TWI744620B - Pooling prediction system - Google Patents

Abstract

An election prediction system is provided. The election prediction system comprises a winning factor analysis module, an online word-of-mouth estimation module, a media pool data accessing module, a public tendency data accessing module, a weight calculation module, and a win probability estimation module. The winning factor analysis module is utilized for generating a candidate strength data. The online word-of-mouth estimation module is utilized for generating an online word-of-mouth estimation data. The media pool data accessing module is utilized for generating a media pool data. The pubic tendency data accessing module is utilized for generating a public tendency data. The weight calculation module is utilized for generating a weight data based on a voter structure data. The win probability estimation module is utilized for generating a candidate predictive win probability.

Description

Election Forecast System

The present invention relates to a prediction system, especially an election prediction system.

With the rapid development of the Internet, Internet public opinion has become a data source that brands/service companies and even policy research institutions attach great importance to. Traditional election prediction methods that rely solely on sample surveys by polling agencies have failed to effectively reach different ethnic groups, and often fail to effectively reflect actual voter preferences, and are prone to inaccurate predictions.

In view of this, the present invention provides an election prediction system to improve the accuracy of prediction.

An embodiment of the present invention provides an election prediction system. This election prediction system includes a candidate winning factor analysis module, an online word-of-mouth prediction module, a media poll data acquisition module, a crowd prediction data acquisition module, a weight calculation module, and an election Rate prediction module.

Among them, the candidate winning factor analysis module is used to retrieve at least one historical election data to analyze the corresponding at least one winning factor The weight is used to generate a candidate's strength data. The internet word-of-mouth prediction module is used to retrieve at least one website data of one website, and analyze the website data to generate an internet word-of-mouth prediction data. The aforementioned online word-of-mouth prediction data includes at least one of voter familiarity data, voter favorability data, and online business performance data. The media poll data acquisition module is used to acquire at least one media poll report to generate a media poll data. The crowd prediction data acquisition module is used to retrieve at least one network statistical data to generate a crowd prediction data.

The weight calculation module is used to generate weight data corresponding to at least one of the aforementioned candidate strength data, Internet word-of-mouth prediction data, media poll data, and crowd prediction data based on a voter structure data. The election rate prediction module is used to adjust the weight of candidate strength data, Internet word-of-mouth prediction data, media poll data, and mass prediction data based on the weight data to generate a candidate's predicted election rate.

Compared with traditional election prediction methods that rely solely on sample surveys by polling agencies, the election prediction system provided by the present invention takes into account both media polling data and Internet crowd prediction data, and uses machine learning to obtain candidate strength data. Get online word-of-mouth data by means of sentiment analysis, and assign appropriate weights to each data. In this way, on the one hand, the electoral group that cannot be touched by traditional election prediction methods can be touched, and on the other hand, other possible influencing factors can be taken into consideration. In this way, the accuracy of election prediction can be improved.

The specific embodiments adopted in the present invention will be further explained by the following embodiments and drawings.

10: Election prediction system

100: Candidate Winning Factor Analysis Module

200: Internet word-of-mouth prediction module

300: Media Poll Data Acquisition Module

400: Crowd prediction data acquisition module

500: Weight calculation module

600: Winning rate prediction module

20: Historical election database

220: Web crawler unit

240: Semantic Analysis Unit

260: Sentiment Analysis Unit

30a, 30b: website

262: Vector Conversion Module

264: Emotion Prediction Module

266: Predicted value generation module

40a, 40b: media

520: weight generation unit

540: Weight adjustment unit

D _H : Historical election information

D1: Candidate's strength information

DOC: website information

D2: Internet word-of-mouth prediction data

V _DOC : text vector

C _P : Positive reliability data

C _N : Negative reliability data

V _S : Mood value

D3: Media polling data

P _M1 , P _M2 : media poll report

P _T : Media trend data

D4: Crowd forecast data

D _NS : Voter structure data

D5: Weight data

D _HNS : historical voter structure data

D _W : Initial estimation weight data

D6: Candidates predict the election rate

D _A : Relative advantage numerical data

P _TL : Release time information

640: Election Rate Calculation Unit

P _N1 ,P _N2 : network statistics

250: Network sound volume statistics unit

620: Advantage Numerical Calculation Unit

50: Prediction Market Website

The first figure is a block diagram of an embodiment of the election prediction system of the present invention.

The second figure is a block diagram of an embodiment of the candidate winning factor analysis module of the first figure.

The third figure is a block diagram of an embodiment of the Internet word-of-mouth prediction module of the first figure.

The fourth diagram is a block diagram of an embodiment of the network sentiment analysis unit in the third diagram.

The fifth figure is a block diagram of an embodiment of the media poll data acquisition module of the first figure.

The sixth figure is a block diagram of an embodiment of the crowd prediction analysis unit in the first figure.

The seventh diagram is a block diagram of an embodiment of the weight calculation module of the first diagram.

The eighth figure is a block diagram of an embodiment of the election rate prediction module of the first figure.

Figure 9 is a flowchart of an embodiment of the election prediction method of the present invention.

The specific embodiments of the present invention will be described in more detail below in conjunction with the schematic diagrams. According to the following description and the scope of patent application, the advantages and features of the present invention will be more clear. It should be noted that the diagrams are in a very simplified form and all use imprecise proportions. They are only used for Conveniently and clearly assist in explaining the purpose of the embodiments of the present invention.

The first figure is a block diagram of an embodiment of the election prediction system 10 of the present invention. As shown in the figure, the election prediction system 10 includes a candidate winning factor analysis module 100, an Internet word-of-mouth prediction module 200, a media poll data acquisition module 300, and a crowd prediction data acquisition module. Group 400, a weight calculation module 500, and an election rate prediction module 600.

Among them, the candidate winning factor analysis module 100 is used to retrieve at least one historical election data D _H to analyze the weight of the winning factors of each constituency to generate a candidate strength data D1. In one embodiment, the aforementioned historical election information D _H includes one candidate data, one political party recommendation data, one election experience data, and one government affairs experience data.

In one embodiment, as shown in the second figure, the candidate winning factor analysis module 100 is connected to a historical election database 20 to retrieve historical election data D _{H of} each constituency. In one embodiment, the historical election database 20 is the election database of the Central Election Commission, but it is not limited to this. The historical election database 20 may also be other public election databases, such as Wikipedia. In addition, the candidate winning factor analysis module 100 can also link multiple historical election databases 20 to retrieve more complete historical election data.

The candidate winning factor analysis module 100 analyzes the aforementioned historical election data D _H by means of regression analysis and/or machine learning to determine the weight of the winning factors of each constituency, and then generates candidate strength data D1. In one embodiment, the candidate winning factor analysis module 100 can use regression analysis to determine various historical election data D _H (such as candidate age, recommended political party, election experience, and government experience) and the candidate's victory. The degree of relevance of the election is used to confirm the weight of the winning factors of each constituency and apply it to the candidate to be predicted to generate candidate strength data D1. In addition, in one embodiment, the candidate winning factor analysis module 100 can use historical election data D _H as training materials, and analyze various historical election data D _H and the candidate's winning by means of machine learning. The correlation function or degree of correlation is applied to the candidate to be predicted to generate candidate strength data D1.

The Internet word-of-mouth prediction module 200 is used to retrieve one of the website data DOC of at least one website, and analyze the website data DOC to generate an Internet word-of-mouth prediction data D2. The aforementioned online word-of-mouth prediction data D2 includes at least one of voter familiarity data, voter favorability data, and online business performance data.

In terms of voter familiarity data, in one embodiment, as shown in the third figure, the Internet word-of-mouth prediction module 200 includes a Internet crawler unit 220, a semantic analysis unit 240, and an Internet voice volume statistics unit 250. With a sentiment analysis unit 260. The web crawler unit 220 is used to retrieve website data DOC of at least one website (in the figure, two websites 30a and 30b are taken as examples), such as text data of a discussion area on the Internet. The semantic analysis unit 240 performs keyword analysis on the retrieved website data DOC to generate Internet word-of-mouth prediction data D2, especially the voter familiarity data therein.

In one embodiment, the voter familiarity data includes internet voice volume, search popularity, number of community fans, number of community interactions, and number of community discussions. Among them, the network sound volume can be obtained through the network sound volume statistics unit 250. The more the candidate’s name appears in the discussion article on the website, that is It means that there is a higher volume of online voice, and it usually means that voters are more familiar. Search heat is how often voters use search engines (such as Google) to search for specific candidates. The higher the search popularity, the higher the voters' familiarity. Information about search popularity can be obtained by using existing statistical software, such as Google trend. The number of community fans, the number of community interactions, and the number of community discussions all represent the influence of the fan group of a particular candidate. The higher the influence of the fan group, the higher the familiarity of voters. The data on the number of community fans, the number of community interactions, and the number of community discussions can be obtained from social websites such as Facebook and Twitter through the web crawler unit 220 and existing statistical software.

In terms of voter favorability, the sentiment analysis unit 260 analyzes the retrieved network data by way of sentiment analysis to generate Internet word-of-mouth prediction data D2, especially the voter favorability data therein. In one embodiment, as shown in FIG. 4, the sentiment analysis unit 260 has a vector conversion module 262, an sentiment prediction module 264, and a prediction value generation module 266. The vector conversion module 262 is used to perform vector conversion on the retrieved website data DOC (especially text data) to generate a text vector V _DOC . The emotion prediction module 264 has a positive emotion prediction model and a negative emotion prediction model. The text vector V _DOC is a combination of a positive emotion prediction model and a negative emotion prediction model to generate a positive reliability data C _P and a negative reliability data _CN . The predictive value generating module 266 generates a sentiment value V _S (the text can be divided into positive, negative, and neutral) based on the positive reliability data C _P and the negative reliability data C _N. This sentiment value V _S can be used as the Internet word-of-mouth prediction data D2, especially the voter favorability data therein.

In addition to the aforementioned method of text sentiment analysis, Internet word-of-mouth The prediction module 200 can also use the data of the internet positive and negative ratings, the likes of social networking sites and emoticons to generate voter favorability data.

Regarding the online business performance data, in one embodiment, the online business performance data includes a community activity index, a fan adhesion index, and a post interaction index. The community activity index represents the degree of enthusiasm in the discussion of specific candidates by netizens. The Fan Adhesion Index represents the active degree of fans interacting with a particular candidate. The Post Interaction Index represents the enthusiastic state of interaction between fans and a specific candidate.

The aforementioned data can be obtained through the web crawler unit 220 and the semantic analysis unit 240. For example, the web crawler unit 220 can retrieve text data of a specific discussion area, such as main text and palindrome. The format of the main text may include article title, article content, article author, article link, publication time, number of likes, and response. Number of essays, number of shares, etc.: The format of the palindrome can include the content of the palindrome, the author of the palindrome, the link to the main text being replied, and the time of reply, etc. The semantic analysis unit 240 can analyze the text data captured by the web crawler unit 220 to determine the candidates involved in each text, and then learn the online business performance of each candidate.

For example, through the aforementioned web crawler unit 220 and semantic analysis unit 240, it is possible to analyze the degree of enthusiasm of netizens’ discussions about specific candidates in specific online discussion forums (for example, based on the number of main posts in the discussion forum involving specific candidates. ), the degree of active interaction (for example, based on the number of likes and shares related to a specific candidate in the discussion area), and the degree of enthusiasm (for example, based on the number of palindromes involving a specific candidate in the discussion area), etc., through these data , You can get network operation performance data.

The media poll data capturing module 300 is used to capture at least one media poll report to generate a media poll data D3. The media poll report is the report data obtained by the credible media using sample survey methods to directly survey voters' support for candidates through telephone and other methods. In one embodiment, as shown in FIG. Fifth, data acquisition module 300 polls media acquired by the at least one media-based media poll reported P _M1, P _M2 (FIG. 40a to two media and 40b, to generate two A media poll report P _M1 , P _M2 as an example), generate a media tendency data P _T and a release time data P _{TL corresponding to each media poll report P M1} , P _M2 , and based on the media tendency data P _T and The release time data P _TL adjusts the data of the media poll reports P _M1 and P _M2 to generate the media poll data D3. For example, if a media poll report has a specific political party tendency, the poll data of the candidates recommended by the specific political party in the report must be downgraded. If a certain media poll report is released recently, it is necessary to increase the influence of the media poll report on the generated media poll data D3. In one embodiment, the historical media poll data of each media can be used to analyze media tendency data to adjust the data of each media poll report.

As shown in FIG. 6, the crowd prediction data acquisition module 400 is used to capture at least one network statistical data P _N1 and P _N2 to generate a crowd prediction data D4. This network statistical data P _N1 may be statistical data of prediction market websites 50, such as Future Event Exchange, Smart Exchange (also known as Taipei Political Economy Exchange), etc. It can also be the network statistical data P _N2 generated by the direct expression of the masses, for example, by means of online voting. Generally speaking, the more participants and the more votes, the greater the influence on the generated mass forecast data D4.

The weight calculation module 500 generates weight data D5 corresponding to at least one of the aforementioned candidate strength data D1, Internet word-of-mouth prediction data D2, media poll data D3, and crowd prediction data D4 based on a voter structure data D _NS . In one embodiment, the voter structure data may include age structure data, geographic data, gender structure data, population movement data, and other data that may affect voting tendencies.

In one embodiment, the weight calculation module 500 can refer to the historical voter structure data and the historical election data D _H to calculate the weight distribution of different electoral districts corresponding to the current voter structure data D _NS. In one embodiment, the weight calculation module 500 may refer to historical voter structure data and previous election forecast data to calculate the weight distribution _{of different electoral districts corresponding to the current voter structure data D NS.}

In one embodiment, as shown in FIG. 7, the weight calculation module 500 has a weight generation unit 520 and a weight adjustment unit 540. The weight generating unit 520 calculates the corresponding initial estimated weight data D _W based on the historical voter structure data D _HNS and the historical election data D _H. The weight adjustment unit 540 distinguishes a network-native voter group from a non-network-native voter group based on the voter structure data D _{NS, and applies different weights to the two groups to obtain the weight data D5.}

For example, the voter structure data D _NS includes the voter age structure data. The weight adjustment unit 540 may define voters younger than forty years old as Internet native voters, and voters older than or equal to 40 years old as non-online native voters. For non-Internet native voters, the weight adjustment unit 540 increases the weight of candidate strength data D1 and media polling data D3. For the original online voters, the weight adjustment unit 540 increases the weights of the online word-of-mouth prediction data D2 and the crowd prediction data D4. But it is not limited to this. In one embodiment, for non-Internet native voters, the weight adjustment unit 540 may ignore the Internet word-of-mouth prediction data D2 and the crowd prediction data D4. For the native voters on the Internet, the weight adjustment unit 540 may ignore the candidate strength data D1 and the media polling data D3.

The election rate prediction module 600 adjusts the weights of candidate strength data D1, Internet word-of-mouth prediction data D2, media poll data D3 and crowd prediction data D4 based on the weight data D5 generated by the weight calculation module 500 to generate a candidate prediction Election rate D6. As shown in FIG. 8, in one embodiment, the winning rate prediction module 600 includes an advantage value calculation unit 620 and a winning rate calculation unit 640. Among them, the advantage value calculation unit 620 can adjust the weight of candidate strength data D1, Internet word-of-mouth prediction data D2, media poll data D3, and crowd prediction data D4 according to weight data D5 to calculate the relative advantage value data D of each candidate. _A. The election rate calculation unit 640 is based on the characteristics of electoral districts, such as differences in electoral systems, such as a single electoral district system (that is, a single elector is elected in a electoral district), a plural electoral district system (that is, multiple electors in a electoral district), etc., Calculate the probability of each candidate's election through the numerical data of relative advantage D _A , that is, the candidate's predicted election rate D6. For example, the election rate prediction module 600 may use machine learning or regression analysis to calculate the election probability of each candidate based on the relative advantage value of each candidate.

Please also refer to the ninth picture. Figure 9 is a flowchart of an embodiment of the election prediction method of the present invention. As shown in the figure, this election prediction method includes the following steps.

In step S120, at least one historical selection data is retrieved Analyze the weight of the corresponding winning factors to generate a candidate's strength data. In step S130, the website data of at least one website is retrieved and the website data is analyzed to generate an Internet word-of-mouth prediction data. In step S140, at least one media poll report is captured to generate a media poll data. In step S150, at least one network statistical data is retrieved to generate a crowd forecast data. Among them, the Internet word-of-mouth prediction data includes at least one of voter familiarity data, voter favorability data, and online business performance data.

The data in the aforementioned steps S120 to S150 can be generated simultaneously or sequentially. The order of the steps is not limited. In this embodiment, these data are generated at the same time to provide subsequent steps for processing.

Subsequently, as shown in step S160, based on a voter structure data, a weight data corresponding to at least one of the aforementioned candidate strength data, Internet word-of-mouth prediction data, media poll data, and crowd prediction data is generated.

Next, as shown in step S180, the weights of candidate strength data, Internet word-of-mouth prediction data, media poll data, and crowd prediction data are adjusted according to the weight data to generate a candidate's predicted election rate.

Compared with traditional election prediction methods that rely solely on sample surveys by polling agencies, the election prediction system provided by the present invention takes into account both media polling data and Internet crowd prediction data, and uses machine learning to obtain candidate strength data. Get online word-of-mouth data by means of sentiment analysis, and assign appropriate weights to each data. In this way, on the one hand, it can touch the ethnic groups that cannot be touched by traditional election forecasting methods. Groups, on the other hand, also take other possible influencing factors into consideration. In this way, the accuracy of election forecasts can be improved.

The above are only preferred embodiments of the present invention, and do not limit the present invention in any way. Any person skilled in the art, without departing from the scope of the technical means of the present invention, makes any form of equivalent replacement or modification or other changes to the technical means and technical content disclosed by the present invention, which does not depart from the technical means of the present invention. The content still falls within the protection scope of the present invention.

10: Election prediction system

100: Candidate Winning Factor Analysis Module

200: Internet word-of-mouth prediction module

300: Media Poll Data Acquisition Module

400: Crowd prediction data acquisition module

500: Weight calculation module

600: Winning rate prediction module

D1: Candidate's strength information

D2: Internet word-of-mouth prediction data

D3: Media polling data

D4: Crowd forecast data

D5: Weight data

D6: Candidates predict the election rate

Claims (4)

An election prediction system, comprising: a candidate winning factor analysis module for extracting at least one historical election data to analyze the corresponding weight of at least one winning factor to generate a candidate's strength data; a network The word-of-mouth prediction module is used to retrieve one of the website data of at least one website, and analyze the website data to generate an online word-of-mouth prediction data. The online word-of-mouth prediction data includes a voter familiarity data, a voter favorability data, and At least one of the network business performance data; a media poll data acquisition module for extracting at least one media poll report to generate a media poll data; a crowd forecast data acquisition module, It is used to retrieve at least one network statistical data to generate a mass prediction data; a weight calculation module is used to generate data corresponding to the candidate's strength, the Internet word-of-mouth prediction data, and the media based on a voter structure data Poll data and one of the mass prediction data; and a selection rate prediction module for adjusting the candidate's strength data, the Internet word-of-mouth prediction data, the media poll data and the mass prediction based on the weight data The weight of the data is used to generate a candidate’s predicted election rate; wherein, the voter structure data includes at least one age structure data, and the weight calculation module distinguishes a network-native voter group from a non-net voter group based on the age structure data Route native voter groups, and increase the weight of the candidate's strength data and the media polling data for the non-online native voter group; Among them, the candidate winning factor analysis module uses regression analysis and/or machine learning to generate the candidate's strength data; wherein, the online word-of-mouth prediction module uses semantic analysis and/or sentiment analysis to analyze The website data is used to generate the voter’s familiarity data and the voter’s favorability data. For example, the election prediction system of item 1 of the scope of patent application, wherein the voter structure data includes at least one geographic data, one gender structure data, one age structure data, and one population movement data. For example, the election prediction system of item 1 of the scope of patent application, in which the media poll data acquisition module obtains a media trend data corresponding to each media poll report, and adjusts the media poll according to the media trend data The reported data is used to generate the media polling data. For example, the election prediction system of item 1 of the scope of patent application, wherein the historical election data includes a candidate data, a political party recommendation data, a candidate experience data, and a government affairs experience data.

Images