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

Abstract

The present invention provides a visualization method for the evolution of dynamic viewpoints of social media events. The method first determines the emotional membership and emotional classification of information in the collected social media event information collection, and then establishes emotional visualization based on the emotional classification of the information. The geometric layout of the graph and color the established geometric layout based on the emotional membership of the information. This method can intuitively display the emotional changes of events in the information flow, and can display more emotional information in terms of event changes and developments, which can help users better identify the turning points and explosive points of events, and predict the development trend of events.

Description

Method and device for visualization of dynamic opinion evolution of social media events

technical field

The invention belongs to the field of Internet information mining and analysis, and in particular relates to viewpoint analysis and visualization methods of various information flows with social media events as the main line. the

Background technique

With the rapid development of web2.0 technology, more and more people can express their opinions and thoughts on the development of events and topics through Twitter, Weibo and other platforms. Emotional content reflects the development of events and people's reactions, and can see the development trend of events. Sentiment analysis plays a very important role in this analysis process. Sentiment analysis includes sentiment classification, opinion extraction and opinion mining, score prediction and other parts. The results of sentiment analysis are usually shown to users in the form of graphs, but they cannot help users better understand the development process of the event unless the user reads a large amount of unstructured data about the event. the

The purpose of visualizing the results of sentiment analysis is to help users better understand the data and discover the development trend and turning point of events by improving the performance of human-computer interaction. Reference 1 proposes opinion triangles and circles to visualize different sentiments of hotel reviews. Reference 2 proposes an “opinion block” visualization method based on different aspects of a product through a comprehensive analysis of opinions. Reference 3 uses pixel-based sentiment blocks and high-density geological maps for visualization. Reference 4 proposes a simple directed-path approach to describe temporal relationships among event emotions. However, most of these visualization methods are based on pie charts or histograms, etc., and cannot show the emotional changes and trends of events over time. the

Reference list:

Reference 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.

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

Reference 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 .

Reference 4, D. Das, A. Kolya, A. Ekbal, S. Bandyopadhyay, Temporal analysis of sentiment events: a visual realization and tracking. CICLing'11 Proceedings of the 12th international conference on Computational linguistics and intelligent text processing.

Contents of the invention

Therefore, the object of the present invention is to overcome the defects of the above-mentioned prior art, provide an effective and more intuitive emotional visualization method, help users understand the evolution of dynamic viewpoints in the event development process, identify the turning point of the event and predict the development trend of the event wait. the

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

On the one hand, the present invention provides a kind of visualization method of the dynamic viewpoint evolution of social media event, and described method comprises:

Step 1) Determine the emotional membership and emotional classification of the information in the collected social media event information collection, the emotional membership of the information indicates how likely the information belongs to a certain emotional classification;

Step 2) Based on the emotional classification of the information, establish the geometric layout of the emotional visualization graph of the information set. In the geometric layout, the horizontal axis represents the time when the information was generated, and the vertical axis represents the information belonging to each emotional classification. quantity;

Step 3) Color the established geometric layout based on the sentiment membership of said information. the

In the above method, the step 1) may include:

Step 11) Select a small part of information from the social media event information collection for artificial emotion labeling;

Step 12) Train a classifier for sentiment classification based on these labeled information;

Step 13) Use the trained classifier to classify the remaining information in the social media event information set that has not been artificially labeled, and determine the emotional membership and emotional classification of each piece of information. the

In the above method, the step 12) can use the text view features and non-text view features of the information to train the classifier, wherein the text view features of the information are the point-by-point mutual information values based on information retrieval of the emotional words in the information The non-text view features of information include emoji features, time features and punctuation features. the

In the above method, in the step 2) in the geometric layout, the curve function of the bottom layer is S ₀ , and the curve function of the i-th layer emotion classification is:

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

Among them, the density function ρ _j is the amount of information belonging to the jth emotion category within a unit time period, 1≤i≤n; S ₀ =0 or n is the number of sentiment categories.

In the above method, the step 3) may include:

Step 31) Establish a mapping between the emotional membership of the information and the color of the information;

Step 32) On the established geometric layout, color the information on each emotion classification layer according to the gradient order of the information color. the

In the above method, the sentiment is classified as positive, neutral or negative. the

In the above method, the step 31) can use the following function to establish the mapping between the emotional membership of the information and the color of the information:

$\mathrm{<span\; class="notranslate">\; RGB</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; 255,255,0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\\ <span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 255</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; 255,0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\end{array}\right.$

Wherein, t represents any information in the social media event information set, n(t) represents the membership degree of the information t to the negative emotion classification, and p(t) represents the membership degree of the message t to the positive emotion classification. the

The above method may further include step 4) labeling emotional tags on the established emotional visualization graph, where the emotional tags are emotional words appearing in the social media event information. the

In the above method, in the step 4), the font size of the emotional label on the visual graph obtained in the step 3) is calculated by the following formula:

F(w)=α*PMI-IR(w)*f(w)

Wherein, w represents any emotional word to be displayed on the visual graph, f(w) is the frequency of occurrence of emotional word w in the information of each emotional classification, and PMI-IR (w) represents the emotional word w Pointwise mutual information value based on information retrieval, α is a scaling factor. the

In yet another aspect, the present invention provides a visualization device for the evolution of dynamic viewpoints of social media events, including:

A device for determining the emotional membership and emotional classification of information in the collected social media event information collection, the emotional membership of the information indicates how likely the information belongs to a certain emotional classification;

A device for establishing the geometric layout of the emotional visualization graph of the information set based on the emotional classification of the information, in the geometric layout, the horizontal axis represents the time when the information is generated, and the vertical axis represents the information belonging to each emotional classification quantity;

Means for coloring the established geometric layout based on the sentiment membership of said information. the

Compared with prior art, the advantage of the present invention is:

It can intuitively display the emotional changes of events in the information flow, including the intensity and evolution of emotions, the extraction of opinion tags, etc., so that more emotional information can be displayed in terms of event changes and developments, and can help users better identify events. Turning points and breaking points, predicting the development trend of events. the

Description of drawings

Embodiments of the present invention will be further described below with reference to accompanying drawings, wherein:

Fig. 1 is a schematic flow chart of a visualization method according to an embodiment of the present invention;

Fig. 2 is a schematic diagram of a geometric layout of a visualization method according to an embodiment of the present invention;

Fig. 3 is the schematic diagram of the sentiment analysis visualization view that obtains according to the visualization method of the embodiment of the present invention;

Fig. 4 is a schematic diagram of a visualization result of sentiment analysis using a visualization method according to an embodiment of the present invention;

Fig. 5 is a schematic diagram of visualization results of sentiment analysis using traditional methods. the

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below through specific embodiments in conjunction with the accompanying drawings. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention. the

In one embodiment of the present invention, a visualization method for dynamic opinion evolution of social media events is provided, which can intuitively display the emotional changes of events in information streams, including the intensity and evolution of emotions. Among them, social media events are actually information such as tweets or microblogs in continuous time streams on different event topics on platforms such as Twitter or Weibo. The method mainly includes the following steps: Step 1) Sentiment classification of information in the collected social media event information collection, that is, determining the emotional membership degree of each piece of information and the emotional classification to which the information belongs; Step 2) Establishing the The geometric layout of the emotion visualization graph of the information set; step 3) color the established geometric layout to visualize the emotional strength; step 4) label the emotion labels on the established emotion visualization graph. the

From June 2008 to September 2009 in the TREC2011 microblog data collection, 41096 microblogs related to "Obama" are used as an example of social media event information collection below, and the social media event according to one embodiment of the present invention The steps of the visualization method for dynamic opinion evolution are discussed in more detail. the

More specifically, in step 1), sentiment classification is performed on each piece of information in the social media event information collection. Sentiment classification of information is actually to calculate the degree of membership of the information belonging to different emotion categories (may be referred to as emotion membership for short) and determine the emotion category (also referred to as emotion category) described by the information. The emotional membership degree of each piece of information refers to the probability that the information belongs to a certain emotional classification. For example, assuming that three sentiment classifications are adopted: positive, neutral and negative, then three different sentiment memberships can be obtained for information t: positive p(t), neutral m(t), and negative n(t). They satisfy the following formula:

p(t)+m(t)+n(t)=1. the

The emotional category of information t is the emotional category represented by the maximum value of the three membership degrees, that is, t∈max(p(t), m(t), n(t)). For example, if m(t)>p(t)>n(t), it means that information t is information of neutral emotion type. It is time-consuming and labor-intensive to perform artificial sentiment classification on the entire information set. Therefore, a classification method based on machine learning can usually be used, that is, a certain labeled data set is used to train the classifier, and then the trained classifier is used to realize the classification of the entire information set. Sentiment classification of each message in . the

In one embodiment, 2250 pieces of information can be evenly selected from the above 41096 for artificial sentiment labeling, for example, labeling as positive, neutral or negative. Then a classifier is trained based on these labeled microblog information, and the trained classifier can be used to classify the sentiment of 38846 microblogs that have not been artificially labeled, and at the same time obtain the emotional membership of each microblog information. Among them, the classifier can use typical classification models such as logistic regression model, decision tree, naive Bayesian method, k-nearest neighbor method, support vector machine (Support Vector Machine, SVM) model, random forest classification model, etc. When training a classification model, the most important thing is which features are used to select samples and train the model. The so-called training classification model is to use the sample data and the features extracted from the sample data to learn the parameters in the classification model through training, so as to be trained. Good classification model. In this embodiment, the classifier is mainly trained with text view features and non-text view features of social media event information. the

Among them, the text view feature of the information can be represented by a vector composed of the PMI-IR value of the emotional word in the information, where the PMI-IR value of the emotional word is the point-wise mutual information (Point-wise mutual information) based on information retrieval of the emotional word Mutual Information and Information Retrieval) weight. The non-text view features of information include emoji features, time features, and punctuation features. The following briefly introduces the steps to obtain the text view features and non-text view features of information:

Step (11) obtains the text view feature of the information. The text view feature of information is a vector composed of PMI-IR values of sentiment words in the information. The sentiment words in each piece of information can be identified by using a sentiment dictionary, and the sentiment words in the sentiment dictionary can usually be collected from well-known WordNet and HowNet. The calculation method of PMI-IR for any emotional word w in the emotional word set P of the information is as follows:

$\mathrm{<span\; class="notranslate">\; PM</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; IR</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; [</span>\frac{\mathrm{<span\; class="notranslate">\; htts</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; wNEARexcellent</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; htts</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; poor</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; htts</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w\; NEAR\; poor</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; htts</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; excellent</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ]</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; p</span>}$

In the above formula, hits() is the number of query results obtained by sending a request to the AltaVista search engine, where w NEAR excellent uses the NEAR operator of AltaVista, that is, the returned results must contain w and excellent, and the word distance between these two words Cannot exceed 10. For example, the emotional word dictionary of WordNet and HowNet is used to identify the emotional words of 41096 microblogs, and then calculate the PMI-IR value of these emotional words, so that the text view characteristics of each microblog information can be obtained. For example, in the microblog data about Obama in June 2008, the following emotional words were identified: "incredible beautiful thank amazing moderate good good admire adorable liked loser dissed shit fucking gay dead worst kill idiot sock stupid" and calculate the PMI of each emotional word -IR value, available (as shown in Table 1):

Table 1

Step (12) obtains the non-text view features of the information. The non-text view features of information include time features, emoji features, and punctuation features. These features can be obtained as follows:

For the time feature, the time feature reflects the time of publishing information; since people often have different behaviors in the morning and evening, at the beginning of the month and at the end of the month, in spring and winter, the time of publishing information can be divided into categories based on hours, days, and weeks. and month time features. For example, you can obtain the release time of the information, and then according to the 15 preset time periods: morning, noon and evening (4 time periods in total), Monday to Sunday (7 time periods in total), week of the month (every There are 4 weeks in a month, a total of 4 time periods), assign a value of 1 to the corresponding time period feature, otherwise it is 0, and form a 15-dimensional 0-1 vector, which constitutes a time feature. the

For the emoji feature, it is obtained based on the emoji table, and each emoji corresponds to an emotion value, that is, positive emotion is +1, negative emotion is -1, and neutral is 0. For example, based on the emoticon feature library obtained from Wikipedia (see this website for the content of the library: http://en.wikipedia.org/wiki/List_of_emoticons), extract all emoticons from the information, for example: (<_ >), >:[, etc., record their sentiment labels as positive (+1), neutral (0), negative (-1). Then, the sum of the numerical values corresponding to the emoticons included in each piece of information can be used as the feature of the emoticon, and if the emoticon is not included, the feature value is set to 0. For example, "Barack obama piss off:(" has an emoji feature value of -1; for "I love Obama.:-)" it has an emoji feature value of 1. the

For the punctuation feature, emphatic identifiers (!), interrogative identifiers (?), and their mixture or repetition express stronger emotions. Therefore, the frequency of each punctuation mark in the information can be used as the value of this feature. For example, the punctuation feature can be extracted by traversing the text data to obtain the number of contained ! and sums. the

The text view features and non-text view features of each piece of information can be extracted through the above method. In step 1), the individual features of each piece of information can be combined into a total feature vector, and the random forests (Random forests) classifier can be trained by using 2250 microblogs that have been labeled with artificial emotions. Then, using the trained random forest classifier, based on the feature vector of each microblog information, sentiment classification is carried out on 38,846 microblogs that have not been artificially labeled, and the emotional membership of each microblog information is obtained at the same time. For example, after training, the classifier can obtain the emotional membership of the microblog "Sisterbies good morning and thank you I think obama is beautiful." p(t)=0.7595, m(t)=0.1288, n(t)= 0.1117), the classification result is that this microblog belongs to the "positive" emotional category (p(t)>m(t)>n(t)). the

Step 2) Based on the emotional classification of the information in the information set, establish the geometric layout of the emotional visualization graphics of the information set

For example, the quantity of information belonging to each emotion category in the above-mentioned information collection is counted separately, and the geometric layout of the emotional visualization graph of the information collection is established. The number of social media messages in the sentiment category. Taking the above example as an example, the emotional categories are positive, neutral, and negative. Then, the emotional category layers can usually be arranged from top to bottom in the direction of the vertical axis. The bottom is the negative emotion category, the middle is the neutral emotion category, and the top category of positive emotions. The three emotion category layers can be arranged from bottom to top based on the horizontal plane, or can be symmetrical with respect to the middle layer of emotion category. the

For example, suppose the curve function at the bottom layer of the visualization graph is S ₀ , and the curve function at the i-level of the emotion category is:

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

Among them, the density function _ρj is the distribution of the quantity of information belonging to the emotional category j, and it can also be understood as the quantity of social media information on the emotional layer within a unit time period. Assuming that there are n kinds of emotional categories (such as the three emotional categories mentioned in the above example, n=3), then the uppermost curve function of the visualization graph is S _n as:

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; <figure-callout\; id="0"\; label="S"\; filenames="DEST\_PATH\_HDA00003198868000012.png,HDA00003065001300021.png"\; state="\{\{state\}\}">S</figure-callout></span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&rho;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

Generally speaking, from a general perspective, emotions can usually be classified into three categories (ie n=3), such as positive, neutral, and negative. However, it is understood that in the method introduced above, there is no special restriction on the value of the emotion category n. For example, n can take a value of 2, or in order to make the visualization effect more detailed, the emotion can also be divided into finer details. , for example n can be 4, 5, 6 and so on. From the above formulas (1) and (2), it can be seen that the curve of each emotional category i is accumulated on the basis of the curves of several previous emotional categories, so that the emotional categories can be arranged from bottom to top Visualization. the

In the above formula, the visualization graph based on the horizontal plane can be obtained by setting the function S ₀ =0; the visualization graph based on space symmetry can be obtained by S ₀ =-S _n . These two graphs illustrate attention through the height or width of the graph. The more social media information, the higher the height of the graph, indicating that more people participated in the discussion. During this period of time, the attention to this event The higher the degree. Nevertheless, spatially symmetric visualizations can more directly represent the ratio of positive and negative emotions by observing the up and down changes of the neutral layer, while the horizontal plane-based visualization cannot. Therefore, in a preferred embodiment, the mathematical expression of the underlying curve function _S0 is set to the following formula (3), so as to obtain the geometric layout of the visual graph symmetrical to the emotional middle layer:

${\mathrm{<span\; class="notranslate">\; <figure-callout\; id="0"\; label="S"\; filenames="DEST\_PATH\_HDA00003198868000012.png,HDA00003065001300021.png"\; state="\{\{state\}\}">S</figure-callout></span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&rho;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

Fig. 2 shows the geometric layout of emotion visualization graphics according to an embodiment of the present invention. As shown in Figure 2, the horizontal axis represents the time when the information was generated, and the vertical axis represents the amount of social media information. The geometric layout of the visualized diagram adopts a symmetrical form. The middle layer is the number of people holding "neutral" views, and the upper and lower layers are the number of "positive" and "negative" people respectively. There is a symmetrical relationship about the middle layer. The thickness of each layer reflects the number of people participating in the discussion and the classification of their emotions during the time period. the

Step 3) Color the geometric layout of the built emotion visualization graph to visualize emotion strength. the

In order to make the visualization effect map not only identify the positive, neutral, and negative emotions, but also reflect the strength of emotion, it is necessary to color the layers of each emotion category in the above visualization graph to visualize the intensity of emotion. In one embodiment, a mapping function of color and emotion membership is used, so that the color gradient is used to represent the change and intensity of emotion. The RGB color model is adjusted through the mapping function, so that the three colors of red, green and blue can be mixed arbitrarily to produce multiple colors. That is to say, each color is determined by these three elements, and the value of each of red, green and blue is between 0 and 255. The mapping function between the attribute value of color and the non-negative sentiment membership weight of information t is defined as follows:

$\mathrm{<span\; class="notranslate">\; RGB</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; 255,255,0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\\ <span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 255</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; 255,0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

For the following three situations:

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

RGB(t)=(255,255,0) means pure yellow; (b)

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

Among them, in formula (4), when the first parameter value (((1-n(t))*255) is larger, the green is lighter; it means that the class is classified as "positive", but it is not very The large probability is divided into "positive", that is, p(t)≥n(t) and m(t)<1; when the second parameter value ((1-p(t))*255) is larger, the red The lighter it is; it means that this class is divided into "negative", but it is divided into "negative" with a not very high probability, that is, p(t)<n(t). Considering that positive and negative emotions are Opposite, so when the information t belongs to the positive or negative emotion category, the degree of membership p(t) cannot be equal to n(t), that is, the emotion of each piece of information is either positive or negative. When the information t belongs to the neutral For the emotional category, if p(t) is equal to n(t), then directly set m(t)=1, p(t)=n(t)=0, as absolute neutrality. Therefore, absolute neutrality is in the middle layer, The color is yellow (2) and the RGB tuple value is (255,255,0).

For example, through the above method, color the layers of the geometric layout graph shown in Figure 2 from top to bottom to get the colored visualization shown in Figure 3: the top is green, that is, the positive class, but the green gradually becomes lighter (The top layer, p(t)=1.0, pure green), followed by those who are classified as positive, but this probability is getting smaller and smaller, that is, the degree of "positive" is getting weaker and weaker. The middle transitions to neutral yellow. The middle pure yellow indicates those tweets that are classified as neutral with a probability of 1.0. The part above the yellow is classified as neutral, but p(t)>n(t), and the part below the yellow is classified as neutral. Part is divided into neutral, but p(t)<n(t). The bottom one is red, that is, the negative class, but the red gradually becomes darker, which means that those are classified as negative classes, and this probability is getting bigger and bigger, that is, the degree of "negative" is getting stronger and stronger, until The bottom layer has n(t)=1.0, pure red. the

Step 4) Label the emotional labels on the established emotional visualization graph. the

The emotional words that appear in social media event information can be displayed on the emotional visualization graph as emotional labels, and the frequency of occurrence of emotional words and the emotional intensity of emotional words can be reflected by the font size of the emotional labels. In this visual graph , the font size F(w) of the emotional label w is calculated as follows:

F(w)=α*PMI-IR(w)*f(w)

Where f(w) is the frequency of the emotional word w appearing in the information of each emotional classification, PMI-IR(w) represents the point-by-point mutual information value of the emotional word w based on information retrieval, α is a scaling factor, and the size of α It has something to do with the scale of the visualization. For example, if the largest font that can be displayed is a font size 9, you can set α to 9. the

For example, it is assumed that in the result of processing the above-mentioned microblog data set in June 2008, the emotional label is "incredible, beautiful, wrong, idolt". Then calculate the font size of each emotional label text according to the font size formula (shown in Table 2), and finally generate the emotional visualization graph as shown in Figure 3. the

Table 2

It should be understood that the above step 4) is optional. Marking emotional labels on the established emotional visualization graph is to further display more emotional information on the visual image to help users better predict the development of events trend. All appearing emotional words can be displayed on the established visual graph, or only some emotional words with high frequency of occurrence and large PMI-IR value can be displayed. the

Figure 4 shows the analysis results of tweets about Obama from June 2008 to May 2009. Figure 4 shows the evolution of sentiment about Obama. The graph shows that the width of the flow increased in November 2008, which indicates that the number of tweets about Obama reached a peak. The changes in the graph can be attributed to Obama on November 5, 2008 Defeated John McCain, was elected as the 44th President of the United States, and declared elected. Besides that, in January 2009, the green layer overtook the red layer, showing that the majority of tweeters posted positive content, at the same time that Obama delivered his inaugural address, which can be considered uplifting Popular and so successful that many people support him. the

Compared with the traditional visualization methods such as pie chart and histogram shown in Figure 5, the pie chart in Figure 5 can clearly illustrate the proportion of each emotion, but it cannot explain the intensity of emotion and does not reflect Emotional ups and downs. A histogram centrally illustrates how data changes over different time series. The color gradient in the histogram can represent the intensity of emotion, but the problem it faces is the same as that based on the horizontal plane. Understanding such a histogram requires the user to associate emotional changes with time. In addition, from Histograms make it difficult to see the evolution of sentiment over time because the time windows are divided manually. Therefore, the information provided by the visualization methods based on pie charts and histograms is relatively limited, and can only provide a macroscopic perspective on event sentiment. The visualization diagram of the present invention shown in FIG. 4 visualizes the emotional intensity through the change of the color gradient, and the tweet density function based on the event sequence makes the turning point and burst point of the event and the prediction of the event more intuitive. And the emotional word labels also reflect the emotional intensity. On the whole, the visualization diagram provided by the present invention shown in Fig. 4 reflects that people's emotions to Obama change over time, especially when some influential events occur, and finally these changes tend to be gentle, at this time There are also fewer tweets with strong emotions. From this point of view, the emotion visualization diagram according to the embodiment of the present invention helps users better understand the intensity of emotions at different times and the trend of evolution over time. the

In yet another embodiment of the present invention, there is also provided a visualization device for the dynamic viewpoint evolution of a social media event, including: a device for determining the emotional membership and emotional classification of information in the collected social media event information collection , the emotional membership degree of the information indicates how likely the information belongs to a certain emotional classification; the device for establishing the geometric layout of the emotional visualization graphics of the information set based on the emotional classification of the information, in the geometric layout , the horizontal axis represents the time of information generation, and the vertical axis represents the quantity of information belonging to each emotion classification; a means for coloring the established geometric layout based on the emotional membership of the information. The apparatus may also include means for labeling emotion labels on the created emotion visualization graph. the

Although the present invention has been described in terms of preferred embodiments, the present invention is not limited to the embodiments described herein, and various changes and changes are included without departing from the scope of the present invention. the

Claims (10)

1. A visualization method of the dynamic viewpoint evolution of a social media event, said method comprising: Step 1) Determine the emotional membership and emotional classification of the information in the collected social media event information collection, the emotional membership of the information indicates how likely the information belongs to a certain emotional classification; Step 2) Based on the emotional classification of the information, establish the geometric layout of the emotional visualization graph of the information set. In the geometric layout, the horizontal axis represents the time when the information was generated, and the vertical axis represents the information belonging to each emotional classification. quantity; Step 3) Color the established geometric layout based on the sentiment membership of said information. 2. The method according to claim 1, said step 1) comprising: Step 11) selecting a small part of information from the social media event information collection for artificial emotion labeling; Step 12) Train a classifier for sentiment classification based on the labeled information; Step 13) Use the trained classifier to classify the remaining information in the social media event information set that has not been artificially labeled, and determine the emotional membership and emotional classification of each piece of information. 3. The method according to claim 2, in the step 12), the classifier is trained with the text view features and non-text view features of the information, wherein the text view features of the information are based on the information of the emotional words in the information The retrieved point-wise mutual information values are represented by vectors, and the non-text view features of information include emoji features, time features, and punctuation features. 4. method according to claim 1, in described step 2) in described geometric layout, the curve function of the bottom layer is S0, and the curve function of the i-th layer emotion classification is: ${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; \&rho;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}$ Among them, the density function ρ _j is the amount of information belonging to the jth emotion category within a unit time period, 1≤i≤n; S ₀ =0 or n is the number of sentiment categories. 5. The method according to any one of the preceding claims, said step 3) comprising: Step 31) Establishing a mapping between the emotional membership of the information and the color of the information; Step 32) On the established geometric layout, color the information on each emotion classification layer according to the gradient order of the information color. 6. The method of claim 5, the sentiment classification being positive, neutral or negative. 7. The method according to claim 6, said step 31) adopts the following function to establish a mapping between the emotional membership of the information and the color of the information: $\mathrm{<span\; class="notranslate">\; RGB</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; 255,255,0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\\ <span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 255</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; 255,0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\end{array}\right.$ Wherein, t represents any information in the social media event information set, n(t) represents the membership degree of the information t to the negative emotion classification, and p(t) represents the membership degree of the message t to the positive emotion classification. 8. The method according to claim 5, further comprising step 4) labeling emotional tags on the established emotional visualization graph, the emotional tags being emotional words appearing in the social media event information. 9. the method according to claim 8, described step 4) in, described emotion label in described step 3) the font size on the visualization graph that obtains is calculated with following formula: F(w)=α*PMI-IR(w)*f(w) Wherein, w represents any emotional word to be displayed on the visual graph, f(w) is the frequency of occurrence of emotional word w in the information of each emotional classification, and PMI-IR (w) represents the emotional word w Pointwise mutual information value based on information retrieval, α is a scaling factor. 10. A visualization device for the evolution of dynamic viewpoints of social media events, said device comprising: A device for determining the emotional membership and emotional classification of information in the collected social media event information collection, the emotional membership of the information indicates how likely the information belongs to a certain emotional classification; A device for establishing the geometric layout of the emotion visualization graph of the information set based on the emotion classification of the information, in the geometric layout, the time when the information is generated is represented by the horizontal axis, and the information belonging to each emotion classification is represented by the vertical axis quantity; Means for coloring the established geometric layout based on the sentiment membership of said information.

Images