CN110096696A - A kind of Chinese long text sentiment analysis method - Google Patents
A kind of Chinese long text sentiment analysis method Download PDFInfo
- Publication number
- CN110096696A CN110096696A CN201810595441.9A CN201810595441A CN110096696A CN 110096696 A CN110096696 A CN 110096696A CN 201810595441 A CN201810595441 A CN 201810595441A CN 110096696 A CN110096696 A CN 110096696A
- Authority
- CN
- China
- Prior art keywords
- word
- sentence
- sentiment
- sentiment analysis
- emotion
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a kind of Chinese long text sentiment analysis methods, which comprises step 1: Text Pretreatment；Step 2: condition random field extracts kernel sentence；Step 3: sentiment analysis is carried out to kernel sentence；Step 4: defining the feeling polarities weight of kernel sentence；Step 5: extending sentiment dictionary used in current sentiment analysis；Step 6: obtaining final sentiment analysis result；Kernel sentence is extracted using condition random field, introduces sentence feeling polarities weight to analyze entire article, improves the accuracy rate of Chinese long text sentiment analysis.
Description
Technical field
The present invention relates to natural language processing fields, and in particular, to a kind of Chinese long text sentiment analysis method.
Background technique
In the past few decades, people's lives increasingly be unable to do without internet, with the development of " internet+", perhaps
More traditional industries and internet combine, internet also in the life style for gradually changing the people, present user from
The taker of preceding simple slave network acquisition information has become the creator of the network information, and people pass through microblogging, circle of friends and opinion
Altar etc. delivers the idea viewpoint of oneself.The growth of index was also presented in the information of internet in recent years.
Often there is huge social and economic benefit in these information, for individual, the viewpoint for understanding other people can be with
Them are helped to make a choice or learn in shopping the knowledge oneself not understood；For businessman or tissue, Ke Yibang
It helps them to understand Market Situation and carrys out timely adjustment strategy.But these massive informations will also result in internet public feelings problem simultaneously.
Social various information is more enriched, the efficiency of propagation is more increased, is covered by the propagation of internet with the approach for comparing propagation in the past
The range of lid is also more comprehensive, if easily socially causing undesirable influence without reasonably monitoring and managing.And
The core of the analysis of public opinion is equally sentiment analysis.
The result of text emotion analysis is generally divided into three kinds, i.e., actively, neutral and passive, can also regard one three classification as
Problem.In order to realize the sentiment analysis of magnanimity internet text, method, machine learning and the nerve that researchers pass through statistics
The methods of network carrys out the emotion tendency of automated analysis text.Text emotion analyzes the research for short texts such as microbloggings at present
It is more, and the research of long text is relatively deficient, and there are many complicated clause different from English for Chinese, so in long text
Difficulty is bigger on sentiment analysis.
The research of sentiment analysis at present is broadly divided into both direction, and one is the sentiment analysis algorithm based on sentiment dictionary,
One is the sentiment analysis algorithm based on machine learning.In the sentiment analysis of sentiment dictionary, the direct shadow of the quality of sentiment dictionary
It rings to final sentiment analysis as a result, so the sentiment dictionary of one high quality of design is most important.In the emotion of machine learning
Analysis is mainly based upon text classification, and choosing word feature, part of speech feature and semantic feature is feature vector, uses support vector machines
Scheduling algorithm carries out sentiment analysis.The core of the sentiment analysis technology of long text is the objective statement for filtering out not emotional expression
Sentence only carries out sentiment analysis to the sentence to show emotion in text, to improve the accuracy and efficiency of sentiment analysis.
Summary of the invention
The purpose of the present invention is overcoming the problems, such as that existing long text sentiment analysis algorithm accuracy rate is not high, provides one kind and is based on
The long text sentiment analysis method of machine learning extracts kernel sentence using condition random field, introduces sentence feeling polarities weight and comes
Entire article is analyzed, the accuracy rate of Chinese long text sentiment analysis is improved.
For achieving the above object, this application provides the present invention is based on the sentiment analysis methods of machine learning, including
Following steps:
(1), Text Pretreatment
To be made pauses in reading unpunctuated ancient writings well to the article of sentiment analysis according to punctuate, then every a word is segmented, filter it is deactivated
Word operation；
(2), condition random field extracts kernel sentence
Use condition random field obtains the evaluation object of entire article, and the sentence comprising the evaluation object is this article
Kernel sentence.
(3), sentiment analysis is carried out to kernel sentence
Using the sentiment analysis algorithm based on sentiment dictionary, all words in the sentence after traversing Text Pretreatment, if deposited
It is in sentiment dictionary, then records corresponding feeling polarities score, finally all scores recorded is added up and are summed, according to
Final score come judge every words feeling polarities, if final score be greater than zero if express positive emotion, if small
Passive emotion is expressed in zero, without apparent emotional expression if being equal to zero, is then defined just according to judging result
Beginning emotion score is+1, -1 and 0.
(4), the feeling polarities weight of kernel sentence is defined
The power that different sentence feeling polarities are distinguished using feeling polarities weight, according to final emotion weight score point
Analyse entire article feeling polarities.
(5), sentiment dictionary used in current sentiment analysis is extended
The emotion word for not including in sentiment dictionary is joined using PMI Mutual Information Rate and Stamford semantic tree, is extended
The use scope of sentiment dictionary.
(6), final sentiment analysis result is obtained
The wherein pretreated specific steps of text in the step (1) are as follows:
(1.1) unified format analysis processing is carried out to the non-structured text of input: removes the non-text of head and the tail of non-structured text
This part obtains plain text part, and if it is empty text is then skipped；
(1.2) made pauses in reading unpunctuated ancient writings to plain text to be analyzed according to fullstop, exclamation mark, ellipsis, question mark, branch；
The word segmentation processing that plain text part is carried out to word part of speech removes the punctuate in word segmentation result for word part of speech
Symbol, onomatopoeia, interjection, auxiliary word, conjunction, preposition, adverbial word, number, quantifier；
Wherein, condition random field extracts kernel sentence method particularly includes:
(2.1), propose that evaluation object marking mode is making by oneself during extracting evaluation object by Ramshaw and Marcus
Justice label label.IOB label label is carried out to the word after participle, is analyzed using Stanford Parser and was segmented above
Sentence afterwards obtains the corresponding semantic tree label of each word.
(2.2) the characteristic function template of condition random field is devised according to word segmentation result, part of speech, semantic structure.
(2.3) the evaluation object word of entire article is counted, that highest word of the frequency of occurrences is just entire article
Evaluation object, the sentence comprising the word are then kernel sentence.If when multiple evaluation object frequency of occurrence highest, these word
It is the evaluation object of this article.
Sentiment analysis wherein is carried out to kernel sentence method particularly includes:
(3.1) to kernel sentence use the sentiment analysis algorithm based on sentiment dictionary, according to sentiment analysis result it is positive, disappear
Pole and the initial emotion score for defining the sentence without obvious emotion are+1, -1 and 0.
Wherein define kernel sentence feeling polarities weight method particularly includes:
(4.1) difference that emotional expression intensity is talked about according to every defines different emotions polarity weight to every words, then gives
The initial emotion score of every words is multiplied by a corresponding weight.
The emotional intensity of emotion word can be reinforced or be weakened to (4.2.1) degree adverb, and this patent devises a degree adverb
Table gives the feeling polarities score of the kernel sentence multiplied by a weight according to different degree adverbs.
(4.2.2) exclamative sentence and interrogative sentence can reinforce the emotional expression of sentence, and this patent devises corresponding auxiliary words of mood
Table, to corresponding kernel sentence multiplied by a weight.
It is that author summarizes the sentence for expressing oneself viewpoint above that (4.2.3), which summarizes sentence, and this patent devises a degree pair
Vocabulary, to corresponding kernel sentence multiplied by a weight.
Wherein extend the specific method of current sentiment dictionary are as follows:
(5.1) word combination for segmenting and removing stop words is input in the resolver of Stamford and carries out syntax parsing, obtained
To the corresponding grammer label of each word.Selection grammer label is adverbial phrase ADVP, Adjective Phrases ADJP, verb phrase VP
And the word of sentiment dictionary is not included in as candidate emotion word.
(5.3) a basic sentiment dictionary is constructed, the Mutual Information Rate of candidate emotion word and basic sentiment dictionary is calculated, meets
Certain threshold value is then added in sentiment dictionary.
(5.3.2) calculates the Mutual Information Rate of each candidate emotion word and basic sentiment dictionary, and calculation formula is
Wherein ciFor candidate emotion word, N is basic sentiment dictionary, riFor the emotion word in basic sentiment dictionary, count
(ci, ri) it is quantity in data bank, it is data bank that internet is chosen in this patent, is existed by a crawler to crawl the combination
Information bar mesh number on internet.The Mutual Information Rate of the word is calculated, it is finally eligible to be added in sentiment dictionary.
Goal of the invention of the invention is achieved in that
The present invention proposes a kind of sentiment analysis method of Chinese long text, obtains every using machine learning condition random field
The evaluation object of words obtains the kernel sentence of entire article according to evaluation object, so carries out sentiment analysis to kernel sentence, and according to every
The difference of word feeling polarities intensity imparts different weights, and final analysis obtains the feeling polarities of entire article.Pass through extraction
Kernel sentence filters out the statement sentence of not emotional expression, has done further division to the emotional intensity of kernel sentence expression, has mentioned
The high accuracy rate of long text sentiment analysis.
A kind of sentiment analysis method of Chinese long text proposed by the present invention simultaneously also has the advantages that
(1), kernel sentence is extracted using condition random field, reduces the quantity for needing sentiment analysis sentence, improves emotion
The efficiency of analysis.
(2), carry out expanding sentiment dictionary using PMI Mutual Information Rate and Stamford semantic tree, increase the emotion of sentiment dictionary
Word quantity improves the recall rate of sentiment analysis.
Detailed description of the invention
Attached drawing described herein is used to provide to further understand the embodiment of the present invention, constitutes one of the application
Point, do not constitute the restriction to the embodiment of the present invention；
Fig. 1 is Chinese long text sentiment analysis method flow diagram；
Fig. 2 is that condition random field extracts kernel sentence flow chart；
Fig. 3 is distich sub-definite emotion weight synthesis flow schematic diagram；
Fig. 4 is sentiment dictionary extension flow diagram.
Specific embodiment
To better understand the objects, features and advantages of the present invention, with reference to the accompanying drawing and specific real
Applying mode, the present invention is further described in detail.It should be noted that in the case where not conflicting mutually, the application's
Feature in embodiment and embodiment can be combined with each other.
In the following description, numerous specific details are set forth in order to facilitate a full understanding of the present invention, still, the present invention may be used also
Implemented with being different from the other modes being described herein in range using other, therefore, protection scope of the present invention is not by under
The limitation of specific embodiment disclosed in face.
Embodiment one:
Fig. 1 is a kind of flow chart of Chinese long text sentiment analysis method of the present invention.
In the present embodiment, as shown in Figure 1, a kind of Chinese long text sentiment analysis method of the present invention, comprising the following steps:
S1, condition random field extract kernel sentence
Conditional random field models are carried out to every words of input and mark evaluation object, statistics entire article frequency of occurrence is most
Word be entire article evaluation object, extract comprising the evaluation object sentence be kernel sentence.To kernel sentence carry out participle with
Remove stop words processing, can take different segmenting methods for different language, in the present embodiment, is used for Chinese
Jieba Chinese word segmentation system carries out participle operation.After obtaining part of speech, removes the punctuation mark in word segmentation result, onomatopoeia, sighs
Word, auxiliary word, conjunction, preposition, adverbial word, number, quantifier.
It is described in detail below with reference to the detailed process that Fig. 2 extracts kernel sentence to condition random field, specific as follows:
S1.1, every words of input are carried out with condition random field extraction evaluation object；
S1.1.1, word segmentation processing is carried out to sentence first, is segmented in the present embodiment using JIeba and carry out participle operation, so
The marking mode of evaluation object is extracted labeled as condition random field using IOB to word segmentation result afterwards, wherein (1) B-EVA, evaluation pair
As adjunctival starts word；(2) I-EVA, evaluation object adjunctival inside word；(3) B-PRO, evaluation object start word
Language；(4) I-PRO, evaluation object inside word；(5) B-ATT, evaluation object association attributes start word；(6) I-ATT, evaluation
Word inside object association attributes；(7) O, other；
Word segmentation result is carried out semantic analysis by S1.1.2, uses Stamford resolver to carry out syntax parsing in the present embodiment,
Obtain corresponding semantic tree label.Corresponding condition random field character modules are designed according to word segmentation result, part of speech and semantic tree label
Plate.
B-PRO and I-PRO is the evaluation object of word in label result.
The number of all evaluation object and appearance, the highest word of frequency of occurrence are in S1.2, statistics entire article
The evaluation object of entire article, the sentence comprising the word are the kernel sentence of this article；
S1.3, the sentiment analysis based on sentiment dictionary then is carried out to kernel sentence.
Emotion word in S1.3.1, traversal sentence, and the cumulative summation of corresponding feeling polarities score is recorded according to sentiment dictionary,
For final result if it is greater than 0, this then expresses positive emotion, and passive emotion is expressed if result is less than 0, if
It as a result is 0, then the sentence does not have apparent emotional expression.
Emotion weight process is defined to sentence below with reference to Fig. 3 to be described in detail, specific as follows:
S2.1, according to kernel sentence sentiment analysis as a result, if expressing positive emotion, define initial emotion score
It is+1, if expressing passive emotion, defining initial emotion score is -1, if without apparent emotional expression, just
Beginning emotion score is 0；
S2.2, according to whether there are degree adverb and clause to corresponding initial emotion score multiplied by a weight；
S2.2.1, emotion word degree adverb table is designed to sentence multiplied by a corresponding weight according to this patent；
Degree adverb is divided into most, more and less three-level by this patent, and corresponding weight is 1.5,1.25 and 0.75.Time
Kernel sentence is gone through, if there is degree adverb then on the initial score of the sentence multiplied by corresponding weight.
S2.3, according to whether for special clause to corresponding initial emotion score multiplied by a weight.
S2.3.1, it is designed according to this patent and sighs with feeling vocabulary to sentence multiplied by a corresponding weight.
This patent has counted common interjection in Chinese and has defined corresponding feeling polarities weight, devises an exclamation
Vocabulary, traverse kernel sentence, if there is on the interjection then initial score of the sentence multiplied by corresponding weight.
S2.3.2, according to this patent summary of Design vocabulary to sentence multiplied by a corresponding weight.
This patent has counted common summary word in Chinese and has defined corresponding feeling polarities weight, devises a summary
Vocabulary traverses kernel sentence, if there is summarizing word then on the initial score of the sentence multiplied by corresponding weight.
S3, final result is analyzed to the cumulative summation of the feeling polarities score polarity for imparting feeling polarities weight.
The feeling polarities score of cumulative all kernel sentences of summation, for final result if it is greater than 0, this article then expresses product
The emotion of pole expresses passive emotion if result is less than 0, if result is 0, this article does not have apparent emotion table
It reaches.
It is described in detail below with reference to detailed process of the Fig. 4 to expanding sentiment dictionary, specific as follows:
S3.1, semantic analysis is carried out to pretreated result herein, is carried out in the present embodiment using Stamford resolver
Semantic analysis chooses candidate emotion word according to semantic label；
Semantic parsing is carried out to kernel sentence using Stamford resolver, according to semantic label, the present embodiment chooses semantic mark
Label be verb phrase VP, Adjective Phrases ADJP and adverbial phrase ADVP and be not present in sentiment dictionary word be candidate feelings
Feel word.
S3.2, the Mutual Information Rate that the word and basic sentiment dictionary are calculated according to the basic sentiment dictionary of this patent；
S3.2.1, the present embodiment devise one and represent positive and Negative Affect basic sentiment dictionary, then calculate every
The Mutual Information Rate of a candidate's emotion word and basic sentiment dictionary, passive Mutual Information Rate calculation formula are
Entry number in S3.2.2, the present embodiment by a crawler come the corresponding word combination of crawler in internet, and
It designs and calculates the positive of the word and basis sentiment dictionary when the entry number of the word and basic sentiment dictionary is greater than 15,000,000
Mutual Information Rate and passive Mutual Information Rate.
S3.2.3, the difference for calculating positive mutual information and passive mutual information, if it is positive emotion word that result, which is canonical, not so
It is then Negative Affect word, emotion word is added by sentiment dictionary according to result.
Although preferred embodiments of the present invention have been described, it is created once a person skilled in the art knows basic
Property concept, then additional changes and modifications may be made to these embodiments.So it includes excellent that the following claims are intended to be interpreted as
It selects embodiment and falls into all change and modification of the scope of the invention.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art
Mind and range.In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to include these modifications and variations.
Claims (7)
1. a kind of Chinese long text sentiment analysis method, which is characterized in that the described method includes:
Step 1: Text Pretreatment: the article to sentiment analysis being made pauses in reading unpunctuated ancient writings according to punctuate number, then to each after punctuate
Word is segmented, filters stop words operation；
Step 2: condition random field extracts kernel sentence: use condition random field obtains the evaluation object of entire article, comments comprising this
The sentence of valence object is the kernel sentence of this article；
Step 3: sentiment analysis being carried out to kernel sentence: using the sentiment analysis algorithm based on sentiment dictionary, analyzes the feelings of every words
Feel polarity, initial emotion score is defined according to judging result；
Step 4: defining the feeling polarities weight of kernel sentence: distinguishing the strong of different sentence feeling polarities using feeling polarities weight
It is weak, according to final emotion weight Fraction analysis entire article feeling polarities；
Step 5: extending sentiment dictionary used in current sentiment analysis: using point Mutual Information Rate and Stamford semantic tree emotion
The emotion word for not including in dictionary joins, the use scope of expanding sentiment dictionary；
Step 6: obtaining final sentiment analysis result.
2. Chinese long text sentiment analysis method according to claim 1, which is characterized in that text is pre- in the step 1
The specific steps of processing are as follows:
Step 1.1: carry out unified format analysis processing to the non-structured text of input: the head and the tail for removing non-structured text are non-textual
Part obtains plain text part, and if it is empty text is then skipped；
Step 1.2: being made pauses in reading unpunctuated ancient writings to plain text to be analyzed according to fullstop, exclamation mark, ellipsis, question mark, branch；
Step 1.3: plain text part being carried out to the word segmentation processing of word part of speech, for word part of speech, is removed in word segmentation result
Punctuation mark, onomatopoeia, interjection, auxiliary word, conjunction, preposition, adverbial word, number, quantifier.
3. Chinese long text sentiment analysis method according to claim 1, which is characterized in that step 2 conditional with
Airport extract kernel sentence specific steps include:
Step 2.1: using evaluation object marking mode to extract the customized label label during evaluation object, to segmenting
Word afterwards carries out evaluation object label label, and the sentence after analyzing participle above using Stamford resolver obtains each
The corresponding semantic tree label of word；
Step 2.2: according to word segmentation result, part of speech, the characteristic function template of semantic structure design condition random field；
Step 2.3: counting the evaluation object word of entire article, that highest word of the frequency of occurrences is the evaluation of entire article
Object, if the sentence comprising the word is then kernel sentence multiple evaluation object frequency of occurrence highests, these words are this
The evaluation object of piece article.
4. Chinese long text sentiment analysis method according to claim 1, which is characterized in that core in the step 3
Sentence carries out the specific steps of sentiment analysis are as follows: the sentiment analysis algorithm based on sentiment dictionary is used to kernel sentence, according to emotion point
The initial emotion score that is positive, passive and defining the sentence without obvious emotion of analysis result is+1, -1 and 0.
5. Chinese long text sentiment analysis method according to claim 1, which is characterized in that definition in the step 4
The specific steps of the feeling polarities weight of kernel sentence are as follows: the difference that emotional expression intensity is talked about according to every, not to every words definition
Feeling of sympathy polarity weight, then to the initial emotion score of every words multiplied by a corresponding weight.
6. Chinese long text sentiment analysis method according to claim 5, which is characterized in that this method is equipped with degree adverb
Table gives the feeling polarities score of the kernel sentence multiplied by a weight according to different degree adverbs；This method is equipped with auxiliary words of mood
Table, to corresponding kernel sentence multiplied by a weight, this method, which is equipped with, summarizes vocabulary, weighs to kernel sentence is summarized accordingly multiplied by one
Value.
7. Chinese long text sentiment analysis method according to claim 1, which is characterized in that extension is worked as in the step 5
The specific steps of sentiment dictionary used in preceding sentiment analysis are as follows:
Step 5.1: the word combination for segmenting and removing stop words being input in the resolver of Stamford and carries out syntax parsing, is obtained
Each corresponding grammer label of word；Selection grammer label be adverbial phrase ADVP, Adjective Phrases ADJP, verb phrase VP simultaneously
And the word of sentiment dictionary is not included in as candidate emotion word；
Step 5.2: building one basic sentiment dictionary calculates the Mutual Information Rate of candidate emotion word and basic sentiment dictionary, meets
Certain threshold value is then added in sentiment dictionary；
Step 5.3: calculating the positive and passive Mutual Information Rate of each candidate emotion word and basic sentiment dictionary, passive Mutual Information Rate
Calculation formula be
Wherein, ciFor candidate emotion word, N is passive basic sentiment dictionary, riFor the emotion word in basic sentiment dictionary, count
(ci, ri) be data bank in quantity.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810595441.9A CN110096696A (en) | 2018-06-11 | 2018-06-11 | A kind of Chinese long text sentiment analysis method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810595441.9A CN110096696A (en) | 2018-06-11 | 2018-06-11 | A kind of Chinese long text sentiment analysis method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110096696A true CN110096696A (en) | 2019-08-06 |
Family
ID=67443510
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810595441.9A Pending CN110096696A (en) | 2018-06-11 | 2018-06-11 | A kind of Chinese long text sentiment analysis method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110096696A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111460158A (en) * | 2020-04-01 | 2020-07-28 | 安徽理工大学 | Microblog topic public emotion prediction method based on emotion analysis |
| CN112464646A (en) * | 2020-11-23 | 2021-03-09 | 中国船舶工业综合技术经济研究院 | Text emotion analysis method for defense intelligence library in national defense field |
| CN113688620A (en) * | 2021-08-26 | 2021-11-23 | 北京阅神智能科技有限公司 | Article emotion analysis method and device |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080249764A1 (en) * | 2007-03-01 | 2008-10-09 | Microsoft Corporation | Smart Sentiment Classifier for Product Reviews |
| CN102663046A (en) * | 2012-03-29 | 2012-09-12 | 中国科学院自动化研究所 | Sentiment analysis method oriented to micro-blog short text |
| CN103646088A (en) * | 2013-12-13 | 2014-03-19 | 合肥工业大学 | Product comment fine-grained emotional element extraction method based on CRFs and SVM |
| CN103678278A (en) * | 2013-12-16 | 2014-03-26 | 中国科学院计算机网络信息中心 | Chinese text emotion recognition method |
| CN104268160A (en) * | 2014-09-05 | 2015-01-07 | 北京理工大学 | Evaluation object extraction method based on domain dictionary and semantic roles |
| CN104731770A (en) * | 2015-03-23 | 2015-06-24 | 中国科学技术大学苏州研究院 | Chinese microblog emotion analysis method based on rules and statistical model |
| CN105005553A (en) * | 2015-06-19 | 2015-10-28 | 四川大学 | Emotional thesaurus based short text emotional tendency analysis method |
| CN106407236A (en) * | 2015-08-03 | 2017-02-15 | 北京众荟信息技术有限公司 | An emotion tendency detection method for comment data |
-
2018
- 2018-06-11 CN CN201810595441.9A patent/CN110096696A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080249764A1 (en) * | 2007-03-01 | 2008-10-09 | Microsoft Corporation | Smart Sentiment Classifier for Product Reviews |
| CN102663046A (en) * | 2012-03-29 | 2012-09-12 | 中国科学院自动化研究所 | Sentiment analysis method oriented to micro-blog short text |
| CN103646088A (en) * | 2013-12-13 | 2014-03-19 | 合肥工业大学 | Product comment fine-grained emotional element extraction method based on CRFs and SVM |
| CN103678278A (en) * | 2013-12-16 | 2014-03-26 | 中国科学院计算机网络信息中心 | Chinese text emotion recognition method |
| CN104268160A (en) * | 2014-09-05 | 2015-01-07 | 北京理工大学 | Evaluation object extraction method based on domain dictionary and semantic roles |
| CN104731770A (en) * | 2015-03-23 | 2015-06-24 | 中国科学技术大学苏州研究院 | Chinese microblog emotion analysis method based on rules and statistical model |
| CN105005553A (en) * | 2015-06-19 | 2015-10-28 | 四川大学 | Emotional thesaurus based short text emotional tendency analysis method |
| CN106407236A (en) * | 2015-08-03 | 2017-02-15 | 北京众荟信息技术有限公司 | An emotion tendency detection method for comment data |
Non-Patent Citations (1)
| Title |
|---|
| 文能: "基于领域本体和CRFS的商品评论倾向型分析", 《中国优秀硕士学位论文全文数据库，I143-27》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111460158A (en) * | 2020-04-01 | 2020-07-28 | 安徽理工大学 | Microblog topic public emotion prediction method based on emotion analysis |
| CN111460158B (en) * | 2020-04-01 | 2022-09-23 | 安徽理工大学 | Microblog topic public emotion prediction method based on emotion analysis |
| CN112464646A (en) * | 2020-11-23 | 2021-03-09 | 中国船舶工业综合技术经济研究院 | Text emotion analysis method for defense intelligence library in national defense field |
| CN113688620A (en) * | 2021-08-26 | 2021-11-23 | 北京阅神智能科技有限公司 | Article emotion analysis method and device |
| CN113688620B (en) * | 2021-08-26 | 2024-03-22 | 北京阅神智能科技有限公司 | Article emotion analysis method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108733653A (en) | A kind of sentiment analysis method of the Skip-gram models based on fusion part of speech and semantic information | |
| CN106484664B (en) | Similarity calculating method between a kind of short text | |
| Ghosh et al. | Fracking sarcasm using neural network | |
| CN108763353B (en) | Baidu encyclopedia relation triple extraction method based on rules and remote supervision | |
| CN104008091B (en) | A kind of network text sentiment analysis method based on emotion value | |
| US10496756B2 (en) | Sentence creation system | |
| CN109376251A (en) | A kind of microblogging Chinese sentiment dictionary construction method based on term vector learning model | |
| CN107025299B (en) | A kind of financial public sentiment cognitive method based on weighting LDA topic models | |
| CN108363687A (en) | Subjective item scores and its construction method, electronic equipment and the storage medium of model | |
| CN101520802A (en) | Question-answer pair quality evaluation method and system | |
| CN105975454A (en) | Chinese word segmentation method and device of webpage text | |
| CN108388554B (en) | Text emotion recognition system based on collaborative filtering attention mechanism | |
| CN106294326B (en) | A kind of news report Sentiment orientation analysis method | |
| CN106547866B (en) | A kind of fine granularity sensibility classification method based on the random co-occurrence network of emotion word | |
| CN109509557B (en) | Chinese electronic medical record information extraction preprocessing method based on big data platform | |
| CN106202584A (en) | A kind of microblog emotional based on standard dictionary and semantic rule analyzes method | |
| CN110263319A (en) | A kind of scholar's viewpoint abstracting method based on web page text | |
| CN106446147A (en) | Emotion analysis method based on structuring features | |
| CN110096696A (en) | A kind of Chinese long text sentiment analysis method | |
| CN107463703A (en) | English social media account number classification method based on information gain | |
| CN111339772B (en) | Russian text emotion analysis method, electronic device and storage medium | |
| CN110399603A (en) | A kind of text-processing technical method and system based on sense-group division | |
| CN106055633A (en) | Chinese microblog subjective and objective sentence classification method | |
| CN107894976A (en) | A kind of mixing language material segmenting method based on Bi LSTM | |
| Mekki et al. | Sentence boundary detection of various forms of Tunisian Arabic |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190806 |
|
| RJ01 | Rejection of invention patent application after publication |