CN109086340A - Evaluation object recognition methods based on semantic feature - Google Patents
Evaluation object recognition methods based on semantic feature Download PDFInfo
- Publication number
- CN109086340A CN109086340A CN201810748969.5A CN201810748969A CN109086340A CN 109086340 A CN109086340 A CN 109086340A CN 201810748969 A CN201810748969 A CN 201810748969A CN 109086340 A CN109086340 A CN 109086340A
- Authority
- CN
- China
- Prior art keywords
- feature
- evaluation object
- semantic
- word
- corpus
- 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
Landscapes
- Machine Translation (AREA)
Abstract
Evaluation object recognition methods based on semantic feature, belong to text mining field, it is characterized by: the problem of the problem of short text evaluation object is extracted is converted into information extraction, first short text corpus is pre-processed, user-defined feature template, corresponding feature is extracted in short text according to template, carries out evaluation object identification using conditional random field models.The present invention not only allows for lexical characteristics, is also introduced into semantic feature, takes full advantage of contextual information；Present invention introduces semantic features, and bluebeard compound feature improves the accuracy of evaluation object extraction.
Description
Technical field
The present invention relates to a kind of evaluation object recognition methods for being based on semantic feature (Semantic features), belong to
Text mining field.
Background technique
In recent years, with the popularity of the internet and the fast development of e-commerce, more and more consumers are liked
The commodity that online purchase oneself is liked, in order to express oneself for the attitude of commodity, consumer can deliver on electric business platform and comment
By this also causes network comment to sharply increase, and forms commercially valuable large data sets.Utilize natural language processing technique
The emotion for including in these network comment data sets is excavated, sentiment analysis is carried out, there is certain finger for businessman and consumer
Lead meaning.It deepens continuously along with the research of sentiment analysis, fine-grained sentiment analysis is also of interest by more and more scholars.
Fine granularity sentiment analysis is exactly the emotional orientation analysis based on evaluation unit, and so-called evaluation unit is exactly by evaluation object
With the feature viewpoint pair of word or phrase composition with emotional color.Therefore the identification of evaluation object is fine granularity sentiment analysis
Basis.
Evaluation object is generally a sentence institute main topic of discussion, and the modification of evaluating word is mainly shown as in comment text
Object.Evaluation object identification is proposed by Liu Bing earliest, and the noun for having upper frequency or nominal phrase are considered as
Evaluation object.The research of evaluation object identification, the method for being generally based on rule/template in the early stage, the rule of formulation includes word
The forms such as sequence rules, part-of-speech rule, syntactic rule.Kim etc. understands the semantic relation between word and word by the template of formulation,
So that semantic role be mirrored in frame, using newsletter archive as data set, identify viewpoint in text, viewpoint holder,
And related subject.In addition to this, some scholars find out the candidate evaluations pair frequently occurred using the method for association rule mining
As then using wrong sample in the method removal candidate target of two kinds of beta prunings.
With the rise of topic model, more and more scholars apply it to sentiment analysis field, and due to evaluation
Object is generally main topic of discussion in sentence, therefore topic model can be used to carry out the identification of evaluation object.There is scholar to make
The evaluation object in comment on commodity text is excavated with the topic model of more granularities, and zhang etc. is on this basis to similar
Evaluation object is clustered, and proposes the recall rate for having changed identification to a certain extent.Then, some scholars are by topic model and maximum entropy
Model combines, identify comment in evaluation object, evaluating word.
In recent years, as conditional random field models are at the natural languages such as Chinese word segmentation, part-of-speech tagging, name Entity recognition
The progress obtained in reason task has scholar to start to propose to be applied in evaluation object identification.The it is proposeds such as Niklas Jako will
Evaluation object identification mission sees the task of marking, and carries out evaluation object identification using single features, single template.Domestic some
The characteristics of person is according to Chinese text proposes evaluation object based on syntactic structure or based on part of speech feature in conjunction with syntactic analysis
Contextual information is preferably utilized in recognition methods.Ge Wang etc. has been carried out using conditional random field models and word feature across neck
The evaluation object in domain identifies.
Summary of the invention
The invention aims to can make full use of the contextual information of short text to improve the precision that evaluation object extracts,
Information extraction problem is converted by the identification problem of short text evaluation object, introduces the semantic feature of short text sentence to introduce
More contextual informations realize a kind of evaluation object recognition methods based on semantic feature.
Evaluation object recognition methods based on semantic feature, it is characterised in that: first comment corpus is pre-processed；It makes by oneself
Adopted feature templates extract corresponding feature in comment corpus according to template；Evaluation object is carried out using conditional random field models
Identification.
The evaluation object recognition methods includes the following steps:
(1) data prediction: corpus of text is obtained in the online comment of electric business website by web crawlers, to comment
Corpus is segmented, part-of-speech tagging, emotion word mark and semantic role are analyzed, and obtains word feature, part of speech from comment corpus
Feature, emotion word feature, semantic role feature；
(2) training pattern: it is used as training set by extracting 4/5ths in the comment corpus of data prediction, from training set
The feature of extraction trains customized feature templates, according to training result obtains the weight of each feature；
(3) evaluation object identifies: i.e. using remaining 1/5th comments corpus as test set, being obtained according to step (2)
Feature templates, corresponding with template word feature, part of speech feature, emotion word feature and semantic role are extracted in test set
Feature, then using conditional random field models design conditions probability P (y | x), x is list entries, and y is output sequence, will be tested
The feature of concentration obtains optimum as output sequence as list entries, evaluation object, evaluating word from list entries.
Semantic feature is introduced, the semantic feature includes emotion word feature, semantic role feature, is estimated by emotion word feature
The position of evaluation object is counted, and actor and word denoting the receiver of an action person in sentence are captured by semantic role feature.
By introducing semantic feature, including emotion word feature, semantic role feature, to make full use of short text sentence
Contextual information.
Example sentence (1): " cell phone appearance sees that delivery is quickly very well."
Emotion word feature (sen): referring to and judge whether current word belongs to emotion word, right if being currently emotion word
The feature answered is 1, is otherwise 0, belongs to Boolean type feature.Emotion word is to contain commentator for modifying evaluation object
To the attitude of commodity, can very likely there be evaluation object near emotion word in a sentence, reflect to a certain extent
Therefore the position feature of evaluation object introduces the identification that emotion word feature is conducive to evaluation object.Such as: in example sentence (1)
" good-looking ", " fast " belong to emotion word.
Semantic role feature (srl): referring to and carry out semantic character labeling (SRL) to each ingredient in sentence, semantic role
Mark is a kind of semantic analysis technology of shallow-layer, marks the opinion that certain phrases in sentence are given predicate (verb, adjective etc.)
First (i.e. semantic role), such as actor, the word denoting the receiver of an action person, the timing, location and method of event.Still language is carried out to example sentence (1)
Adopted role analysis, analysis result is as shown in Figure 1, wherein there are two predicate " good-looking ", " fast ", and by taking " good-looking " as an example, " very " is it
Degree or mode (generally being indicated with ADV), and the noun phrase of " cell phone appearance " composition is its actor (generally with A0 table
Show), that is, the evaluation object that the sentence unit includes.Therefore, semantic role analysis facilitates the extraction of information, to evaluation pair
The identification of elephant has certain directive significance, and the identification of evaluation object will be may consequently contribute to by introducing semantic feature.
Present invention introduces semantic features, and bluebeard compound feature improves the accuracy of evaluation object extraction.
Detailed description of the invention
Fig. 1 is the flow chart of the evaluation object recognition methods based on semantic feature.
Specific embodiment
Evaluation object identifying system based on semantic feature includes data prediction, training pattern and identification three phases.
Corpus is tested from web crawlers, is to have carried out crawling for mobile phone comment from major electric business website, it is collected
One shares original word number 245221407 in comment on commodity, altogether 4904600 comments, of different Chinese characters occurs
Number is 32757.Training corpus, which takes, crawls 4/5ths of corpus, and remaining corpus is as testing material.
Step 1: pre-processing to experiment corpus, pretreatment stage includes the following aspects:
1, it segments, is the sequence labelling problem based on word sequence by participle task modeling.For inputting the word sequence of sentence,
Model marks the label on a mark word boundary to each word in sentence, by calling in language technology platform (LTP)
Ltp_test main program, detailed step are as follows:
1) language technology platform (LTP) item file, model file downloading；
2) create project folder D: myprojects LTP (can be optionally)；Ltp_ after model file is decompressed
Data file is put into project folder；Dll, exe file after item file is decompressed all copies to project folder
Under, last complete listed files；
3) using python routine call in project folder configured ltp_test, in a program setting ginseng
The option of number last_stage is ws (participle), completes the participle task to corpus of text.
2, part-of-speech tagging is similar to participle task, is word-based sequence labelling problem by part-of-speech tagging task modeling,
When carrying out part-of-speech tagging, method that reference pair corpus is segmented, it is only necessary in a program by parameter last_stage's
Option is set as pos (part-of-speech tagging), and part-of-speech tagging task can be completed.
3, emotion word marks, and after participle, part-of-speech tagging, is manually marked to the emotion word occurred in text, such as
The fruit word belongs to emotion word and is then designated as 1, is otherwise designated as 0, emotion word includes adjective emotion word and verb emotion word.
4, semantic role feature is extracted, and the extraction of semantic role is broadly divided into two subtasks, one is predicate is (generally
Verb, adjective) identification, secondly be exactly argument (semantic role, such as actor A0, word denoting the receiver of an action person A1) identification and divide
Class.Sentence progress syntactic analysis is obtained into syntax tree, and beta pruning is carried out to syntax tree, removes the mark that can not become semantic role
Unit is infused, the argument selected then is given using classifier, is finally classified to argument, classification is all tag along sorts, is returned
The data returned are xml format, therefrom extract semantic role feature.Semantic role list of labels such as table 1:
1 semantic role list of labels of table
Step 2: training set marks, it is to the evaluation object (CO) in the comment text after pretreatment, evaluating word
(CC) and other words (OT) are labeled, would generally will be some nominal short according to the granularity of participle, when participle
Language is separated, such as: " cell phone appearance " in mobile phone comment is " mobile phone ", " appearance " after participle, in hotel's comment
The nominal phrase of this type such as " administrative between " and " big bed room " be respectively after participle " administration ", " " and " big bed ",
" room ".When sequence labelling, it is intended that the evaluation object of this type is marked into entire phrase, rather than by two
Word marks respectively, therefore we are using BIO mark system herein, and BIO mark is to be labeled as each element " B-X ", and generation refers to X
Beginning, " I-X ", generation refers to the centre of X, " O ", and generation refers to other words or punctuation mark, and mark example is as shown in table 1:
Table 1 marks example
Step 3: user-defined feature template, in order to find optimal template window, we define 7 kinds of template temp1-
Temp7 difference is as follows:
Temp1=(- 1,0) indicates centered on current word, considers the previous word of the word, window size 2.
Temp2=(- 1,0,1) is indicated centered on current word, considers the previous word and the latter word of the word, window
Size is 3.
Temp3=(0,1) indicates centered on current word, considers the latter word of the word, window size 2.
Temp4=(- 2, -1,0) indicates centered on current word, considers the first two words of the word, window size 3.
Temp5=(- 2, -1,0,1,2) is indicated centered on current word, considers the first two words and latter two word of the word,
Window size is 5.
Temp6=(0,1,2) indicates centered on current word, considers latter two word of the word, window size 3.
Temp7=(- 3, -2, -1,0,1,2,3) indicates centered on current word, considers first three word and rear three of the word
A word, window size 7.
The meaning of template is as shown in table 2:
2 feature templates meaning of table
The meaning of U01 is the 0th column, the 1st row in table 2；The meaning of U02 is the 0th column, the 2nd row, other several principle classes
Seemingly, and so on.
Step 4: being trained using training corpus to feature templates, the weight parameter of correlated characteristic is obtained.Condition random
One main task of field model is exactly that the weight λ of feature is estimated by training dataset, and logarithm maximum likelihood parameter
λ=(λ is estimated from stand-alone training data1,λ2,...,λn), wherein λiIt is to be estimated to obtain by L-BFGS method.
Assuming that known training set D={ (X1,Y1),(X2,Y2),...,(XΓ,YΓ), it is used according to maximum entropy model maximum
The method of likelihood estimates parameter, then, for conditional probability model p=(y | x, λ), the log-likelihood function of training set D is
Shown in formula (1):
It is the experienced probability distribution of training set sample, the formula of conditional probability is shown in formula (2):
Wherein Z (x) is normalization factor.Therefore, the probability of experience distribution and the mathematic expectaion of conditional probability pass through condition
Random field models acquire, shown in expression formula such as formula (3), (4):
According to logarithm maximum likelihood function, by seeking the available corresponding parameter of its first derivative, L-BFGS method is than passing
Iteration method of scales, the gradient descent method of system are more efficient.This method can be regarded as flight data recorder optimization program it is only necessary to
The first derivative for wanting majorized function is provided, shown in expression formula such as formula (5):
According to maximum entropy model principle, the feature distribution expectation of conditional probability model is equal to the expectation of experience distribution, then
The problem of parameter Estimation, can be solved by optimization method.
In description above, the gradient calculation expression of log-likelihood function L (λ) is given, i.e. experience is distributedMathematic expectaion subtract the mathematic expectaion of conditional probability p (y | λ, x), which is to pass through conditional random field models
It obtains.The mathematic expectaion of experience distribution is that training data concentrates the quantity for meeting the stochastic variable (x, y) of feature constraint, and condition is general
The mathematic expectaion of rate is substantially to calculate p (y | λ, x).
Step 4: evaluation object identifies, test set is equally pre-processed, is marked, the method for pre-processing and marking is same
Training set, in order to guarantee its consistency, the correlated characteristic of training corpus text be extracted, in conjunction with the trained mould of step 3
Plate introduces the identification that conditional random field models carry out evaluation object, for the sequence X (X of input1,X2,...,Xn), output sequence
Y(Y1,Y2,...,Yn) probability calculation formula such as shown in (6):
Z (x) is normalization factor herein, it be all Y shape probability of states and, use Z (x) as denominator, in this way may be used
The length of list entries X, X are indicated to ensure required probability less than 1, niIndicate each text feature of input, YiIndicate output sequence
The state of column being likely to occur, by text feature XiA series of Y can be obtained after inputiProbability value, we take its maximum value make
It is identification as a result, the expression of Z (x) are as follows:
Y herein |eAnd y |vRespectively indicate side and the node of the non-directed graph being made of annotated sequence, tiIndicate the transmitting of side e
Characteristic function, skThe state characteristic function being defined on node v, μkAnd λiRespectively indicate the weight of node feature and side feature.
tiAnd skAll there is relationship with position, is local feature function.In general, characteristic function tiAnd skValue is 1 or 0；When meeting feature item
Value is 1 when part, is otherwise 0.Condition random field is completely by characteristic function ti、skWith corresponding weight λi、μkIt determines.
Claims (4)
1. the evaluation object recognition methods based on semantic feature, it is characterised in that: first pre-processed to comment corpus；It is customized
Feature templates extract corresponding feature in comment corpus according to template；Evaluation object knowledge is carried out using conditional random field models
Not.
2. the evaluation object recognition methods based on semantic feature according to claim 1, it is characterised in that: the evaluation object
Recognition methods includes the following steps:
(1) corpus of text pre-acquiring and comment corpus pretreatment；I.e. through web crawlers in the online comment of electric business website
Corpus of text is obtained, comment corpus is segmented, part-of-speech tagging, emotion word mark and semantic role are analyzed, and from comment
Word feature, part of speech feature, emotion word feature, semantic role feature are obtained in corpus；
(2) training pattern；I.e. the comment corpus of extraction 4/5ths is extracted from training set as training set from comment corpus
Feature train customized feature templates, according to training result, obtain the weight of each feature；
(3) evaluation object identifies；I.e. using remaining 1/5th comments corpus as test set, the spy obtained according to step (2)
Template is levied, it is special that word feature corresponding with template, part of speech feature, emotion word feature and semantic role are extracted in test set
Then sign calculates probability P (y | x) using conditional random field models, x is list entries, and y is output sequence, from list entries
Obtain optimum.
3. the evaluation object recognition methods based on semantic feature according to claim 1, it is characterised in that: introduce semantic special
Sign, the semantic feature include emotion word feature, semantic role feature, by the position of emotion word feature assessment evaluation object,
And actor and word denoting the receiver of an action person in sentence are captured by semantic role feature.
4. the evaluation object recognition methods based on semantic feature according to claim 1, it is characterised in that: will be in test set
For feature as list entries x, evaluation object, evaluating word, which are used as to export, is used as output sequence y, and design conditions probability P (y | x), take
Conditional probability P(y | maximum value x) is as optimum.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810748969.5A CN109086340A (en) | 2018-07-10 | 2018-07-10 | Evaluation object recognition methods based on semantic feature |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810748969.5A CN109086340A (en) | 2018-07-10 | 2018-07-10 | Evaluation object recognition methods based on semantic feature |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109086340A true CN109086340A (en) | 2018-12-25 |
Family
ID=64837392
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810748969.5A Pending CN109086340A (en) | 2018-07-10 | 2018-07-10 | Evaluation object recognition methods based on semantic feature |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109086340A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110516241A (en) * | 2019-08-26 | 2019-11-29 | 北京三快在线科技有限公司 | Geographical address analytic method, device, readable storage medium storing program for executing and electronic equipment |
| CN110705268A (en) * | 2019-09-02 | 2020-01-17 | 平安科技（深圳）有限公司 | Article subject extraction method and device based on artificial intelligence and computer-readable storage medium |
| CN110807330A (en) * | 2019-09-09 | 2020-02-18 | 腾讯科技（深圳）有限公司 | Semantic understanding model evaluation method and device and storage medium |
| CN111309910A (en) * | 2020-02-13 | 2020-06-19 | 南京云问网络技术有限公司 | Text information mining method and device |
| CN111581474A (en) * | 2020-04-02 | 2020-08-25 | 昆明理工大学 | Evaluation object extraction method of case-related microblog comments based on multi-head attention system |
| CN111966822A (en) * | 2019-05-20 | 2020-11-20 | 北京京东尚科信息技术有限公司 | Method and device for determining emotion category of evaluation information |
| CN111967980A (en) * | 2020-09-02 | 2020-11-20 | 上海积成能源科技有限公司 | System and method for bank credit analysis by adopting maximum entropy model based on enterprise financial and newspaper and energy consumption data |
| CN112256826A (en) * | 2020-10-19 | 2021-01-22 | 网易（杭州）网络有限公司 | Emotion analysis method, evaluation method and emotion analysis model training method and device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103049435A (en) * | 2013-01-04 | 2013-04-17 | 浙江工商大学 | Text fine granularity sentiment analysis method and text fine granularity sentiment analysis device |
| CN103631961A (en) * | 2013-12-17 | 2014-03-12 | 苏州大学张家港工业技术研究院 | Method for identifying relationship between sentiment words and evaluation objects |
| 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 |
| CN104298665A (en) * | 2014-10-16 | 2015-01-21 | 苏州大学 | Identification method and device of evaluation objects of Chinese texts |
| US9679226B1 (en) * | 2012-03-29 | 2017-06-13 | Google Inc. | Hierarchical conditional random field model for labeling and segmenting images |
-
2018
- 2018-07-10 CN CN201810748969.5A patent/CN109086340A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9679226B1 (en) * | 2012-03-29 | 2017-06-13 | Google Inc. | Hierarchical conditional random field model for labeling and segmenting images |
| CN103049435A (en) * | 2013-01-04 | 2013-04-17 | 浙江工商大学 | Text fine granularity sentiment analysis method and text fine granularity sentiment analysis device |
| CN103631961A (en) * | 2013-12-17 | 2014-03-12 | 苏州大学张家港工业技术研究院 | Method for identifying relationship between sentiment words and evaluation objects |
| 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 |
| CN104298665A (en) * | 2014-10-16 | 2015-01-21 | 苏州大学 | Identification method and device of evaluation objects of Chinese texts |
Non-Patent Citations (2)
| Title |
|---|
| 夏 圆, 张 征: ""基于条件随机场的评价对象抽取"", 《计算机系统应用》 * |
| 王荣洋等: ""基于ＣＲＦｓ的评价对象抽取特征研究"", 《中文信息学报》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111966822A (en) * | 2019-05-20 | 2020-11-20 | 北京京东尚科信息技术有限公司 | Method and device for determining emotion category of evaluation information |
| CN110516241A (en) * | 2019-08-26 | 2019-11-29 | 北京三快在线科技有限公司 | Geographical address analytic method, device, readable storage medium storing program for executing and electronic equipment |
| CN110705268A (en) * | 2019-09-02 | 2020-01-17 | 平安科技（深圳）有限公司 | Article subject extraction method and device based on artificial intelligence and computer-readable storage medium |
| WO2021042517A1 (en) * | 2019-09-02 | 2021-03-11 | 平安科技（深圳）有限公司 | Artificial intelligence-based article gist extraction method and device, and storage medium |
| CN110705268B (en) * | 2019-09-02 | 2024-06-25 | 平安科技（深圳）有限公司 | Article subject matter extraction method and device based on artificial intelligence and computer readable storage medium |
| CN110807330A (en) * | 2019-09-09 | 2020-02-18 | 腾讯科技（深圳）有限公司 | Semantic understanding model evaluation method and device and storage medium |
| CN110807330B (en) * | 2019-09-09 | 2023-04-07 | 腾讯科技（深圳）有限公司 | Semantic understanding model evaluation method, device and storage medium |
| CN111309910A (en) * | 2020-02-13 | 2020-06-19 | 南京云问网络技术有限公司 | Text information mining method and device |
| CN111581474A (en) * | 2020-04-02 | 2020-08-25 | 昆明理工大学 | Evaluation object extraction method of case-related microblog comments based on multi-head attention system |
| CN111581474B (en) * | 2020-04-02 | 2022-07-29 | 昆明理工大学 | Evaluation object extraction method of case-related microblog comments based on multi-head attention system |
| CN111967980A (en) * | 2020-09-02 | 2020-11-20 | 上海积成能源科技有限公司 | System and method for bank credit analysis by adopting maximum entropy model based on enterprise financial and newspaper and energy consumption data |
| CN112256826A (en) * | 2020-10-19 | 2021-01-22 | 网易（杭州）网络有限公司 | Emotion analysis method, evaluation method and emotion analysis model training method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109086340A (en) | Evaluation object recognition methods based on semantic feature | |
| WO2018028077A1 (en) | Deep learning based method and device for chinese semantics analysis | |
| CN106599032B (en) | Text event extraction method combining sparse coding and structure sensing machine | |
| CN108363725B (en) | Method for extracting user comment opinions and generating opinion labels | |
| CN106649519B (en) | Product characteristic mining and evaluating method | |
| Bhonde et al. | Sentiment analysis based on dictionary approach | |
| CA3088692A1 (en) | Visualizing comment sentiment | |
| CN103853824A (en) | In-text advertisement releasing method and system based on deep semantic mining | |
| Kwaik et al. | An Arabic tweets sentiment analysis dataset (ATSAD) using distant supervision and self training | |
| CN109960756A (en) | Media event information inductive method | |
| CN103903164A (en) | Semi-supervised automatic aspect extraction method and system based on domain information | |
| Zhao et al. | Sentiment analysis on the online reviews based on hidden Markov model | |
| CN114971730A (en) | Method for extracting file material, device, equipment, medium and product thereof | |
| Banik et al. | Gru based named entity recognition system for bangla online newspapers | |
| CN112860896A (en) | Corpus generalization method and man-machine conversation emotion analysis method for industrial field | |
| US9652997B2 (en) | Method and apparatus for building emotion basis lexeme information on an emotion lexicon comprising calculation of an emotion strength for each lexeme | |
| CN108536673B (en) | News event extraction method and device | |
| Cui et al. | KNET: A general framework for learning word embedding using morphological knowledge | |
| CN105389303A (en) | Automatic heterogenous corpus fusion method | |
| Kumar et al. | Aspect-Based Sentiment Analysis of Tweets Using Independent Component Analysis (ICA) and Probabilistic Latent Semantic Analysis (pLSA) | |
| CN113868431A (en) | Financial knowledge graph-oriented relation extraction method and device and storage medium | |
| Dandashi et al. | Arabic named entity recognition—a survey and analysis | |
| Tayal et al. | DARNN: Discourse Analysis for Natural languages using RNN and LSTM. | |
| Abudureheman | Design of a User Comment Management System Based on Text Mining: Innovative Organization Management for E-Commerce | |
| KR101909537B1 (en) | System and method for classifying social data |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20181225 |