 5) in the language feature extraction of sentence level comprises: Through the sentence structure of X but Y, determine the Twitter text data using turning conjunctions (such as but, yet, etc.), mark the part X before the conjunction and the part Y after the conjunction, and update the emotional score impact factor formula IF _t : ${\mathrm{<span\; class="notranslate">\; IF</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; IF</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 10</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; sgn</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; C</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; N</span>}\mathrm{<span\; class="notranslate">\; J</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$ where, if the independent text unit t is in X, then If in Y, then Through the following three sentence structures: If X, Y, If X(,) then Y, and Y, if X., determine the Twitter text data using conditional sentences, where X is a conditional sentence and Y is a result sentence in the sentence structure, and update Emotional score impact factor formula IF _t : ${\mathrm{<span\; class="notranslate">\; IF</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$ where the independent text unit t is in X. 10. a kind of sentiment analysis method based on structured feature according to claim 1, is characterized in that, step 6) described computing emotion polarity value, comprises: (1) Calculate the basic sentiment polarity value of each independent text unit t, let L be the dictionary set of sentiment polarity value used, L _t = {l∈L _t |t∈l} means the sentiment including the independent text unit t A subset of the dictionary, the basic emotional polarity value s _t of each independent text unit t is obtained by the following formula: ${\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\frac{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}}\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; c</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; |</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; \&NotEqual;</span>\mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 9</span>}<span\; class="notranslate">\; )</span>$ Where score(l, t) is the basic emotional polarity value of each independent text unit t given in the dictionary l, |L _t | indicates the number of emotional polarity value dictionaries containing independent text unit t; (2) For each independent text unit t, update the basic emotional polarity value s _t of each independent text unit t by using the emotional score influence factor IF _t obtained in step 5): ${\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; IF</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ;</span>$ (3) Calculate the overall emotional score S _{T for each Twitter text data T} : S _T =∑ _t∈T s _t (11).