CN103699626B - Method and system for analysing individual emotion tendency of microblog user - Google Patents

Abstract

The invention discloses a method for analysing individual emotion tendency of microblog user. The method comprises the following steps of: acquiring data, separating words, loading a word bank and an emoticon bank, establishing the interested topic bank of the user, dividing short sentences, extracting emotion elements, establishing the individual locution list of the user, calculating a locution emotion value, calculating the topic emotion tendency of the user, and calculating the overall emotion tendency of the user. The invention further discloses a system for analysing individual emotion tendency of microblog user. According to the method and the system disclosed by the invention, an emotion analysis on the single microblog user is realized, and the emotion analysis on the user is combined with specific topics to avoid an indistinct and stiff analysis mode, thus the emotion analysis on the user is more meticulous and directional, and the accuracy of the emotion tendency analysis is improved.

Description

A kind of microblog users individualized emotion trend analysis method and system

Technical field

The present invention relates to microblog data process field, particularly to a kind of microblog users individualized emotion trend analysis method And system.

Background technology

Microblogging is the random platform of a freedom, and its information is short and small, issues rapid, user delivers oneself frequently by microblogging Subjective feeling to various events and comment object, share with other people oneself value, suggestion, emotion etc..In micro-blog information Contain a lot of emotion words, contain abundant emotion information.The environment freely loosening, makes emotion in user's microblog data Expression information can deeper into, exactly reflect user Sentiment orientation.

The sentiment analysis research work of Chinese microblogging, is primarily directed to certain particular event and theme is carried out at present, analysis All related micro-blog information texts, extract Emotion element, carry out statistical analysis, the emotion information of microblogging are carried out point Class, mark and prediction, achieve certain achievement.But current research is primarily upon the sentiment analysis of micro-blog information or colony uses The heartbeat conditions at family, for single microblog users Sentiment orientation analysis and research not yet carry out in a deep going way, not individually for The Sentiment orientation analysis of user.And, the analysis to Sentiment orientation, it does not refine to each specific aspect in social life yet, The specific aim that this results in sentiment analysis is not strong, the accuracy of analysis and prediction, comprehensive waits to improve further.

The emotional expression mode of microblogging is personalized it is necessary to the individualized feature introducing user just can obtain more accurately Analysis result.The much-talked-about topic conversion of microblogging is very fast, and active user is relatively stable.The trend of impact event development is institute There are the microblog users of participation, the emotion model of user is metastable.By the analysis to user individual emotion, Ke Yigeng Emotion, the development of predicted events and the change of accurate more meticulously labelling micro-blog information.The user feeling analysis information set up is also Can be with life-time service, the accumulation with data can be more and more accurate.

By the individualized emotion analytical technology to microblog users, can analyze judge they to hot issue, particularly By liking or hatred degree of, special object or product, excavate business therein and social value, before there is wide application Scape, such as 1) public sentiment monitoring, the trend analysis and prediction of much-talked-about topic, sentiment analysis of social group etc.；2) stock market, popular disease The trend analysis and prediction such as disease, election；3) user behavior analysis based on big data, such as propensity to consume, user preferences etc..Micro- The research of rich user individual Sentiment orientation analysis method has important learning value and social meaning.

Content of the invention

In order to overcome disadvantages mentioned above and the deficiency of prior art, it is an object of the invention to provide a kind of microblog users individual character Change Sentiment orientation analysis method it is achieved that sentiment analysis to microblogging unique user, make the sentiment analysis of user more careful, more Directional.

Another object of the present invention is to providing a kind of microblog users individualized emotion trend analysis system.

The purpose of the present invention is achieved through the following technical solutions:

A kind of microblog users individualized emotion trend analysis method, comprises the following steps:

(1) gather all data of the microblogging homepage of each user, be stored in data base；

(2) text data in the microblog data that step (1) is collected carries out participle, obtains participle set and part of speech mark Note；

(3) dictionary, emoticon storehouse needed for loading；Described dictionary includes hownet sentiment word lexicon, degree dictionary, negative Dictionary, personal pronoun dictionary, function word is connected dictionary, cyberspeak dictionary and classified lexicon；

(4) using the interlayer based on word frequency, aggregating algorithm sets up user's topic of interest storehouse upwards:

(4-1) set up topic tree: filter out the word not having topic meaning in user version data, obtain there is obvious topic The word of information, using classified lexicon, counts word frequency, sets up topic tree；Described topic tree is hierarchical structure, and ground floor is one Level topic classification, the second layer is two grades of sub-topic classification, and third layer is classified for three-level sub-topic；The described word not having topic meaning Language includes degree word, negative word, personal pronoun, function word, link word, adjective；

(4-2) according to topic tree, by the aggregating algorithm upwards of the interlayer based on word frequency, successively extract high frequency topic；

(4-3) set up a main split, for placing popular spy on the topic word that cannot be included into father's layer topic and network There are topic word or phrase, obtain conventional topic storehouse；Word in microblog data is mated with the word in conventional topic storehouse Corresponding, the topic word language that occurrence number in the microblog data of user is exceeded threshold value extracts, and also serves as high frequency topic；

(4-4) the high frequency topic obtaining step (4-2) and (4-3), as user's topic of interest word, sets up user's sense Interest topic storehouse；

(5) microblog data that step (1) is collected divide short sentence it is ensured that each short sentence at most contain one interested Topic word；

(6) extract the Emotion element in each short sentence, calculate the initial emotion value of each short sentence:

(6-1) set of words of short sentence is carried out mating mapping with each dictionary and emoticon storehouse, mark all kinds of emotions Element；Described Emotion element includes emotion word, degree word, negative word, punctuation mark, emoticon, wherein degree word and punctuate Symbol is all used for adjusting the degree of emotion word, and negative word is used for adjusting the polarity of emotion word；

(6-2) the emotion value of calculating short sentence Chinese version:

The weights of setting Emotion element: positive emotion word weights are "+1 ", negative emotion word weights are " -1 "；Negative word is weighed It is worth for " -1 "；Degree word and punctuation mark, according to the depth of its degree, arrange weights, and weights scope is between 0 to 3；Degree word The emotion word being affected with punctuation mark follows nearby principle, and that is, each degree word or punctuation mark affect apart from its nearest feelings The emotion degree of sense word；

Emotion value i of short sentence Chinese version_wordsComputational methods be:

$i_{words}=b\·\underset{i=1}{\overset{m}{\σ}}(\underset{j=1}{\overset{n}{\σ}}{c}_{ij}\·f_{ij})\·q_i$

In formula, q_iRepresent i-th emotion word, c_ijRepresent and modify q_iJ-th degree word weights, f_ijRepresent and modify q_i? J negative word weights；If q_iDo not attach degree word, then c_ijTake default value 1；If q_iDo not attach negative word, then f_ijTake acquiescence Value 1；N takes modification q_iDegree word number and modify q_iNegative word number in maximum, m represents the number of emotion word, b table The corresponding weights of indicating point symbol, i, j are positive integer；

(6-3) calculate short sentence in emoticon emotion value:

The expression providing for microblogging operator, it is divided into front for the contribution of Sentiment orientation, negatively, neutral three kinds Situation: the weights of front emoticon are set to "+1 ", the weights of negative emoticon are set to " -1 ", the weights of neutral expression's symbol It is set to " 0 "；

Emoticon emotion value i in short sentence_marksComputational methods be:

$i_{marks}=\underset{i=1}{\overset{l}{\σ}}{m}_i$

In formula, m_iRepresent the expression of i-th table front, negative or neutral emotion, i is positive integer, and l is emoticon Number；

(6-4) calculate initial emotion value i of short sentence₀:

i₀=i_words+i_marks；

(7) to the text data after step (2) process, extract the word combination of high frequency using word slip window sampling, obtain To user individual idiom list；

(8) statistical analysiss are carried out to the initial emotion value of all short sentences comprising each bar idiom, draw the feelings of idiom Inductance value；

For every idiom, find out all short sentences containing this idiom, its initial emotion value sum-average arithmetic calculates Method is as follows:

$i_g=\frac{1}{p}\underset{i=1}{\overset{p}{\σ}}{i}_{0i}$

In formula, i_0iFor the initial emotion value of i-th short sentence comprising this idiom, p is the short sentence number containing this idiom, i_gEmotion initial value for this idiom；

By i_gValue be mapped in [- 3,3], obtain emotion value i of idiom '_g, record usual in the personalization of this user In language emotion labelling table；

(9) calculate the individualized emotion value of each short sentence, computational methods are:

$i=i_0-\underset{i=1}{\overset{m^{\′}}{\σ}}(\underset{j=1}{\overset{n^{\′}}{\σ}}{c}_{gij}\·f_{gij})\·q_{gi}+\underset{k=1}{\overset{r}{\σ}}{i}_{gk}^{\′}$

In formula, i₀For the initial emotion value of short sentence, q_giRepresent i-th word, c_gijRepresent and modify q_giJ-th degree word Weights, f_gijRepresent and modify q_giJ-th negative word weights；If q_giDo not attach degree word, then c_gijTake default value 1；If q_giNot yet There is subsidiary negative word, then f_gijTake default value 1；N' takes modification q_giThe number of degree word and modify q_giThe number of negative word in Maximum；M' represents the number of word, and i, j are positive integer；i'_gkRepresent the emotion value of k-th idiom, r represents that this is short The number of idiom in sentence；

(10) Sentiment orientation of calculating user's topic of interest:

For any user topic of interest word in user's topic of interest storehouse, it is calculated as follows its emotion value:

$i_{{\mathrm{topic}}_i}=\frac{1}{w}\underset{j=1}{\overset{w}{\σ}}{i}_j$

i_jFor comprising the individualized emotion value of j-th short sentence of this user's topic of interest word, w is to comprise this user to feel emerging The short sentence sum of funny remarks epigraph,Emotion value for this user's topic of interest word；WillValue be mapped in [- 3,3], Obtain the topic Sentiment orientation value of final user, using these values, set up user individual microblog topic emotion value list.

After carrying out step (10), can also follow the steps below:

A () repeats step (1)～(7), for the idiom adding new in user individual idiom list, by step Suddenly (8) are calculated emotion value i of this idiom '_g；It is reported in user individual idiom list for before this circulation In idiom, then by the following method update i '_g: to every idiom, carry out step (8) first, obtain the i of epicycle circulation_g's Value of calculation i_{g_new}If, i_{g_prev}The i obtaining for last round of circulation_gValue of calculation, then i_gIt is updated to:

i_g=ω₁i_{g_prev}+ω₂i_{g_new}

In formula, ω₁For i_{g_prev}Weights, ω₂For i_{g_new}Weights；

By i_gValue be mapped in [- 3,3], obtain emotion value i of idiom '_g；

B i ' that () obtains according to step (a)_g, carry out step (9)～(10).

After carrying out step (10), can also follow the steps below:

Repeat step (1)～(9), for emerging in the new user's sense adding in user's topic of interest storehouse in this circulation Funny remarks is write inscription, and calculates the topic Sentiment orientation of user by the method for step (10)；It is reported in user for before this circulation User's topic of interest word in topic of interest storehouse, then update the topic Sentiment orientation i of user by the following method_topic: first Carry out step (10), obtain the i of epicycle circulation_topicValue of calculation i_{topic_new}If, i_{topic_prev}Last round of circulation obtains i_topicValue of calculation, then i_topicIt is updated to:

i_topic=ω '₁i_{topic_prev}+ω'₂i_{topic_new}

ω'₁For i_{topic_prev}Weights, ω '₂For i_{topic_new}Weights.

After carrying out step (10), also follow the steps below:

According to following formula calculating user's overall emotion propensity value:

$i_{user}=\frac{1}{s}\underset{i=1}{\overset{s}{\σ}}{i}_i$

In formula, i_iRepresent the emotion value of i-th short sentence, s represents short sentence sum.

All data of the described microblogging homepage gathering each user of step (1), particularly as follows:

In units of user, collect all data in its homepage；Described data includes oneself sending out in user home page face Cloth and forwarding microblog data, the data of " comment sending ", the data of "@mentions mine ", microblogging name and user of interest Microblogging name, self-introduction, spontaneous or forward the title in the url webpage and video link that comprise in microblogging.

Step (4-2) described according to topic tree, by the aggregating algorithm upwards of the interlayer based on word frequency, successively extract high frequency words Topic, particularly as follows:

To each topic word, if its occurrence number is higher than the threshold value setting, this topic is high frequency topic, otherwise, will The occurrence number of this topic word passes to father's layer topic word, successively calculates and extracts high frequency topic；The occurrence number of topic word includes this The occurrence number of topic itself and the occurrence number of the sub-topic of this topic.

The described data collecting step (1) of step (5) divides short sentence, particularly as follows:

Each microblog data of analysis, if not having topic word or pertaining only to a topic, whole piece in a microblog data Microblogging is as a short sentence；

If one in microblog data contain multiple topic word, combine punctuation mark degree of priority analyze: if two away from Have punctuate between maximum topic word, be then split into two short sentences at punctuate, if the maximum topic word of two distances it Between there is no punctuate, then checking time the topic word of big distance, if all there is no punctuate, not splitting, whole piece microblog data conduct One short sentence；Described distance is to reach another topic word tree number to be passed through from a topic word in topic tree；If There are multiple punctuates between two topic word, then choose and divided at the high punctuate of relative importance value: in punctuation mark relative importance value, fullstop > branch > comma.

Text data after the described process to step (2) of step (7), extracts the word combination of high frequency, particularly as follows:

If the length of window of sliding window is w, w is the word number that sliding window comprises, and successively takes 1,2,3,4 respectively；Profit Count in all short sentences with word sliding window, the total degree of each word or phrase appearance, will appear from number of times and be more than threshold value User individual idiom list listed in word or phrase；Collect daily statement word, set up daily statement phrase database；By daily table Predicate is rejected from user individual idiom list, obtains user individual idiom list.

Realize the microblog users individualized emotion trend analysis system of above-mentioned analysis method, including

Data acquisition module, for gathering all data of the microblogging homepage of each user, is stored in data base；

Word-dividing mode, for carrying out participle to the text data in the data collecting, obtains participle set and part of speech mark Note；

Dictionary load-on module, for loading required dictionary, emoticon storehouse；Described dictionary includes hownet emotion word word Storehouse, degree dictionary, negates dictionary, personal pronoun dictionary, and function word is connected dictionary, cyberspeak dictionary and classified lexicon；

Module is set up in user's topic of interest storehouse, for using the interlayer based on word frequency, aggregating algorithm sets up user's sense upwards Interest topic storehouse；

Short sentence division module, divide short sentence for the data that arrives data collecting module collected it is ensured that each short sentence extremely Contain a topic of interest word more；

Emotion element extraction module, for extracting the Emotion element in each short sentence, calculates the initial emotion of each short sentence Value；

Module is set up in user individual idiom list, for the data that data acquisition module is collected, using word Slip window sampling extracts the word combination of high frequency, obtains user individual idiom list；

Idiom emotion value computing module, for carrying out statistical to the emotion value of all short sentences comprising each bar idiom Analysis, draws the emotion value of idiom；

Short sentence emotion value computing module, for calculating the individualized emotion value of each short sentence；

User's topic Sentiment orientation computing module, for calculating the Sentiment orientation of each topic of interest of user.

Described microblog users individualized emotion trend analysis system, also includes user's overall emotion tendency computing module, Overall emotion for calculating user is inclined to.

Compared with prior art, the present invention has advantages below and a beneficial effect:

(1) present invention achieves sentiment analysis to microblogging unique user, by the sentiment analysis of user with to concrete topic Combine, it is to avoid general mechanical analytical model, make the sentiment analysis to user more careful, more directional.From user Habitual expression way consider, using user individual idiom as one of the element of sentiment analysis, be conducive to improving emotion The accuracy of trend analysis.

(2) present invention is directed to user, carries out sentiment analysis to its microblog data, user can be helped to be appreciated more fully from itself And the hobby of other users；When the new speech focus of one section of appearance in network, it is possible to use the result that this method obtains, quickly Draw interest level and the Sentiment orientation of this user, to predict speech and the reaction of user；Product development business, operator and wide Accuse business and utilize this method, user interested in pushed away commodity or service can be found out, contribute to product development business offer fuller The commodity of sufficient demand, contributing to operator provides more humane intimate service, contributes to advertiser and is directed to user's input advertisement； Market demand reference can be provided for many industries；Contribute to public sentiment monitoring.

Brief description

Fig. 1 is the flow chart of the microblog users individualized emotion trend analysis method of embodiments of the invention.

Fig. 2 is the data acquisition of microblog users individualized emotion trend analysis method and the pretreatment of embodiments of the invention The particular flow sheet of process.

Fig. 3 is the particular flow sheet setting up user's topic of interest storehouse of embodiments of the invention.

Fig. 4 is the schematic diagram of the topic tree set up by the microblog data of certain user in embodiments of the invention.

Fig. 5 is the particular flow sheet successively extracting high frequency topic in embodiments of the invention from topic tree.

Fig. 6 is the concrete setting schematic diagram of the emoticon weights of embodiments of the invention.

Fig. 7 is the exemplary plot (w takes 1) of the utilization word sliding window extraction word combination of embodiments of the invention.

Fig. 8 is the exemplary plot (w takes 3) of the utilization word sliding window extraction word combination of embodiments of the invention.

Fig. 9 is the idiom emotion value calculation flow chart of embodiments of the invention.

Figure 10 is step (7)～(8) of microblog users individualized emotion trend analysis method of embodiments of the invention Particular flow sheet.

Figure 11 is step (5)～(11) of microblog users individualized emotion trend analysis method of embodiments of the invention Particular flow sheet.

Figure 12 is the structural representation of the microblog users individualized emotion trend analysis system of embodiments of the invention.

Specific embodiment

With reference to embodiment, the present invention is described in further detail, but embodiments of the present invention not limited to this.

Embodiment

As shown in figure 1, the microblog users individualized emotion trend analysis method of the present embodiment, comprise the following steps:

(1) gather all microblog data of the microblogging homepage of each user, be stored in data base:

In units of user, collect all data in its homepage；Described data includes oneself sending out in user home page face Cloth and forwarding microblog data, the data of " comment sending ", the data of "@mentions mine ", microblogging name and user of interest Microblogging name, self-introduction, spontaneous or forward the title in the url webpage and video link that comprise in microblogging.

(2) text data in the microblog data that step (1) is collected is read, using Chinese in units of microblogging paragraph by paragraph The segmenting method of morphological analysis system ictclas carries out participle operation, obtains participle set and corresponding part-of-speech tagging.

(3) dictionary, emoticon storehouse needed for loading；Described dictionary includes hownet sentiment word lexicon, degree dictionary, negative Dictionary, personal pronoun dictionary, function word is connected dictionary, cyberspeak dictionary and classified lexicon；Wherein, hownet sentiment word lexicon, Degree dictionary, each word in negative dictionary is with weights；Chinese word is only taken, by it in hownet sentiment word lexicon In positive emotion word and front evaluates word be classified as positive emotion word, by negative emotion word therein and unfavorable ratings Word is classified as negative emotion word；With regard to degree word and negative word, using self-built degree word dictionary and negative word dictionary, wherein Degree word dictionary comprises degree word 219, and negative dictionary comprises negative word 48.Classified lexicon is using the qq input method of improvement Classified lexicon, is hierarchical tree structure；Dictionary ground floor be maximum kind, comprise game, disciplines, hobby, sports and amusement, Culture and arts, area etc., the dictionary second layer is the subclasses of ground floor, for example, is divided into reason below " disciplines " big class again Workers and peasants doctor, social sciences economy, education military affairs etc., dictionary third layer is the subclasses of second layer classification, such as divides below social sciences economy For equity fund, law, commodity, finance etc.；Under the catalogue of whole classified lexicon, have 6223 and do not subdivide dictionary；Expression Symbolic library passes through to collect the emoticon of microblog and provide weights to obtain.

Step (1)～(3) are data acquisition and preprocessing process, and particular flow sheet is shown in Fig. 2.

(4) using the interlayer based on word frequency, aggregating algorithm sets up user's topic of interest storehouse upwards, as shown in figure 3, step As follows:

(4-1) set up topic tree: filter out the word not having topic meaning in user version data, obtain there is obvious topic The word of information, using classified lexicon, counts word frequency, sets up topic tree；Described topic tree is hierarchical structure, and ground floor is one Level topic classification, the second layer is two grades of sub-topic classification, and third layer is classified for three-level sub-topic；The described word not having topic meaning Language includes degree word, negative word, personal pronoun, function word, link word, adjective；

Fig. 4 is the topic tree set up by the microblog data of certain user.In square frame, corresponding numeral is this topic word and its son The number of times that topic word occurs.

(4-2) according to topic tree, by the aggregating algorithm upwards of the interlayer based on word frequency, successively extract high frequency topic: to every One topic word, if its occurrence number is higher than the threshold value setting, this topic is high frequency topic, otherwise, going out this topic word Occurrence number passes to his father's layer topic word, successively calculates and extracts high frequency topic；The occurrence number of topic word includes this topic itself Occurrence number and the occurrence number of its sub-topic；

Operation example as shown in figure 5, in figure represents the high frequency topic extracting with dashed rectangle, extracted altogether four high Frequency topic.

(4-3) set up a main split, for placing popular spy on the topic word that cannot be included into father's layer topic and network There are topic word or phrase, obtain conventional topic storehouse, this branch does not have layer architecture；By the word in microblog data and conventional words Word in exam pool carries out coupling and corresponds to, and the topic word language that occurrence number in the microblog data of user is exceeded threshold value extracts Come, also serve as high frequency topic；

(4-4) the high frequency topic obtaining step (4-2) and (4-3), as user's topic of interest word, sets up user's sense Interest topic storehouse.

(5) data that step (1) collects is divided short sentence it is ensured that each short sentence at most contains a topic of interest Word；

The described data collecting step (1) divides short sentence, particularly as follows: each microblog data of analysis, if one is micro- There is no topic word in rich data or pertain only to a topic, then whole piece microblogging is as a short sentence；If one containing in microblog data There are multiple topic word, then combine punctuation mark degree of priority and analyze: if having punctuate between the topic word of two distance maximums, It is split into two short sentences at punctuate；If there is no punctuate between the topic word of two distance maximums, then check time big distance Topic word；If all there is no punctuate, do not split, whole piece microblog data is as a short sentence；Described distance is in topic tree In reach another topic word tree number to be passed through from a topic word；If having multiple punctuates between two topic word, Choose and divided at the high punctuate of relative importance value: in punctuation mark relative importance value, fullstop > branch > comma.

(6) extract the Emotion element in each short sentence, calculate the initial emotion value of each short sentence:

(6-1) set of words of short sentence is carried out mating mapping with each dictionary and emoticon storehouse, mark all kinds of emotions Element；Described Emotion element includes emotion word, degree word, negative word, punctuation mark, emoticon, wherein degree word and punctuate Symbol is all used for adjusting the degree of emotion word, and negative word is used for adjusting the polarity of emotion word；

(6-2) the emotion value of calculating short sentence Chinese version:

The weights of setting Emotion element: positive emotion word weights are "+1 ", negative emotion word weights are " -1 "；Negative word is weighed It is worth for " -1 "；Degree word and punctuation mark, according to the depth of its degree, arrange weights, and weights scope is 0 to 3；Degree word and mark The emotion word that point symbol is affected follows nearby principle, and that is, each degree word or punctuation mark affect apart from its nearest emotion word Emotion degree；

Weights setting citing such as table 1～4:

Table 1 hownet emotion word

Front word Language	Negation words Language
		Love is doted on	Regret
Love and esteem	Annoyed
		Good and sound	Dim
…	…

Degree word weights commonly used by table 2

Degree word	Degree value
		Too	3
Very	2.5
		Very	2
…	…

Negative word weights commonly used by table 3

Table 4 punctuation mark weights

Punctuation mark	Degree coefficient
		.(fullstop)	1
, (comma)	1
		！！！…！	2
？？？…？	1.5
		…	…

Emotion value i of short sentence Chinese version_wordsComputational methods be:

$i_{words}=b\·\underset{i=1}{\overset{m}{\σ}}(\underset{j=1}{\overset{n}{\σ}}{c}_{ij}\·f_{ij})\·q_i$

In formula, q_iRepresent i-th emotion word, c_ijRepresent and modify q_iJ-th degree word weights, f_ijRepresent and modify q_i? J negative word weights；If q_iDo not attach degree word, then c_ijTake default value 1；If q_iDo not attach negative word, then f_ijTake acquiescence Value 1；N takes modification q_iDegree word number and modify q_iNegative word number in maximum, m represents the number of emotion word, b table The corresponding weights of indicating point symbol, i, j are positive integer；

(6-3) calculate short sentence in emoticon emotion value:

The expression providing for microblogging operator, it is divided into front for the contribution of Sentiment orientation, negatively, neutral three kinds Situation: the weights of front emoticon are set to "+1 ", the weights of negative emoticon are set to " -1 ", the weights of neutral expression's symbol It is set to " 0 "；

The concrete setting citing of emoticon weights is as shown in Figure 6.

Emoticon emotion value i in short sentence_marksComputational methods be:

$i_{marks}=\underset{i=1}{\overset{l}{\σ}}{m}_i$

In formula, m_iRepresent the expression of i-th table front, negative or neutral emotion, i is positive integer, and l is emoticon Number；

(6-4) calculate initial emotion value i of short sentence₀:

i₀=i_words+i_marks；

(7) to the text data after step (2) process, extract the word combination of high frequency using word slip window sampling, obtain To user individual idiom list；

If the length of window of sliding window is w, w is the word number that sliding window comprises, and successively takes 1,2,3,4 respectively；Profit Count in all short sentences with word sliding window, the total degree of each word or phrase appearance, will appear from number of times and be more than threshold value (no Have different threshold values with window, length of window is less, required threshold value is higher) word or phrase list in user individual be used to Term list；Wherein, during w=1, only statistical disposition is carried out to adjective and modal particle.For phrase, with the most word of word number Group lists idiom list in.In addition, the no obvious feelings being caused due to Chinese grammer using self-built daily statement phrase database rejecting The daily statement phrase that sense is pointed to, specific practice is: artificially collects Chinese everyday expressions collocation, as " I ", " present " " rise Come " etc., to compare the daily statement phrase that the no obvious emotion rejected in idiom list is pointed to.Thus, obtain user personality Change idiom list.

Fig. 7 takes word sliding window when 1 to extract the exemplary plot of word combination for w.

Fig. 8 takes word sliding window when 3 to extract the exemplary plot of word combination for w.

(8) the initial emotion value of all short sentences comprising each bar idiom is carried out with statistical analysiss, calculates each bar idiom Emotion value, as shown in figure 9, detailed process is as follows:

For every idiom, find out all short sentences containing this idiom, its initial emotion value sum-average arithmetic calculates Method is as follows:

$i_g=\frac{1}{p}\underset{i=1}{\overset{p}{\σ}}{i}_{0i}$

In formula, i_0iFor the initial emotion value of i-th short sentence comprising this idiom, p is the short sentence number containing this idiom, i_gEmotion initial value for this idiom；

In view of i in theory_gValue be infinity, but its most of numerical value is distributed near 10 again, so according to Formula (1) is by i_gThe value of (x as in formula) is mapped in [- 3,3], obtains emotion value i of idiom '_g(y as in formula), Record is in the personalized idiom emotion labelling table of this user；

$y=\left\{\begin{array}{cc}0.42\sqrt{x},& 0\&le;x<50\\ -0.42\sqrt{x},& -50<x<0\\ 3,& 0>50\\ -3,& x>-50\end{array}\right.---\left(1\right)$

The particular flow sheet of step (7)～(8) is shown in Figure 10.

(9) calculate the individualized emotion value of each short sentence, computational methods are: deduct comprise in initial emotion value idiom or The Emotion element value of calculation of any word in idiom, then adds the idiom calculating according to user individual idiom list Emotion value, obtain the individualized emotion value of short sentence it may be assumed that

$i=i_0-\underset{i=1}{\overset{m^{\&prime;}}{\&sigma;}}(\underset{j=1}{\overset{n^{\&prime;}}{\&sigma;}}{c}_{gij}\&centerdot;f_{gij})\&centerdot;q_{gi}+\underset{k=1}{\overset{r}{\&sigma;}}{i}_{gk}^{\&prime;}$

In formula, i₀The initial emotion value of the short sentence for above calculating, c_gij、f_gij、q_giIt is the correlometer of word in idiom Calculation value, q_giRepresent i-th word, c_gijRepresent and modify q_giJ-th degree word weights, f_gijRepresent and modify q_giJ-th negative Word weights；If q_giDo not attach degree word, then c_gijTake default value 1；If q_giDo not attach negative word, then f_gijTake default value 1； N' takes modification q_giThe number of degree word and modify q_giThe number of negative word in maximum, m' represents the number of word, i, J is positive integer；i'_gkRepresent the emotion value of k-th idiom, r represents the number of idiom in this short sentence.

(10) Sentiment orientation of calculating user's topic of interest:

For any user topic of interest word in user's topic of interest storehouse, it is calculated as follows its emotion value:

$i_{{\mathrm{topic}}_i}=\frac{1}{w}\underset{j=1}{\overset{w}{\&sigma;}}{i}_j$

In formula, i_jFor comprising the individualized emotion value of j-th short sentence of this user's topic of interest word, w is to comprise this use The short sentence sum of family topic of interest word,Emotion value for this user's topic of interest word；

Will according to formula (1)The value of (x as in formula) is mapped in [- 3,3], obtains final user interested Topic Sentiment orientation value (y as in formula).Using these values, set up user individual microblog topic emotion value list, such as table 5 Shown.

Table 5 user individual microblog topic emotion value list

After carrying out step (10), also the overall emotion tendency of user can be calculated, that is, follow the steps below:

(11) overall emotion calculating user is inclined to:

According to following formula calculating user's overall emotion propensity value:

$i_{user}=\frac{1}{s}\underset{i=1}{\overset{s}{\&sigma;}}{i}_i$

In formula, i_iRepresent the emotion value of i-th short sentence, s represents short sentence sum.

The particular flow sheet of step (5)～(11) is shown in Figure 11.

It is, in general, that micro-blog information amount is bigger, the topic word in user's topic of interest storehouse is then abundanter, the emotion of user Trend analysis are also more accurate.So, the Sentiment orientation analysis to microblog users should regularly repeat, and covers old result.This It is a kind of mechanism of the timing upgrading of the sentiment analysis to user, result can be made more comprehensively, accurately, more ageing.

Wherein, due in renewal process the list of user individual idiom have new idiom and add, the present embodiment Microblog users individualized emotion trend analysis method is updated to the analysis result of step (1)～(11) using following methods:

After carrying out step (11), also follow the steps below:

(12) step (1)～(7) are repeated, for the idiom adding new in user individual idiom list, by Step (8) is calculated emotion value i of this idiom '_g；It is reported in user individual idiom row for before this circulation Idiom in table, then update i ' by the following method_g: to every idiom, carry out step (8) first, obtain the i of epicycle circulation_g Value of calculation i_{g_new}If, i_{g_prev}The i obtaining for last round of circulation_gValue of calculation, then i_gIt is updated to:

i_g=ω₁i_{g_prev}+ω₂i_{g_new}

In formula, ω₁For i_{g_prev}Weights, ω₂For i_{g_new}Weights；

By i_gValue be mapped in [- 3,3], obtain emotion value i of idiom '_g；

(13) i ' being obtained according to step (12)_g, carry out follow-up step (9)～(11).

Because in renewal process, user's topic of interest storehouse has new user's topic of interest word addition, the present embodiment Microblog users individualized emotion trend analysis method is updated to the analysis result of step (1)～(11) using following methods:

After carrying out step (11), also follow the steps below:

(14) step (1)～(9) are repeated, for the user in the new addition in user's topic of interest storehouse in this circulation Topic of interest word, is calculated the topic Sentiment orientation of user by the method for step (10)；It is reported in for before this circulation User's topic of interest word in user's topic of interest storehouse, then update the topic Sentiment orientation i of user by the following method_topic: Carry out step (10) first, obtain the i of epicycle circulation_topicValue of calculation i_{topic_new}If, i_{topic_prev}Obtain for last round of circulation i_topicValue of calculation, then i_topicIt is updated to:

i_topic=ω '₁i_{topic_prev}+ω'₂i_{topic_new}

ω'₁For i_{topic_prev}Weights, ω '₂For i_{topic_new}Weights.

The update method in conventional topic storehouse: periodically check much-talked-about topic one hurdle in microblogging homepage, topic therein is added To in conventional topic storehouse；In addition, the renewal of qq input method classified lexicon is checked in timing, new entry is added in classified lexicon.

As shown in figure 12, the microblog users individualized emotion trend analysis system of the present embodiment, including

Data acquisition module, for gathering all data of the microblogging homepage of each user, is stored in data base；

Word-dividing mode, for carrying out participle to the text data in the data collecting, obtains participle set and part of speech mark Note；

Dictionary load-on module: for loading required dictionary, emoticon storehouse；Described dictionary includes hownet emotion word word Storehouse, degree dictionary, negates dictionary, personal pronoun dictionary, and function word is connected dictionary, cyberspeak dictionary and classified lexicon；

Module is set up in user's topic of interest storehouse: for using the interlayer based on word frequency, aggregating algorithm sets up user's sense upwards Interest topic storehouse:

Short sentence division module, divide short sentence for the data that arrives data collecting module collected it is ensured that each short sentence extremely Contain a topic of interest word more；

Emotion element extraction module, for extracting the Emotion element in each short sentence, calculates the initial emotion of each short sentence Value；

Module is set up in user individual idiom list, for the data that data acquisition module is collected, using word Slip window sampling extracts the word combination of high frequency, obtains user individual idiom list；

Idiom emotion value computing module, for carrying out statistical to the emotion value of all short sentences comprising each bar idiom Analysis, draws the emotion value of idiom:

Short sentence emotion value computing module, for calculating the individualized emotion value of each short sentence；

The topic Sentiment orientation computing module of user, for calculating the Sentiment orientation of each topic of interest of user；

User's overall emotion is inclined to computing module, and the overall emotion for calculating user is inclined to.

Above-described embodiment is the present invention preferably embodiment, but embodiments of the present invention are not subject to described embodiment Limit, other any spirit without departing from the present invention and the change made under principle, modification, replacement, combine, simplify, All should be equivalent substitute mode, be included within protection scope of the present invention.

Claims (10)

1. a kind of microblog users individualized emotion trend analysis method is it is characterised in that comprise the following steps:

(1) gather all data of the microblogging homepage of each user, be stored in data base；

(2) text data in the microblog data that step (1) is collected carries out participle, obtains participle set and part-of-speech tagging；

(3) dictionary, emoticon storehouse needed for loading；Described dictionary includes hownet sentiment word lexicon, degree dictionary, negative word Storehouse, personal pronoun dictionary, function word is connected dictionary, cyberspeak dictionary and classified lexicon；

(4) using the interlayer based on word frequency, aggregating algorithm sets up user's topic of interest storehouse upwards:

(4-1) set up topic tree: filter out the word not having topic meaning in user version data, obtain there is obvious topic information Word, using classified lexicon, count word frequency, set up topic tree；Described topic tree is hierarchical structure, and ground floor is talked about for one-level Topic classification, the second layer is two grades of sub-topic classification, and third layer is classified for three-level sub-topic；The described word bag not having topic meaning Include degree word, negative word, personal pronoun, function word, link word, adjective；

(4-2) according to topic tree, by the aggregating algorithm upwards of the interlayer based on word frequency, successively extract high frequency topic；

(4-3) set up a main split, for placing popular peculiar words on the topic word that cannot be included into father's layer topic and network Epigraph or phrase, obtain conventional topic storehouse；By the word in microblog data and the word in conventional topic storehouse carry out coupling corresponding, The topic word language that occurrence number in the microblog data of user is exceeded threshold value extracts, and also serves as high frequency topic；

(4-4) the high frequency topic obtaining step (4-2) and (4-3), as user's topic of interest word, sets up user interested Topic storehouse；

(5) microblog data that step (1) collects is divided short sentence it is ensured that each short sentence at most contains a topic of interest Word；

(6) extract the Emotion element in each short sentence, calculate the initial emotion value of each short sentence:

(6-1) set of words of short sentence is carried out mating mapping with each dictionary and emoticon storehouse, mark all kinds of Emotion elements； Described Emotion element includes emotion word, degree word, negative word, punctuation mark, emoticon, wherein degree word and punctuation mark all For adjusting the degree of emotion word, negative word is used for adjusting the polarity of emotion word；

(6-2) the emotion value of calculating short sentence Chinese version:

The weights of setting Emotion element: positive emotion word weights are "+1 ", negative emotion word weights are " -1 "；Negative word weights are “-1”；Degree word and punctuation mark, according to the depth of its degree, arrange weights, and weights scope is between 0 to 3；Degree word and mark The emotion word that point symbol is affected follows nearby principle, and that is, each degree word or punctuation mark affect apart from its nearest emotion word Emotion degree；

Emotion value i of short sentence Chinese version_wordsComputational methods be:

$i_{words}=b\&centerdot;\underset{i=1}{\overset{m}{\&sigma;}}(\underset{j=1}{\overset{n}{\&sigma;}}{c}_{ij}\&centerdot;f_{ij})\&centerdot;q_i$

In formula, q_iRepresent i-th emotion word, c_ijRepresent and modify q_iJ-th degree word weights, f_ijRepresent and modify q_iJ-th no Determine word weights；If q_iDo not attach degree word, then c_ijTake default value 1；If q_iDo not attach negative word, then f_ijTake default value 1；n Take modification q_iDegree word number and modify q_iNegative word number in maximum, m represents the number of emotion word, and b represents punctuate The corresponding weights of symbol, i, j are positive integer；

(6-3) calculate short sentence in emoticon emotion value:

The expression providing for microblogging operator, it is divided into front for the contribution of Sentiment orientation, negatively, neutral three kinds of feelings Condition: the weights of front emoticon are set to "+1 ", the weights of negative emoticon are set to " -1 ", and the weights of neutral expression's symbol set For " 0 "；

Emoticon emotion value i in short sentence_marksComputational methods be:

$i_{marks}=\underset{i=1}{\overset{l}{\&sigma;}}{m}_i$

In formula, m_iRepresent the expression of i-th table front, negative or neutral emotion, i is positive integer, and l is emoticon number；

(6-4) calculate initial emotion value i of short sentence₀:

i₀=i_words+i_marks；

(7) to the text data after step (2) process, extract the word combination of high frequency using word slip window sampling, used Family personalization idiom list；

(8) statistical analysiss are carried out to the initial emotion value of all short sentences comprising each bar idiom, draw the emotion value of idiom；

For every idiom, find out all short sentences containing this idiom, by its initial emotion value sum-average arithmetic, computational methods As follows:

$i_g=\frac{1}{p}\underset{i=1}{\overset{p}{\&sigma;}}{i}_{0i}$

In formula, i_0iFor the initial emotion value of i-th short sentence comprising this idiom, p is the short sentence number containing this idiom, i_gFor The emotion initial value of this idiom；

By i_gValue be mapped in [- 3,3], obtain emotion value i of idiom '_g, record is in the personalized idiom feelings of this user In sense labelling table；

(9) calculate the individualized emotion value of each short sentence, computational methods are:

$i=i_0-\underset{i=1}{\overset{m^{\&prime;}}{\&sigma;}}(\underset{j=1}{\overset{n^{\&prime;}}{\&sigma;}}{c}_{gij}\&centerdot;f_{gij})\&centerdot;q_{gi}+\underset{k=1}{\overset{r}{\&sigma;}}{i}_{gk}^{\&prime;}$

In formula, i₀For the initial emotion value of short sentence, q_giRepresent i-th word, c_gijRepresent and modify q_giJ-th degree word weights, f_gijRepresent and modify q_giJ-th negative word weights；If q_giDo not attach degree word, then c_gijTake default value 1；If q_giNot attached Band negative word, then f_gijTake default value 1；N' takes modification q_giThe number of degree word and modify q_giThe number of negative word in Big value, m' represents the number of word, and i, j are positive integer；i'_gkRepresent the emotion value of k-th idiom, r represents in this short sentence The number of idiom；

(10) Sentiment orientation of calculating user's topic of interest:

For any user topic of interest word in user's topic of interest storehouse, it is calculated as follows its emotion value:

$i_{{\mathrm{topic}}_i}=\frac{1}{w}\underset{j=1}{\overset{w}{\&sigma;}}{i}_j$

i_jFor comprising the individualized emotion value of j-th short sentence of this user's topic of interest word, w is to comprise this user words interested The short sentence sum of epigraph,Emotion value for this user's topic of interest word；WillValue be mapped in [- 3,3], obtain The topic Sentiment orientation value of final user, using these values, sets up user individual microblog topic emotion value list.

2. microblog users individualized emotion trend analysis method according to claim 1 is it is characterised in that carry out step (10), after, also follow the steps below:

A () repeats step (1)～(7), for the idiom adding new in user individual idiom list, by step (8) it is calculated emotion value i of this idiom '_g；It is reported in user individual idiom list for before this circulation Idiom, then by the following method update i '_g: to every idiom, carry out step (8) first, obtain the i of epicycle circulation_gMeter Calculation value i_{g_new}If, i_{g_prev}The i obtaining for last round of circulation_gValue of calculation, then i_gIt is updated to:

i_g=ω₁i_{g_prev}+ω₂i_{g_new}

In formula, ω₁For i_{g_prev}Weights, ω₂For i_{g_new}Weights；

By i_gValue be mapped in [- 3,3], obtain emotion value i of idiom '_g；

B i ' that () obtains according to step (a)_g, carry out step (9)～(10).

3. microblog users individualized emotion trend analysis method according to claim 1 is it is characterised in that carry out step (10), after, also follow the steps below:

Repeat step (1)～(9), for the user's words interested in the new addition in user's topic of interest storehouse in this circulation Epigraph, is calculated the topic Sentiment orientation of user by the method for step (10)；Emerging for being reported in user's sense before this circulation User's topic of interest word in funny remarks exam pool, then update the topic Sentiment orientation i of user by the following method_topic: carry out first Step (10), obtains the i of epicycle circulation_topicValue of calculation i_{topic_new}If, i_{topic_prev}The i obtaining for last round of circulation_topic's Value of calculation, then i_topicIt is updated to:

i_topic=ω '₁i_{topic_prev}+ω'₂i_{topic_new}

ω'₁For i_{topic_prev}Weights, ω '₂For i_{topic_new}Weights.

4. the microblog users individualized emotion trend analysis method according to any one of claims 1 to 3 it is characterised in that After carrying out step (10), also follow the steps below:

According to following formula calculating user's overall emotion propensity value:

$i_{user}=\frac{1}{s}\underset{i=1}{\overset{s}{\&sigma;}}{i}_i$

In formula, i_iRepresent the emotion value of i-th short sentence, s represents short sentence sum.

5. microblog users individualized emotion trend analysis method according to claim 1 is it is characterised in that step (1) institute State all data of the microblogging homepage gathering each user, particularly as follows:

In units of user, collect all data in its homepage；Described data includes oneself issue in user home page face Micro- with user's of interest with the microblog data forwarding, the data of " comment sending ", the data of "@mentions mine ", microblogging name Rich name, self-introduction, spontaneous or forward the title in the url webpage and video link that comprise in microblogging.

6. microblog users individualized emotion trend analysis method according to claim 1 is it is characterised in that step (4-2) Described high frequency topic, by the aggregating algorithm upwards of the interlayer based on word frequency, is successively extracted according to topic tree, particularly as follows:

To each topic word, if its occurrence number is higher than the threshold value setting, this topic is high frequency topic, otherwise, this is talked about The occurrence number of epigraph passes to father's layer topic word, successively calculates and extracts high frequency topic；The occurrence number of topic word includes this topic The occurrence number of itself and the occurrence number of the sub-topic of this topic.

7. microblog users individualized emotion trend analysis method according to claim 1 is it is characterised in that step (5) institute State the data that step (1) is collected and divide short sentence, particularly as follows:

Each microblog data of analysis, if not having topic word or pertaining only to a topic, whole piece microblogging in a microblog data As a short sentence；

If one containing multiple topic word in microblog data, in conjunction with the analysis of punctuation mark degree of priority: if two distances are There is punctuate between big topic word, be then split into two short sentences at punctuate, if do not had between the topic word of two distance maximums There is punctuate, then checking time topic word of big distance, if all there is no punctuate, not splitting, whole piece microblog data is as one Short sentence；Described distance is to reach another topic word tree number to be passed through from a topic word in topic tree；If two There are multiple punctuates between topic word, then choose and divided at the high punctuate of relative importance value: in punctuation mark relative importance value, fullstop > point Number > comma.

8. microblog users individualized emotion trend analysis method according to claim 1 is it is characterised in that step (7) institute State the text data after step (2) is processed, extract the word combination of high frequency, particularly as follows:

If the length of window of sliding window is w, w is the word number that sliding window comprises, and successively takes 1,2,3,4 respectively；Using word Language sliding window counts in all short sentences, the total degree of each word or phrase appearance, will appear from the word that number of times is more than threshold value Or phrase lists user individual idiom list in；Collect daily statement word, set up daily statement phrase database；By daily statement word Reject from user individual idiom list, obtain user individual idiom list.

9. realize the microblog users individualized emotion trend analysis system of analysis method described in any one of claim 1～8, it is special Levy and be, including

Data acquisition module, for gathering all data of the microblogging homepage of each user, is stored in data base；

Word-dividing mode, for carrying out participle to the text data in the data collecting, obtains participle set and part-of-speech tagging；

Dictionary load-on module, for loading required dictionary, emoticon storehouse；Described dictionary includes hownet sentiment word lexicon, journey Degree dictionary, negates dictionary, personal pronoun dictionary, and function word is connected dictionary, cyberspeak dictionary and classified lexicon；

Module is set up in user's topic of interest storehouse, for using the interlayer based on word frequency upwards aggregating algorithm to set up user interested Topic storehouse；

Short sentence division module, for by data collecting module collected to data divide short sentence it is ensured that each short sentence at most contains There is a topic of interest word；

Emotion element extraction module, for extracting the Emotion element in each short sentence, calculates the initial emotion value of each short sentence；

Module is set up in user individual idiom list, for the data that data acquisition module is collected, is slided using word Window technique extracts the word combination of high frequency, obtains user individual idiom list；

Idiom emotion value computing module, for statistical analysiss are carried out to the emotion value of all short sentences comprising each bar idiom, Draw the emotion value of idiom；

Short sentence emotion value computing module, for calculating the individualized emotion value of each short sentence；

User's topic Sentiment orientation computing module, for calculating the Sentiment orientation of each topic of interest of user.

10. microblog users individualized emotion trend analysis system according to claim 9 is it is characterised in that also include using Family overall emotion tendency computing module, the overall emotion for calculating user is inclined to.