CN103455705A - Analysis and prediction system for cooperative correlative tracking and global situation of network social events - Google Patents

Abstract

The present invention is a collaborative correlation tracking and global situation analysis and prediction system of network social events. The system has: an information fusion unit, which fuses the multi-modal characteristics of network social event data, and is used to obtain the multi-modal characteristics of social events. Data fusion information builds a semantic description model of multi-modal data across social events; the tracking unit is connected with the information fusion unit, based on the multi-modal data semantic description model of cross-social events, facing network content containing rich multimedia information, Aiming at the cross-modal attributes, cross-platform attributes and cross-time-space attributes of network social events, the semantic correlation tracking data of social events on each attribute is obtained; the analysis and prediction unit is connected with the tracking unit, and based on the semantic correlation tracking data, the social-based Global situational analysis and forecast data of events.

Description

Collaborative Correlation Tracking and Global Situation Analysis and Prediction System of Network Social Events

technical field

The invention belongs to the technical field of association tracking and prediction of network social events, and relates to a collaborative association tracking and overall situation analysis and prediction framework system of network social events.

Background technique

With the rapid popularization of the Internet, the Internet has become an indispensable tool in most people's lives. The number of Internet users is growing rapidly and has gradually become a huge group. According to the statistical report on the development of the Internet in China, as of the end of December 2012, the number of Chinese web pages was 122.7 billion, an increase of 41.7% over the same period in 2011. The above statistics show that we already live in a massive amount of network data. We not only need to store these network data reasonably, but also need to study how to deal with the massive data in the network, so as to obtain the potentially useful information hidden behind the massive network data. In the face of massive text, image, and video data on the Internet, how to effectively organize, acquire, mine, and monitor large-scale data generated by social hot events has become an urgent application requirement.

A hot news event is constantly evolving and changing over time. With the development of the event, new and important news stories occur and old stories disappear. When users search for this news event, an important task is to discover news topics, find out the correlation between each news story, and construct a process map of event development, and establish the development context of this event theme. Therefore, when searching for a hot event, the search results that users want to see are not a series of news items arranged only according to textual relevance, allowing users to find the relationship and relationship between the various stories of the event in a pile of messy news items. The progress of the event should be based on the development of the event, and the information such as the relationship between the event evolution, the development of the event, and the evolution process of the event theme can be displayed to the user in a clear and graphical interface, so that the user can quickly know the entire The development and evolution process of the event, and allows users to quickly understand and analyze the theme of the event. How to quickly, accurately, and appropriately provide the information users need, and to a certain extent, reveal the correlation between information, present the complete evolution process of a hot event to users, provide users with intelligent information retrieval services, and allow users to quickly Understanding and analyzing the subject of this event has become a topic of common concern in both academia and industry.

Due to the characteristics of the Internet, such as virtuality, concealment, divergence, permeability and randomness, when a hot event occurs, more and more netizens are willing to express their views through network channels. Today, the dissemination of information and the interaction of opinions are unprecedentedly fast, and the demands of expressing online public opinion are increasingly diverse. If it cannot be effectively guided, negative Internet public opinion will pose a greater threat to social public security. Looking at the international environment and my country's national conditions, network security, social stability and the supervision of hot events are all facing major challenges. The rapid spread of the Jasmine Revolution in the Middle East in various countries and the interaction of network information, as well as the recent "Jiangsu Qidong 728 Incident" and "Sino-Japanese Diaoyu Islands Dispute" in my country, have attracted hundreds of millions of netizens on the Internet. The rapid dissemination and extensive participation of netizens all show that we are facing an extremely severe situation in the analysis of social events. my country is in a critical period of building a well-off and harmonious society in an all-round way, and a safe and stable political and social environment is especially needed. Therefore, the government and relevant managers urgently need to effectively mine, analyze and process public opinion information on the Internet, and grasp and handle crisis events best time. However, it is difficult to deal with the collection and processing of massive network information only by manual methods. It is necessary to integrate various information technologies, and analyze current social hotspot events and emergencies through appropriate computer technology, and to analyze negative public opinion and misleading information on the Internet. Respond to speech in a timely and rapid manner, dig out the causes of various events in the network, and stifle rumors and crises invisible, thereby improving the ability to supervise and deal with hot events and emergencies.

Existing tracking and forecasting algorithms for social events only use textual information from a single platform. In terms of understanding text information, clustering technology is the basic technology of data mining and pattern recognition. The themes of social events are obtained by clustering text information. Traditional clustering methods are not suitable for the explosive and regional characteristics of social events. , the accuracy of topic mining is not high, and it is difficult to describe a complete event semantically. Because in addition to text information, events also have their rich visual information. For an event, there will be different user comments on different websites, however, the two websites may have very similar visual information, such as images or videos, which can be used to construct an event item evolution across time and across websites. Bridges are very useful. For example, for the event "2012 U.S. Presidential Election", the items of this event are highly correlated with images of Obama. Therefore, using multimodal information fusion can describe social events more correctly. Moreover, different platforms can also complement and enhance each other. For example, most events on Google News are from official media, but they also have many user comments on Flickr. Therefore, information on different platforms can help each other, especially when the advantages of one platform complement the weaknesses of another platform is more effective. The key challenge is how to find the most effective way to bridge the semantic gap between the two platforms. In response to this situation, if visual information can be integrated, a unified multi-modal information description can be established for social events, and social events can be effectively collaboratively correlated on different platforms, so as to realize cross-platform multi-modal information fusion. Event semantic description system. The collaborative tracking of social events based on cross-platform multimedia information and the global situation analysis and prediction method can effectively combine multi-platform and multi-modal information to improve this deficiency.

The existing research on tracking and forecasting methods for network social events is still in the initial stage, and there is no complete process and framework system so far. The present invention is a framework system for collaborative correlation tracking and prediction based on network social events, which effectively makes up for the shortcomings of traditional methods, and realizes cross-modal, cross-platform and cross-time-space collaborative correlation tracking based on multi-modal information semantic fusion and global situation analysis and prediction.

Contents of the invention

(1) Technical problems to be solved

The purpose of the present invention is to provide a complete framework system based on collaborative tracking and forecasting of network social events. According to the multi-modal characteristics of network social event data, we can effectively track hot social events in different modes, platforms, and time and space; after tracking multiple events, we can track the entire event process over time. Visual display; based on the tracking results, a complete description of a social event can also be obtained. Through the association analysis of these information and the topic mining model, the theme and public opinion orientation of the social event can be mined, and the statistical learning method can be used to learn about the social events. The overall situation of the event can realize the analysis and prediction of the global situation based on social events. Therefore, a framework for collaborative correlation tracking and global situation analysis and prediction of network social events is proposed.

(2) Technical solutions

In order to achieve the above object, the present invention provides a system for collaborative correlation tracking and overall situation analysis and forecasting of network social events, which includes:

The information fusion unit integrates the multi-modal characteristics of network social event data, uses natural language understanding and image and video processing and analysis technology to obtain fusion information of multi-modal data of social events, and constructs a multi-modal cross-social event Semantic description model of state data;

The tracking unit is connected with the information fusion unit, based on the multi-modal data semantic description model of cross-social events, facing the network content containing rich multimedia information, aiming at the cross-modal attributes, cross-platform attributes and cross-time-space attributes of network social events, Combined with collaborative correlation tracking technology, the semantic correlation tracking data of social events on each attribute is obtained;

The analysis and prediction unit is connected with the tracking unit, and based on the semantic association tracking data, the whole process information of a social event over time is obtained, and the process information is mined through the social event theme mining model, social event pattern prediction and social event statistical prediction model. The theme and public opinion guidance of the social event of the process information, so as to know the overall situation of the social event, and obtain the global situation analysis and forecast data based on the social event.

Beneficial effects of the present invention: the present invention adopts the multi-modal information fusion and modeling of social events, uses the collaborative correlation tracking of social events based on three different attributes of cross-platform, cross-modal and cross-time and space, and excavates each The theme and public opinion orientation of social events, and use statistical learning methods to predict their subsequent development trajectory, and finally put forward a framework system for collaborative correlation tracking and overall situation analysis and prediction of social events on the Internet, including social media information about multimedia information. Event collaborative correlation tracking and global situation analysis and prediction methods. This invention solves the problem of describing the complete evolution process of different hot social events on the network in different modalities, platforms, and time and space, as well as tracking and predicting them. Using the proposed synergistic probability multi-hypothesis association tracking algorithm can greatly improve the system's social awareness. The tracking accuracy of events in each time period is accurate, so as to provide a more accurate development and evolution process of the entire event, which is convenient for users to understand and analyze and predict public opinion.

Description of drawings

Fig. 1a and Fig. 1b are the complete framework system of the present invention oriented to the network social event cooperative association tracking and prediction system;

Fig. 2 is a schematic diagram of semantic description of text-based topic mining in the present invention;

Fig. 3 is a schematic diagram of semantic description of topic mining based on image and video in the present invention;

Fig. 4 is a schematic diagram of cross-modal semantic association in the present invention;

Fig. 5 is a schematic diagram of cross-platform semantic association in the present invention;

Fig. 6 is a schematic diagram of temporal-spatial semantic association in the present invention;

Fig. 7 is a schematic diagram of multimodal topic mining based on social events in the present invention;

Figure 8 is a schematic picture of the distribution of road congestion in various periods of a highway in recent years;

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be described in further detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

As shown in Figure 1a and Figure 1b, the present invention is oriented to the complete framework system of the network social event cooperative tracking and prediction system. A complete description from start to finish. In this way, the entire event process can be visualized and displayed over time, and displayed to users in a clear and graphical interface, so that users can quickly know the development and evolution process of the entire event, and can mine the semantic theme of each event. Knowing the public opinion orientation of social events, and using statistical learning methods to learn the overall situation of this social event, can realize the analysis and prediction of the overall situation based on social events. The structure of the present invention is shown in Figure 1a and Figure 1b, and it comprises three components. The system includes: an information fusion unit 1, a tracking unit 2 and an analysis and prediction unit. The technical solution for realizing the system is as follows:

1 Multimodal information fusion and modeling of social events

Use the information fusion unit 1 to fuse the multi-modal characteristics of network social event data, use natural language understanding and image and video processing and analysis technology to obtain fusion information of multi-modal data of social events, and construct cross-social event data Semantic description model for multimodal data;

Wherein, the semantic description model of constructing the multi-modal data of cross-social events is used to extract the features of the text and visual cross-media information of each social event, and perform semantic analysis on the text and visual cross-media information of each social event. Hierarchical mining, so as to realize the semantic description of social events, build a semantic knowledge base based on multi-modal information fusion, and obtain multi-modal semantic descriptions of various social events.

Wherein, the multimodal data fusion information of social events includes text analysis based on natural language understanding, and an event semantic description system based on image and video processing to realize multimodal information fusion, and realize multimodal semantic information fusion.

Among them, based on text analysis, and topic mining of images and videos, semantic descriptions are adopted, and effective text topics and image and video topics are selected by a lifting algorithm.

1.1 Text analysis based on natural language understanding

For text information, consider using the deformed topic model (LDA, Latent DirichletAllocation) to obtain the topic information of each event, and use the context information of the text to further realize the description of the event text and establish the semantics centered on entities, events and topics The description system is mainly discussed from the following aspects:

a) Entity recognition and association

Extract named mentions, pronominal mentions, and nominal mentions in the text, and connect the mentions referring to the same entity concept, and also use the method of combining statistical translation and text retrieval to establish the correspondence between Chinese and English entities, Realize the association and integration of multilingual information.

b) Event extraction and description

Considering that event category recognition and element extraction are closely related to verb-centered syntactic and semantic analysis, we make full use of syntactic and semantic features and surface features for event category recognition and event element extraction. The Deformed Topic Model (LDA) is used to describe social event texts.

1.2 Based on image and video processing technology

Through images and videos, we can easily and effectively understand various social events. We use sparse learning and dictionary learning by considering the structural information of images and videos, so as to establish a unified visual semantic description system based on the bag-of-words model.

1.3 Multimodal Semantic Information Fusion

Social event content contains multi-modal information such as text, image, video, and sound. For text processing, we consider using the deformed topic model (LDA) to obtain the description of each event. For image and video processing, consider its spatial structure relationship , using sparse learning and dictionary learning to construct a visual topic model, so as to obtain the visual topics of each event, and further realize the semantic description of events according to the visual topics.

As shown in Fig. 2, the text-based topic mining semantic description of the present invention, given all document collections D, we use the deformed topic model (LDA) to mine event topics Z ₁ , Z ₂ ...Z _i ...Z _k , and then Words _{W 1 ,} W _{2 .} . . _{W j .}

As shown in Figure 3, the topic mining semantic description based on image and video in the present invention, based on the topic mining semantic description of the image and video set i, given the pictures or video sets of all events, the topic category C of the pictures or videos of each event is mined _1. C ₂ ... C _i ... C _k , based on the topic categories C ₁ , C ₂ ... C _i ... C _k, each feature descriptor W ₁ , W ₂ ... W _j ... W _n of semantic association of pictures or videos is established.

Based on the topic mining semantic description of text, image and video, the lifting algorithm is adopted to select effective text topics and image and video topics, so as to realize the event semantic description system of multi-modal information fusion, and realize the social event description of multi-modal information fusion .

2 Collaborative Association Tracking Model of Social Events

Utilize the connection between the tracking unit 2 and the information fusion unit 1, based on the multi-modal data semantic description model of cross-social events, in the face of network content containing rich multimedia information, the cross-modal attributes, cross-platform attributes and cross-platform attributes of network social events Spatio-temporal attributes, combined with collaborative association tracking technology, to obtain semantic association tracking data of social events on each attribute;

Wherein, the semantic correlation tracking data of each attribute of the social event is used to gather semantically related network content in the Internet to form a set that can reflect a common theme of the social event.

Among them, aiming at the cross-platform attribute of network social events, a collaborative probability multi-hypothesis tracking method is proposed to realize the semantic collaborative correlation tracking of multiple social event data on two platforms, which is used to realize the cross-modal and cross-temporal association of network social events track.

The collaborative association tracking model of social events gathers semantically related network content in the Internet to form a collection that can reflect the common theme of social events. Facing the network content containing rich multimedia information, aiming at the cross-platform, cross-modal and cross-time and space attributes of network social events, combined with collaborative learning and association technologies, the collaborative correlation tracking of social events on various attributes is realized.

2.1 Cross-modal semantic association

For cross-modal semantic association, its purpose is to realize the association of social events between different modes, as shown in Figure 4 cross-modal semantic association in the present invention, for the two networks of the US presidential election and the launch of the Mars survey orbiter Social events make cross-modal semantic associations between multiple texts in the text domain and multiple images in the visual domain, and realize cross-modal semantic associations between related texts and images that occurred in the same period.

We intend to use multimedia feature fusion analysis at the feature layer to mine potential semantic correlations between different modalities and media spaces, map information from different media to a higher-dimensional public space, and select appropriate measures in the same space to measure The similarity of information samples is characterized by an implicit semantic association matrix of cross-media features. On this basis, a latent semantic index of modal features is constructed, which is used to project cross-media samples into a latent semantic space, and establish the mapping and association of cross-media content on different feature spaces.

2.2 Cross-platform Semantic Association

Here, the cross-platform association refers to the association relationship between multi-source platforms (such as People's Daily Online and Sina Weibo). The occurrence and development of social events and the various sub-events experienced often exist on multiple platforms at the same time on the Internet. For example, information about an event will be described on both platforms (People. Complement each other, so it is necessary to study the complex relationship across platforms, so as to discover and mine the semantic relationship between information samples across platforms. The cross-platform attribute of social events is usually triggered by a certain sub-event of the social event, and then a large number of online speeches are generated on various platforms in cyberspace. Therefore, the key to cross-platform semantic association is based on each sub-event through synergy and complementarity. Association to realize information correspondence on different platforms. Figure 5 shows the cross-platform semantic correlation in the present invention. This is the social event of "Greek Protest" and "Sino-Japanese Diaoyu Islands Dispute". The evolution process of each sub-event has a corresponding description on People's Daily Online and Sina Weibo. Using the synergy probabilistic multi-hypothesis association tracking model we proposed to establish associations between the data of these two platforms, the information of these two social events on the People's Daily Online and Sina Weibo platforms at each moment can be correlated and supplemented. In this way, we can know more clearly the specific information of this event at this moment, can accurately analyze each sub-event, and better monitor and predict social events.

2.3 Semantic association across time and space

The cross-time-space here refers to the difference in geographical location and time. Because after a social event occurs, people's views on this event are different in different spaces, and will change with time. Therefore, we need to establish a cross-temporal semantic association to describe the different information presented by an event in different regions, so that decision makers can grasp the interconnections and differences between regions, and better guide the development of public opinion on events. , to serve the country. We take the words in this event text as the sample space, and for each information sample, we know the "word-time-place" triplet, we will first establish the similarity measure of time scale and geospatial scale, and calculate the Then, for a word in the sample space "dictionary", the following four frequency values are counted: the frequency that appears in a certain time interval and a certain area at the same time, and the frequency that does not appear in a certain time interval But the frequency of appearing in a certain area, the frequency of appearing in a certain time interval but not in a certain area, and the frequency of not appearing in a certain time interval nor in a certain area, these four quantities, through the word based on space and time Frequency analysis obtains the popularity coefficient of each word in a certain space and time, and integrates the correlation between geographical location and time according to this popularity coefficient, and then obtains the semantic association across time and space. As shown in Figure 6, the temporal-spatial semantic correlation in the present invention, this is the spatio-temporal evolution process of the spread of the event "H7N9", its spread and outbreak are regional and temporal, Shanghai and Anhui are the latest discovery, and then began to spread to surrounding provinces At the beginning, there was a small expansion of time and space in the city, but there will be a sudden moment in the middle, such as on April 3, when H7N9 infections were suddenly found in Beijing, and the spread of H7N9 also showed time and space characteristics. A cross-temporal semantic association, which describes the different information presented by an event in different regions at different times, so that decision makers can grasp the interrelationships and differences between regions, and better have a rational analysis of events and judgment.

3 Analysis and prediction of the overall situation of social events

Using the analysis and prediction unit 3 to connect with the tracking unit 2, based on the semantic association tracking data, obtain the whole process information of a social event over time, through the mining process of the social event theme mining model, social event pattern prediction and social event statistical prediction model Information, dig out the theme and public opinion orientation of the social event in the process information, so as to know the overall situation of the social event, and obtain the global situation analysis and forecast data based on the social event.

Among them, the overall situation analysis and forecast data of the social events, obtain the summary details and event evolution trends of multiple events in different time periods, obtain a complete description of each social event from the beginning to the end, and follow the entire event process Visual display over time can show the relationship between event evolution, event development, and event theme evolution process information to users in a clear, graphical interface according to the development of events; through these complete social event information Correlation analysis to dig out the theme and public opinion orientation of this social event, and use statistical learning methods to predict its subsequent development trajectory, reveal the characteristics of communication behavior, and know the overall situation of this social event.

Wherein, the global situation analysis and prediction data based on social events is based on all the information of each event obtained from the tracking data, adopts the method of topic mining model and statistical learning, predicts the theme and trend of each social event, and is used for accurate To analyze and predict the overall situation of social events.

The dissemination of a social event is a typical dynamic process, a process that evolves over time and with the direction of public opinion, and no matter from the time dimension or the space dimension, any social event behavior is not simply isolated, each behavior There is an intricate relationship among them, and the occurrence of every social event also has obvious or hidden signs. There will be a large amount of multi-modal information such as text, images, and videos on social events on the Internet. Using the cross-modal information fusion and modeling of social events proposed by us, the evolution process of a social event can be expressed in detail , through the association analysis of these information, we can dig out the theme and public opinion orientation of social events, and use the method of statistical learning to know the overall situation of this social event, we creatively proposed three prediction models for different social events.

3.1 Mining themes of social events

There will be a large amount of multi-modal information such as text, images, and videos on social events on the Internet. Through cross-modal and cross-platform information fusion and modeling of social events, the evolution process of a social event can be expressed in detail. Then carry out topic modeling for multi-modal information, so that the entire event process can be visualized and displayed over time. On the one hand, information such as the relationship between event evolution, event development, and event theme evolution process can be analyzed according to the development of the event. , displayed to users with a clear, graphical interface, allowing users to quickly know the development and evolution of the entire event, and allowing users to quickly understand and analyze the theme of the event, on the other hand, it can better analyze social events The overall situation and public opinion information to carry out corresponding public opinion guidance and public opinion monitoring.

After tracking various social events, it is possible to carry out corresponding text and visual theme mining, and finally get the topic information of each evolution process of social events, which is of great help to the tracking and defense of social events and the monitoring of public opinion. Figure 7 shows the multi-modal topic mining based on social events in the present invention. By tracking the event ("2012 U.S. Presidential Election") across modes, platforms, and time and space, the results of the event at each moment are obtained. In the evolution process, the theme mining is carried out on the text and images of the whole process of the event through the theme model, and the text theme and visual theme of the event at each moment in Figure 7 are obtained, and then these text themes and visual themes at each moment are associated. and statistics, to realize the mining of event themes, and finally visualized display, which is convenient for users to understand and analyze public opinion.

3.2 Statistical prediction of social events

The evolution process of social events can be described in detail through our proposed method, and the topic distribution of this social event can be obtained, so we can analyze and predict this social event accordingly. We plan to use statistical learning methods to predict the trend of various social events based on all the information of each event obtained from the tracking results, so that we can accurately analyze and monitor social events, and government decision makers can take timely corresponding measures. countermeasures to nip negative public opinion in the bud. In the entire forecast cycle, a baseline will be established through basic data, and then as time goes by, the information of the event obtained according to the tracking results will become more and more abundant, so that the future trend of the event will become more and more obvious, with the help of statistical data The method of mining realizes the prediction of events.

3.3 Social event pattern prediction

Many social events have a common spatio-temporal pattern, and they will appear in a time period of the year. We can analyze the same social events in detail based on the occurrence of social events in recent years, using our proposed modeling method , to get the subject distribution of the social event, we can analyze and predict this social event accordingly, so as to know when this similar social event is most likely to occur in the future, and preventive measures can be taken in advance.

Figure 8 shows the distribution of road congestion in various periods of a certain highway in recent years. The road congestion varies with time in a year, but the development trajectory of this social event in recent years is similar. We can model these common events and use pattern mining methods such as graph matching to analyze and predict the development trajectory of this event in each period, so as to provide reliable information for decision makers. Take appropriate preventive and remedial measures.

4. The collaborative association tracking and overall situation analysis and prediction method of network social events of the present invention comprises the following steps:

Step S1: Extract the features of the text and visual information of each social event, and carry out semantic level mining on these cross-media information, so as to realize the semantic description of social events, and build a semantic knowledge base based on multi-modal information fusion of social events , so as to obtain the multimodal semantic description model of each social event.

Step S2: Gather semantically related network content in the Internet to form a collection that can reflect the common theme of social events. Facing the network content containing rich multimedia information, aiming at the cross-platform, cross-modal and cross-time-space attributes of network social events, combined with collaborative learning and association technologies, a collaborative correlation tracking model of social events on various attributes is realized.

Step S3: Through step 2, the summary details and event evolution trends of multiple events in different time periods can be obtained, so that a complete description of each social event from the beginning to the end can be obtained. In this way, the entire event process can be visualized and displayed over time, and information such as the relationship between event evolution, event development, and event theme evolution process can be displayed to users in a clear and graphical interface according to the development of the event. . Through the association analysis of these complete social event information, the theme and public opinion orientation of this social event can be excavated, and the method of statistical learning can be used to predict its subsequent development trajectory, reveal the characteristics of communication behavior, and know the social event. global situation.

5 Implementation effect

In order to evaluate the present invention, we conducted some experimental research, selected 18 popular topics from Google News (Google News) and Flickr as the hot social events of research, the data set is shown in Table 1, which covers politics, economy, technology , entertainment, military, social and other themes. There are 3583 documents and 6742 images on Google News and 4356 documents and 4356 images on Flickr.

Table 1 Google English news and Flickr picture information dataset of the present invention's implementation effect

Table 2 is the comparison of the present invention at the cross-platform and single-platform tracking results on the two platforms of Google English News and Flickr picture information. It can be seen from Table 2 that the method of the present invention has a significant effect on the correlation tracking of various social events across platforms. Compared with single information such as text or vision, using multi-modal information fusion technology can obtain more Good description of social events. After joining multiple platforms and using cross-platform collaborative correlation tracking, better tracking accuracy can be obtained compared with the tracking method of a single platform.

Table 2 Comparison of cross-platform and single-platform tracking results in the Google English news data set of the present invention

Table 3 is the topic mining information of various social events in Google English News and Flickr image information data sets according to the present invention. As can be seen from Table 3, the method of the present invention has a good effect on topic mining information on various social events across platforms. Event 1 is "the dispute over the Diaoyu Islands between China and Japan", and event 2 is "the legalization process of marijuana in the United States". Event 3 is "2011 Norwegian Attack", Event 4 is "2008 China Milk Scandal", and Event 5 is "2012 US Presidential Election Process". The keywords excavated from Event 1 include China, Japan, Diaoyu Islands, Taiwan, surveillance, military, etc., which clearly reflect the subject information of Event 1 and prove the reliability of the system again.

Table 3 The present invention excavates the topics of social events in the Google English news data set

event 1 event 2 event 3 event 4 event 5 China marijuana breivik milk Obama Japan cannabis police melamine Romney islands medical Norway food Republican Senkaku the law Norwegian Chinese president Taiwan possession Oslo scandal election watch decriminalize july children state military US attacks Sanlu US

The above experiments have fully confirmed the effectiveness and completeness of the invention.

The above is only a specific implementation mode in the present invention, but the scope of protection of the present invention is not limited thereto. Anyone familiar with the technology can understand the conceivable transformation or replacement within the technical scope disclosed in the present invention. All should be covered within the scope of the present invention, therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims (8)

1. A collaborative association tracking and global situation analysis and prediction system of network social events, characterized in that the system includes: The information fusion unit integrates the multi-modal characteristics of network social event data, uses natural language understanding and image and video processing and analysis technology to obtain fusion information of multi-modal data of social events, and constructs a multi-modal cross-social event Semantic description model of state data; The tracking unit is connected with the information fusion unit, based on the multi-modal data semantic description model of cross-social events, facing the network content containing rich multimedia information, aiming at the cross-modal attributes, cross-platform attributes and cross-time-space attributes of network social events, Combined with collaborative correlation tracking technology, the semantic correlation tracking data of social events on each attribute is obtained; The analysis and prediction unit is connected with the tracking unit, and based on the semantic association tracking data, the whole process information of a social event over time is obtained, and the process information is mined through the social event theme mining model, social event pattern prediction and social event statistical prediction model. The theme and public opinion guidance of the social event of the process information, so as to know the overall situation of the social event, and obtain the global situation analysis and forecast data based on the social event. 2. as claimed in claim 1, the collaborative correlation tracking and global situation analysis and prediction system of network social events, is characterized in that, the semantic description model of the multi-modal data of described construction cross social event, is used for extracting each social event The characteristics of text and visual cross-media information, and carry out semantic level mining on the text and visual cross-media information of various social events, so as to realize the semantic description of social events, build a semantic knowledge base based on multi-modal information fusion, and thus A multimodal semantic description of each social event is obtained. 3. as claimed in claim 1, the collaborative association tracking and global situation analysis and prediction system of network social events, is characterized in that, the semantic association tracking data of said social events on each attribute is used to integrate semantically related networks in the Internet. The content is aggregated to form a collection that can reflect the common theme of social events. 4. as claimed in claim 1, the collaborative correlation tracking and global situation analysis and prediction system of network social events, is characterized in that, the global situation analysis and forecast data of described social events, obtains the abstract details of multiple events in different time periods Information and event evolution trend, get a complete description of each social event from the beginning to the end, and visualize the entire event process over time, and can develop the relationship between event evolution, event development, and event theme according to the development of the event. The evolution process information is displayed to users in a clear and graphical interface; through the association analysis of these complete social event information, the theme and public opinion orientation of this social event are excavated, and the statistical learning method is used to predict The trajectory of its subsequent development reveals the characteristics of the communication behavior and the overall situation of this social event. 5. The collaborative association tracking and global situation analysis and prediction system of network social events as claimed in claim 1, is characterized in that, the multi-modal data fusion information of described social events includes text analysis based on natural language understanding, and image-based and video processing, realize the event semantic description system of multi-modal information fusion, and realize the multi-modal semantic information fusion. 6. as claimed in claim 5, the collaborative association tracking and global situation analysis and prediction system of network social events is characterized in that, based on text analysis and topic mining semantic descriptions of images and videos, a lifting algorithm is adopted to select effective text topics and Image and video themes. 7. as claimed in claim 1, the collaborative association tracking and global situation analysis and prediction system of social network events, is characterized in that, for the cross-platform attributes of social network events, a collaborative probability multi-hypothesis tracking method is proposed to realize multiple assumptions on two platforms. Semantic collaborative association tracking of social event data, which is used to realize cross-modal and cross-temporal association tracking of network social events. 8. as claimed in claim 1, the collaborative association tracking and global situation analysis and prediction system of network social events, is characterized in that, said global situation analysis based on social events and forecast data are all information of each event obtained according to the tracking data. Information, using topic mining models and statistical learning methods, to predict the topics and trends of various social events, and to accurately analyze and predict the overall situation of social events.

Images