CN103324662A - Visual method and equipment for dynamic view evolution of social media event - Google Patents

Abstract

The invention provides a visual method for the dynamic view evolution of a social media event. The method comprises the following steps: the emotional membership degree and emotional classification of information in the collection of collected social media event information is determined firstly; then the geometrical layout of an emotional visual graph is built based on the emotional classification of the information, and the built geometrical layout is colored based on the emotional membership degree of the information. The method can intuitively display the emotional change of the event in an information stream, can display more emotional information in the aspects of the change and development of the invent, can help a user recognize the turning point and the bursting point of the event better, and predicts the development trend of the event.

Description

Method for visualizing and equipment that the dynamic viewpoint of Social Media event develops

Technical field

The invention belongs to Web mining and analysis field, relate in particular to viewpoint analysis and the method for visualizing of the various information flows of serving as theme with the Social Media event.

Background technology

Along with the fast development of web2.0 technology, so that increasing people can deliver by platforms such as Twitter, microbloggings their suggestion and idea to the development of event and topic.Content with emotion has reflected the development of event and people's reaction, and can see the development trend of outgoing event.Sentiment analysis has very important effect in this analytic process.Sentiment analysis has comprised emotional semantic classification, and viewpoint extracts and opinion mining, the parts such as score in predicting.The result of sentiment analysis represents to the user with the form of chart usually, but but can not help the user better to understand the evolution of event, unless user oneself goes to read a large amount of non-structured data about this event.

Be to help the user and better understand data by promoting Man machine interaction with the purpose of the result visualization of sentiment analysis, find development trend and the turning point of event.List of references 1 has proposed the different emotions that viewpoint trigonometric sum ring comes visual hotel to comment on.List of references 2 has proposed " suggestion piece " method for visualizing based on a product different aspect by the multianalysis to viewpoint.List of references 3 usefulness carry out visual based on mood piece and the high density geology figure of pixel.The method that list of references 4 has proposed a kind of simple oriented path is described the time relationship between the event emotion.But these method for visualizing major parts are based on pie chart or histogram etc., emotion temporal evolution and variation tendency that can not presented event.

The list of references tabulation:

List of references 1, Y.Wu, F.Wei, S.Liu, N.Au, W.Cui, H.Zhou, and H.Qu, OpinionSeer:Interactive Visualization of Hotel Customer Feedback, IEEE Trans.16 (6): 1109 – 1118,2010.

List of references 2, B.Alper, H.Yang, E.Haber and Kandogan, OpinionBlocks:Visualizing Consumer Reviews, IEEE VisWeek2011Workshop on Interactive Visual Text Analytics for Decision Making.

List of references 3, M.Hao, C.Rohrdantz, H.Janetzko, U.Dayal, D.Keim, L.Haug and M.Hsu, Visual Sentiment Analysis on Twitter Data Streams.IEEE Symposium on Visual Analytics Science and Technology.2011.

List of references 4, D.Das, A.Kolya, A.Ekbal, S.Bandyopadhyay, Temporal analysis of sentiment events:a visual realization and tracking.CICLing'11Proceedings of the12th international conference on Computational linguistics and intelligent text processing.

Summary of the invention

Therefore, the object of the invention is to overcome the defective of above-mentioned prior art, provide a kind of effective, more intuitively the emotion method for visualizing, help the user to understand the dynamically differentiation of viewpoint in the event evolution, the turning point of identification event and the development trend of predicted events etc.

The objective of the invention is to be achieved through the following technical solutions:

On the one hand, the invention provides a kind of method for visualizing of dynamic viewpoint differentiation of Social Media event, described method comprises:

Emotion degree of membership and the emotional semantic classification of information in the Social Media event information set that step 1) is determined to gather, the emotion degree of membership of described information represents that this information belongs to a certain emotional semantic classification with much probability;

Step 2) based on the emotional semantic classification of described information, set up the geometric layout of the emotion visualized graphs of described information aggregate, in the described geometric layout, represent the time that information produces with transverse axis, represent to belong to the quantity of the information of each emotional semantic classification with the longitudinal axis;

Step 3) is carried out painted based on the emotion degree of membership of described information to the geometric layout of setting up.

In the said method, described step 1) can comprise:

Step 11) is selected small part information and is carried out the artificial emotion mark from described Social Media event information set;

Step 12) is used for carrying out the sorter of emotional semantic classification based on these information training that marked;

Step 13) uses the sorter train not mark all the other information and classify carrying out artificial emotion in the described Social Media event information set, determines emotion degree of membership and the emotional semantic classification of every information.

In the said method, described step 12) can be trained described sorter with text view feature and the non-text view feature of information, wherein, the vector that the pointwise mutual information value based on information retrieval of emotion word forms in this information of text view feature of information characterizes, and the non-text view feature of information comprises emoticon feature, temporal characteristics and punctuation mark feature.

In the said method, in described step 2) in described geometric layout, the curvilinear function of the bottom is S ₀, the curvilinear function of i layer emotional semantic classification is:

$S_i={\mathrm{<figure-callout\; id="0"\; label="S"\; filenames="DEST\_PATH\_HDA00003198868000012.png,HDA00003065001300021.png"\; state="\{\{state\}\}">S</figure-callout>}}_0+{\mathrm{\&Sigma;}}_{j=1}^i{\mathrm{\&rho;}}_j$

Wherein, density function ρ _jFor in the unit interval section, belonging to the information content of j kind emotional semantic classification, 1≤i≤n; S ₀=0 or N is the quantity of emotional semantic classification.

In the said method, described step 3) can comprise:

Step 31) sets up mapping between the color of the emotion degree of membership of described information and information;

Step 32) on the geometric layout of setting up, sequentially be that information on each emotional semantic classification layer is painted according to the gradual change of information color.

In the said method, described emotional semantic classification is positive, neutral or passive.

In the said method, described step 31) can adopt set up such as minor function as described in mapping between the color of the emotion degree of membership of information and information:

$\mathrm{RGB}\left(t\right)=\left\{\begin{array}{c}((1-n\left(t\right))*\mathrm{255,255,0}),p\left(t\right)\&GreaterEqual;n\left(t\right)\\ (255,(1-p\left(t\right))*\mathrm{255,0}),p\left(t\right)<n\left(t\right)\end{array}\right.$

Wherein, t represents the arbitrary information in the set of described Social Media event information, and n (t) represent the degree of membership of this information t to passive emotional semantic classification, and p (t) represents the degree of membership that this message t classifies to positive emotion.

In the said method, can comprise that also step 4) is at the emotion visualized graphs mark emotion label of setting up, the emotion word of described emotion label for occurring in described Social Media event information.

In the said method, in the described step 4), the font size of described emotion label on the visualized graphs that described step 3) obtains calculated with following formula:

F(w)=α*PMI-IR(w)*f(w)

Wherein, indicate arbitrary emotion word of showing at described visualized graphs of w, f (w) is the frequency that emotion word w occurs in the information of every kind of emotional semantic classification, and PMI-IR (w) represents the pointwise mutual information value based on information retrieval of this emotion word w, and α is scale factor.

Another aspect the invention provides the visualization device that a kind of dynamic viewpoint of Social Media event develops, and comprising:

Be used for the emotion degree of membership of definite Social Media event information set information that gathers and the device of emotional semantic classification, the emotion degree of membership of described information represents that this information belongs to a certain emotional semantic classification with much probability;

Be used for the emotional semantic classification based on described information, set up the device of geometric layout of the emotion visualized graphs of described information aggregate, in the described geometric layout, represent the time that information produces with transverse axis, represent to belong to the quantity of the information of each emotional semantic classification with the longitudinal axis;

Be used for based on the emotion degree of membership of described information the geometric layout of setting up being carried out painted device.

Compared with prior art, the invention has the advantages that:

The emotion that can show intuitively event in the information flow changes, comprise extraction of the intensity of emotion and evolution, viewpoint label etc., thereby can show more emotion information aspect the change and progress of event, can help the user better to identify turning point and the bursting point of event, the development trend of predicted events.

Description of drawings

Embodiments of the present invention is further illustrated referring to accompanying drawing, wherein:

Fig. 1 is the schematic flow sheet according to the method for visualizing of the embodiment of the invention;

Fig. 2 is the geometric layout schematic diagram according to the method for visualizing of the embodiment of the invention;

Fig. 3 is the signal according to the resulting sentiment analysis visualization view of the method for visualizing of the embodiment of the invention;

Fig. 4 is for adopting the sentiment analysis visualization result schematic diagram according to the method for visualizing of the embodiment of the invention;

Fig. 5 is for adopting the sentiment analysis visualization result schematic diagram of classic method.

Embodiment

In order to make purpose of the present invention, technical scheme and advantage are clearer, and the present invention is described in more detail by specific embodiment below in conjunction with accompanying drawing.Should be appreciated that specific embodiment described herein only in order to explain the present invention, is not intended to limit the present invention.

In one embodiment of the invention, the method for visualizing that provides a kind of dynamic viewpoint of Social Media event to develop can show intuitively that the emotion of event in the information flow changes, and comprises the intensity of emotion and evolution etc.Wherein, the Social Media event in fact be exactly Twitter or microblogging etc. above the platform about information such as the tweet in continuous time stream of different event topic or microbloggings.The method comprises the following steps: that mainly step 1) carries out emotional semantic classification to the information in the set of the Social Media event information that gathers, namely determines the emotion degree of membership of every information and the emotional semantic classification under this information; Step 2) sets up the geometric layout of the emotion visualized graphs of described information aggregate; Step 3) is carried out painted to the geometric layout of setting up, so that emotion intensity is visual; The emotion visualized graphs mark emotion label that step 4) is being set up.

Hereinafter with from the TREC2011 microblogging data acquisition from 41096 microbloggings in year September in June, 2008 to 2009 relevant " Obama " example as the set of Social Media event information, the step of the method for visualizing that the dynamic viewpoint of Social Media event is according to an embodiment of the invention developed is discussed in more detail.

More specifically, every information is carried out emotional semantic classification in step 1) is gathered the Social Media event information.In fact information is carried out emotional semantic classification calculates the degree of membership that this information belongs to different emotions classification (can referred to as the emotion degree of membership) and determines the described emotional semantic classification of this information (also can be described as the emotion classification).The emotion degree of membership of every information refers to that this information belongs to a certain emotional semantic classification with much probability.For example, suppose to adopt three kinds of emotional semantic classifications: positive, neutral and passive, can obtain three kinds of different emotions degrees of membership for information t so: positive p (t), neutral m (t), passive n (t).They satisfy following formula:

p(t)+m(t)+n(t)＝1。

And the emotion classification of information t is the emotion classification of the maximal value representative in these 3 degrees of membership, i.e. t ∈ max (p (t), m (t), n (t)).For example, if m (t)〉p (t)〉n (t), represent that then information t is the information of neutral affective style.Whole information aggregate is carried out the artificial emotion classification wastes time and energy, therefore usually can adopt the sorting technique based on machine learning, namely utilize certain incompatible training classifier of the data set that has marked, then utilize the sorter train to realize emotional semantic classification to every information in the whole information aggregate.

In one embodiment, can from above-mentioned 41096, evenly select 2250 information and carry out the artificial emotion mark, for example, be labeled as actively, neutral or passive.Then based on these micro-blog information training classifiers that has marked, and use the sorter that trains to carry out emotional semantic classification to 38846 microbloggings that do not carry out the artificial emotion mark, obtain simultaneously the emotion degree of membership of every micro-blog information.Wherein, sorter can adopt the typical disaggregated model such as Logic Regression Models, decision tree, naive Bayesian method, k-arest neighbors method, support vector machine (Support Vector Machine, SVM) model, random forest disaggregated model.When train classification models, most critical be which feature to select sample and training pattern with, so-called train classification models be exactly utilize sample data and the feature from sample data, extracted by training to learn out the parameter in the disaggregated model, thereby the disaggregated model that obtains training.In this embodiment, mainly come training classifier with text view feature and the non-text view feature of Social Media event information.

Wherein, the vector that the text view feature of information can form with the PMI-IR value of emotion word in this information characterizes, and wherein the PMI-IR value of emotion word is the pointwise mutual information based on information retrieval (the Point-wise Mutual Information and Information Retrieval) weight of this emotion word.The non-text view feature of information comprises emoticon feature, temporal characteristics and punctuation mark feature.The below introduces the text view feature of obtaining information and the step of non-text view feature simply:

The text view feature of step (11) obtaining information.The vector that the text view feature of information is comprised of the PMI-IR value of emotion word in this information.Emotion word in every information can utilize sentiment dictionary to identify, and the emotion word in the sentiment dictionary can be collected from known WordNet and HowNet usually.Account form for the PMI-IR of any emotion word w among the emotion set of words P of this information is as follows:

$\mathrm{PM}-\mathrm{IR}\left(w\right)={\log }_2\left[\frac{\mathrm{htts}\left(\mathrm{wNEARexcellent}\right)\mathrm{htts}\left(\mathrm{poor}\right)}{\mathrm{htts}\left(\mathrm{wNEARpoor}\right)\mathrm{htts}\left(\mathrm{excellent}\right)}\right]w\&Element;p$

Hits () is the number that sends the Query Result of request acquisition to the AltaVista search engine in following formula, wherein w NEAR excellent has used the NEAR operational character of AltaVista, they be exactly that return results will comprise w and excellent, and the word of these two words distance can not be above 10.For example, utilize the emotion word dictionary of WordNet and HowNet that 41096 microbloggings are carried out the identification of emotion word, then these emotion words are calculated its PMI-IR value, thereby can obtain the text view feature of every micro-blog information.As in the microblogging data of in June, 2008 about Obama, identify following emotion word: " incredible beautiful thank amazing moderate good admire adorable liked loser dissed shit fucking gay dead worst kill idiot sock stupid ", calculate the PMI-IR value of each emotion word, can get (as shown in table 1):

Table 1

The non-text view feature of step (12) obtaining information.The non-text view feature of information comprises temporal characteristics, emoticon feature and punctuation mark feature.Can obtain as follows these features:

For temporal characteristics, what temporal characteristics reflected is the time of delivering information; Because people are in the morning and evening, the beginning of the month and the end of month, spring with different behaviors is often arranged winter, therefore the time of delivering information can be divided into based on hour, day, week and month temporal characteristics.For example, issuing time that can obtaining information, then according to predefined 15 time periods: evening at noon (totally 4 time periods) morning, Monday to Sunday (totally 7 time periods), month which in week (being arranged, totally 4 time periods 4 weeks every month), giving the time period feature assignment on the correspondence is 1, otherwise be 0, form the 0-1 vector of 15 dimensions, namely consist of temporal characteristics.

For the emoticon feature, be based on the emoticon table and obtain, every kind of corresponding emotion value of emoticon, namely positive emotion is+1, and passive emotion is-1, and neutrality is 0.For example (specifically the content in this storehouse is referring to this network address: http://en.wikipedia.org/wiki/List_of_emoticons) extract all emoticons from this information based on the expressive features storehouse that obtains from wikipedia Wikipedia, such as: (＜_): [etc., their emotion label is designated as actively (+1), neutral (0), passive (1).Then can be with the corresponding numerical value sum of the emoticon that comprises in every information as the emoticon feature, do not contain emoticon, its eigenwert is set to 0.For example, be-1 for " Barack obama piss off:(" its emoticon eigenwert; For " I love Obama.:-) ", its emoticon eigenwert is 1.

For the punctuation mark feature, emphasize identifier (! ), query identifier () and their mixing or the emotion that repeats to give expression to are stronger.The frequency that therefore, each punctuation mark can be occurred in information is as the value of this feature.For example can obtain by the traversal text data and comprise! With number extract the punctuation mark feature.

Just can extract text view feature and the non-text view feature of every information by said method.Each Feature Combination of every information can be become a total proper vector in step 1), utilize 2250 microbloggings that carried out the artificial emotion mark that random forest (Random forests) sorter is trained.Then, use the random forest sorter after training, based on the proper vector of every micro-blog information 38846 microbloggings that do not carry out the artificial emotion mark are carried out emotional semantic classification, obtain simultaneously the emotion degree of membership of every micro-blog information.For example, the emotion degree of membership that sorter after training obtains microblogging "@Sisterbies good morning and thank you I think obama is beautiful. " is respectively p (t)=0.7595, m (t)=0.1288, n (t)=0.1117), then classification results belongs to " actively " emotion class (p (t)〉m (t)〉n (t)) for this microblogging.

Step 2) based on the emotional semantic classification to information in the described information aggregate, sets up the geometric layout of the emotion visualized graphs of described information aggregate

For example, add up respectively the quantity that belongs to the information of each emotion classification in the above-mentioned information aggregate, set up the geometric layout of the emotion visualized graphs of described information aggregate, in this geometric layout, represent the time that information produces with transverse axis, represent to belong to the quantity of the Social Media information of each emotion classification with the longitudinal axis.With the above-mentioned example that is exemplified as, the emotion classification be actively, neutral, passive three kinds, usually can arrange from top to bottom emotion classification layer at y direction so, the below is passive emotion classification, the centre is neutral emotion classification, the top is the positive emotion classification.These three emotion classification layers can be based on surface level and be arranged in order from bottom to top, also can be symmetrical with respect to emotion classification middle layer.

For example, the curvilinear function of supposing the visual figure bottom is S ₀, the curvilinear function of emotion classification i layer is:

$S_i={\mathrm{<figure-callout\; id="0"\; label="S"\; filenames="DEST\_PATH\_HDA00003198868000012.png,HDA00003065001300021.png"\; state="\{\{state\}\}">S</figure-callout>}}_0+{\mathrm{\&Sigma;}}_{j=1}^i{\mathrm{\&rho;}}_j---\left(1\right)$

Wherein, density function ρ _jFor the information content that belongs to emotion classification j distributes, also can be understood as the quantity of the Social Media information on this emotion layer in the unit interval section.Suppose that n kind emotion classification is arranged (three kinds of emotion classifications, i.e. n=3 for example mentioning in example above), the superiors' curvilinear function of so visual figure is S _nFor:

$S_n={\mathrm{<figure-callout\; id="0"\; label="S"\; filenames="DEST\_PATH\_HDA00003198868000012.png,HDA00003065001300021.png"\; state="\{\{state\}\}">S</figure-callout>}}_0+{\mathrm{\&Sigma;}}_{j=1}^n{\mathrm{\&rho;}}_j---\left(2\right)$

In general, look up from generous, emotion can be three classification systems (being n=3) usually, and is for example positive, neutral, passive.But understand, in the method for above introducing, the value of emotion classification n is not done special restriction, for example, n can value be 2, perhaps in order to make visual effect more careful, also can carry out thinner division to emotion, but for example the n value is 4,5,6 etc.Can find out by above-mentioned formula (1) and (2), the curve of each emotion classification i is to add up on the curve basis of several emotion classifications of its front to obtain, thereby can produce the effect of visualization that the emotion classification is arranged in order from bottom to up.

In above-mentioned formula, by with function S ₀=0 obtains the visual figure based on surface level; Pass through S ₀=-S _nCan obtain the visual figure based on space symmetr.These two figure are that height or the width by figure illustrates attention rate, and the quantity of Social Media information is more, and the height of figure is higher, represent that more people participate in discussing, and the attention rate of part is higher as to this within this time period.However, by observing the up and down variation of eutral zone, the visual figure of space symmetr can more directly express the ratio of positive and passive emotion, but can not based on the visual figure of surface level.Therefore, in a preferred embodiment, with bottom curvilinear function S ₀Mathematic(al) representation be set to following formula (3), thereby obtain the geometric layout with respect to the visualized graphs of emotion middle layer symmetry:

${\mathrm{<figure-callout\; id="0"\; label="S"\; filenames="DEST\_PATH\_HDA00003198868000012.png,HDA00003065001300021.png"\; state="\{\{state\}\}">S</figure-callout>}}_0=-\frac{1}{2}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&rho;}}_j---\left(3\right)$

The geometric layout of the according to an embodiment of the invention emotion visualized graphs that Fig. 2 has provided.As shown in Figure 2, transverse axis represents the time that information produces, and the longitudinal axis represents the quantity of Social Media information.The geometric layout of this visual figure adopts symmetrical form, and the middle layer is the quantity of holding the people of " neutrality " viewpoint, is respectively up and down the people's of " actively " and " passiveness " quantity, is the relation of a symmetry about the middle layer.The thickness of every one deck just reflects the classification situation of what and the emotion thereof that participate in the number discussed in this time period.

Step 3) is carried out painted to the geometric layout of the emotion visualized graphs set up, so that emotion intensity is visual.

For make effect of visualization figure can not only pick out emotion actively, neutral, passive, can embody the power of emotion simultaneously, need to carry out each emotion classification layer in the above-mentioned visualized graphs painted so that emotion intensity is visual.In one embodiment, adopt the mapping function of a kind of color and emotion degree of membership, represented variation and the intensity thereof of emotion with this gradual change with color.Adjusted the RGB color model by this mapping function, so that three kinds of colors of RGB can be mixed arbitrarily the generation multiple color.That is to say that each color is all determined by these three elements, the value of each is all between 0 to 255 in the RGB.Mapping function between the non-negative affect degree of membership weight of the property value of color and information t is as giving a definition:

$\mathrm{RGB}\left(t\right)=\left\{\begin{array}{c}((1-n\left(t\right))*\mathrm{255,255,0}),p\left(t\right)\&GreaterEqual;n\left(t\right)\\ (255,(1-p\left(t\right))*\mathrm{255,0}),p\left(t\right)<n\left(t\right)\end{array}\right.---\left(4\right)$

For following three kinds of situations:

RGB(t)=(0,255,0) represent pure green; (a)

RGB(t)=(255,255,0) represent gilvous; (b)

RGB(t)=(255,0,0) expression is pure red; (c)

Wherein, in the formula (4), (((1-n (t)) * 255) is larger, and green is lighter when first parameter value; Represent that such is divided into " actively ", but to be not " actively " that very large probability is divided into, i.e. p (t) 〉=n (t) and m (t)＜1; ((1-p (t)) * 255) is larger when the second parameter value, and redness is lighter; Representing that such is divided into " passiveness ", still, is not to be " passiveness " that very large probability is divided into, i.e. p (t)＜n (t).Consider that positive and passive emotion opposes, so when information t belonged to positive or passive emotion classification, degree of membership p (t) can not equal n (t), i.e. the emotion of every information or be positive, or be passive.When information t belongs to neutral emotion classification, if p (t) equals n (t), then directly make m (t)=1, p (t)=n (t)=0 is as absolute neutrality.Therefore, the absolute neutrality figure layer that mediates, color is yellow (2), the RGB tuple values is (255,255,0).

For example, pass through said method, obtain visual figure after painted shown in Figure 3 to the figure layer of geometric layout figure shown in Figure 2 is painted from top to bottom: uppermost is green, be positive class, (top layer, p (t)=1.0 but green shoals gradually, pure green), and then represent that those are to be divided into positive class, but this probability is more and more less, namely the degree of " actively " more and more a little less than.Middle transition is to neutral yellow, middle gilvous represents it is those tweet that are divided into neutrality with 1.0 probability, and yellow top part is exactly to be divided into neutrality, but p (t)〉n (t), yellow part on the lower is exactly to be divided into neutrality, but p (t)＜n (t).Nethermost is red, i.e. passive class, but redness deepens gradually, represents that namely those are to be divided into passive class, and this probability is increasing, and namely the degree of " passiveness " is more and more stronger, until undermost n (t)=1.0 is pure red.

The emotion visualized graphs mark emotion label that step 4) is being set up.

The emotion word that occurs in the Social Media event information can be presented on this emotion visualized graphs as the emotion label, and can reflect by the font size of emotion label the frequency of emotion word appearance and the emotion intensity of emotion word, in this visualized graphs, the font size F (w) of emotion label w is calculated as follows:

F(w)=α*PMI-IR(w)*f(w)

Wherein f (w) is the frequency that emotion word w occurs in the information of every kind of emotional semantic classification, PMI-IR (w) represents the pointwise mutual information value based on information retrieval of this emotion word w, α is scale factor, the α size has relation with the ratio of visual figure, for example, if displayable maximum font is No. 9 words, then can be set to 9 by α.

For example, suppose that the emotion label is " incredible, beautiful, wrong, idolt " in the above-mentioned result that the microblogging data set in June, 2008 is processed.Can be calculated the font size (shown in the table 2) of each emotion label character according to the font size formula, the final emotion visualized graphs that generates as shown in Figure 3 again.

Table 2

Should be understood that above-mentioned step 4) is optional, at the emotion visualized graphs mark emotion label of setting up, is in order further to show more emotion information on this visual image, to help the better development trend of predicted events of user.Show the emotion word of all appearance on the visualized graphs that can set up, also can an emotion word that the display part frequency of occurrences is high and the PMI-IR value is large.

What Fig. 4 represented is from the tweet analysis result of year May in June, 2008 to 2009 about Obama.Fig. 4 has shown the emotional evolution process about Obama, the width that has shown in November, 2008 stream on the figure increases, this explanation has reached the peak about the tweet quantity of Obama, variation among the figure can attribution and on November 5th, 2008 Obama defeated John McCain, be elected as the 44th president of the U.S., and declare elected.In addition, in January, 2009, green figure layer has surpassed red figure layer, shown that most tweet user has delivered positive content, meanwhile, Obama has delivered inaugural speech, to such an extent as to can think that his inspiring and very successful a lot of people of inaugural speech support him.

Contrast with traditional method for visualizing shown in Figure 5 such as pie chart and histogram, such as the pie chart of Fig. 5 the ratio that each emotion is shared can be described clearly, but the intensity of emotion but can not be described, and not embody the fluctuations of emotion.Histogram is concentrated the situation of change of data in the different time sequence has been described.Color gradient in the histogram can represent the intensity of emotion, but the problem that it faces is the same with visual figure based on surface level, understanding such histogram needs the user that emotion is changed and time correlation, in addition, be difficult to allow from histogram and see emotion differentiation in time, because time window is divided by manual type.Therefore, more limited based on the information that pie chart and histogrammic method for visualizing provide, can only provide the event emotion visual angle of macroscopic view relatively.Of the present invention visual figure shown in Figure 4 makes emotion intensity visual by the variation of color gradient, based on the tweet density function of sequence of events so that the prediction of the turning point of event and bursting point and event is all more directly perceived.And emotion word label also reflects emotion intensity.On the whole, provided by the invention visual figure shown in Figure 4 has embodied people to be changed along with the time the emotion of Obama, particularly when some influential events occur, final these variations tend towards stability, the tweet(that at this time emotion is strong pushes away the spy) also become seldom, in view of this, help the user better to understand emotion in the Strength Changes of different time and the trend that develops in time according to the visual figure of the emotion of the embodiment of the invention.

In yet another embodiment of the present invention, a kind of visualization device of dynamic viewpoint differentiation of Social Media event also is provided, comprise: be used for the emotion degree of membership of definite Social Media event information set information that gathers and the device of emotional semantic classification, the emotion degree of membership of described information represents that this information belongs to a certain emotional semantic classification with much probability; Be used for the emotional semantic classification based on described information, set up the device of geometric layout of the emotion visualized graphs of described information aggregate, in the described geometric layout, represent the time that information produces with transverse axis, represent to belong to the quantity of the information of each emotional semantic classification with the longitudinal axis; Be used for based on the emotion degree of membership of described information the geometric layout of setting up being carried out painted device.This equipment can also comprise for the device at the emotion visualized graphs mark emotion label of setting up.

Although the present invention is described by preferred embodiment, yet the present invention is not limited to embodiment as described herein, also comprises without departing from the present invention various changes and the variation done.

Claims (10)

1. the method for visualizing that develops of the dynamic viewpoint of a Social Media event, described method comprises:

Emotion degree of membership and the emotional semantic classification of information in the Social Media event information set that step 1) is determined to gather, the emotion degree of membership of described information represents that this information belongs to a certain emotional semantic classification with much probability;

Step 2) based on the emotional semantic classification of described information, set up the geometric layout of the emotion visualized graphs of described information aggregate, in the described geometric layout, represent the time that information produces with transverse axis, represent to belong to the quantity of the information of each emotional semantic classification with the longitudinal axis;

Step 3) is carried out painted based on the emotion degree of membership of described information to the geometric layout of setting up.

2. method according to claim 1, described step 1) comprises:

Step 11) is selected small part information and is carried out the artificial emotion mark from described Social Media event information set;

Step 12) is used for carrying out the sorter of emotional semantic classification based on these information training that marked;

Step 13) uses the sorter train not mark all the other information and classify carrying out artificial emotion in the described Social Media event information set, determines emotion degree of membership and the emotional semantic classification of every information.

3. method according to claim 2, text view feature and non-text view feature with information in the described step 12) are trained described sorter, wherein, the vector that the pointwise mutual information value based on information retrieval of emotion word forms in this information of text view feature of information characterizes, and the non-text view feature of information comprises emoticon feature, temporal characteristics and punctuation mark feature.

4. method according to claim 1 is in described step 2) in described geometric layout, the curvilinear function of the bottom is S0, the curvilinear function of i layer emotional semantic classification is:

$S_i=S_0+{\mathrm{\&Sigma;}}_{j=1}^i{\mathrm{\&rho;}}_j$

Wherein, density function ρ _jFor in the unit interval section, belonging to the information content of j kind emotional semantic classification, 1≤i≤n; S ₀=0 or N is the quantity of emotional semantic classification.

5. according to the described method of one of the claims, described step 3) comprises:

Step 31) sets up mapping between the color of the emotion degree of membership of described information and information;

Step 32) on the geometric layout of setting up, sequentially be that information on each emotional semantic classification layer is painted according to the gradual change of information color.

6. method according to claim 5, described emotional semantic classification be actively, neutral or passive.

7. method according to claim 6, described step 31) adopt set up such as minor function as described in mapping between the color of the emotion degree of membership of information and information:

$\mathrm{RGB}\left(t\right)=\left\{\begin{array}{c}((1-n\left(t\right))*\mathrm{255,255,0}),p\left(t\right)\&GreaterEqual;n\left(t\right)\\ (255,(1-p\left(t\right))*\mathrm{255,0}),p\left(t\right)<n\left(t\right)\end{array}\right.$

Wherein, t represents the arbitrary information in the set of described Social Media event information, and n (t) represent the degree of membership of this information t to passive emotional semantic classification, and p (t) represents the degree of membership that this message t classifies to positive emotion.

8. method according to claim 5 comprises that also step 4) is at the emotion visualized graphs mark emotion label of setting up, the emotion word of described emotion label for occurring in described Social Media event information.

9. method according to claim 8, in the described step 4), the font size of described emotion label on the visualized graphs that described step 3) obtains calculated with following formula:

F(w)=α*PMI-IR(w)*f(w)

Wherein, indicate arbitrary emotion word of showing at described visualized graphs of w, f (w) is the frequency that emotion word w occurs in the information of every kind of emotional semantic classification, and PMI-IR (w) represents the pointwise mutual information value based on information retrieval of this emotion word w, and α is scale factor.

10. the visualization device that develops of the dynamic viewpoint of a Social Media event, described equipment comprises:

Be used for the emotion degree of membership of definite Social Media event information set information that gathers and the device of emotional semantic classification, the emotion degree of membership of described information represents that this information belongs to a certain emotional semantic classification with much probability;

Be used for the emotional semantic classification based on described information, set up the device of geometric layout of the emotion visualized graphs of described information aggregate, in the described geometric layout, represent the time that information produces with transverse axis, represent to belong to the quantity of the information of each emotional semantic classification with the longitudinal axis;

Be used for based on the emotion degree of membership of described information the geometric layout of setting up being carried out painted device.

Images