Academia.eduAcademia.edu

Figure 2 – uploaded by Robson de Oliveira Albuquerque

See full PDFdownloadDownload figure
Figure 2. Data collected on Twitter between 15 June and 31 July 2018. Histogram with the amount of tweets and retweets collected per day (a). Total amount of tweets and retweets (b). Total number of users and hashtags (c). Figure 2 represents through the visualization layer the total set of data collected by the environment. In Figure 2a, peaks of tweets and retweets that were posted on the Brazilian team match days (June 17th (31,957), June 22th (33,949), June 27th (35,204), July 2nd (34,968) and July 6th (25,506)) can be noticed, the most commented hashtag about the topic in the quarterfinals being detailed in Section 6.8, in Section 6.9, the identification of outliers (users with discrepant activities over a period of time), and in Section 6.10, the analysis of a botnet used to spread tweets. The biggest spike of published tweets and retweets was observed on July 7th (42,733), a day after the quarterfinals (July 6th), the match in which the Brazilian National Soccer Team was eliminated from the FIFA World Cup 2018. On the other hand, there is also a drop after July 10th, indicating that the theme “Brazilian National Soccer Team” was no longer so present on Twitter. It is also possible to identify that 122,975 users posted 730,850 posts and 4240 hashtags (Figure 2b,c).

Figure 2 Data collected on Twitter between 15 June and 31 July 2018. Histogram with the amount of tweets and retweets collected per day (a). Total amount of tweets and retweets (b). Total number of users and hashtags (c). Figure 2 represents through the visualization layer the total set of data collected by the environment. In Figure 2a, peaks of tweets and retweets that were posted on the Brazilian team match days (June 17th (31,957), June 22th (33,949), June 27th (35,204), July 2nd (34,968) and July 6th (25,506)) can be noticed, the most commented hashtag about the topic in the quarterfinals being detailed in Section 6.8, in Section 6.9, the identification of outliers (users with discrepant activities over a period of time), and in Section 6.10, the analysis of a botnet used to spread tweets. The biggest spike of published tweets and retweets was observed on July 7th (42,733), a day after the quarterfinals (July 6th), the match in which the Brazilian National Soccer Team was eliminated from the FIFA World Cup 2018. On the other hand, there is also a drop after July 10th, indicating that the theme “Brazilian National Soccer Team” was no longer so present on Twitter. It is also possible to identify that 122,975 users posted 730,850 posts and 4240 hashtags (Figure 2b,c).

Related Figures (33)

Table 1. OctopusViz highlighted features compared to related works. 4. Problem Statement and Proposed Solution
Table 1. OctopusViz highlighted features compared to related works. 4. Problem Statement and Proposed Solution
Figure 1. OctopusViz architecture. 4.2. Proposed Environment Architecture As a fundamental requirement, and for security and privacy reasons, the environment should ensure that the source of the collection is anonymous in order to not create an opportunity for data change trends or the possibility of influencing the data source. Therefore, one of the ways to achieve this requirement is by using a VPN, so that any collecting sources are not easily identified, avoiding the risk of data contamination by the source. Figure 1 shows the proposed environment’s architecture. The environment’s physical architecture employs a host connected in a Demilitarized Zone (DMZ) and to a protected network of the University of Brasilia (UnB) Research Laboratory. In this host, there are three virtual machines.
Figure 1. OctopusViz architecture. 4.2. Proposed Environment Architecture As a fundamental requirement, and for security and privacy reasons, the environment should ensure that the source of the collection is anonymous in order to not create an opportunity for data change trends or the possibility of influencing the data source. Therefore, one of the ways to achieve this requirement is by using a VPN, so that any collecting sources are not easily identified, avoiding the risk of data contamination by the source. Figure 1 shows the proposed environment’s architecture. The environment’s physical architecture employs a host connected in a Demilitarized Zone (DMZ) and to a protected network of the University of Brasilia (UnB) Research Laboratory. In this host, there are three virtual machines.
Table 2. Host features.
Table 2. Host features.
Table 3. Guest systems and their settings.
Table 3. Guest systems and their settings.
Table 4. Function for translation and correction of tweets.
Table 4. Function for translation and correction of tweets.
Table 5. Corpus words stop words and special characters. Table 6. Function for cleaning tweets.
Table 5. Corpus words stop words and special characters. Table 6. Function for cleaning tweets.
Table 7. Function for tokenization of tweets. 5.3. Phase 3: Classification Sublayer
Table 7. Function for tokenization of tweets. 5.3. Phase 3: Classification Sublayer
Table 8. Function for tweets’ classification.
Table 8. Function for tweets’ classification.
Figure 3. Data collected on Twitter between 15 June and 31 July 2018. Message classification into tweets and retweets (a). Histogram with the messages’ classification into tweets and retweets collected per day (b). with 35,709 retweets. A difference (29,105 retweets-4647 tweets) also can be noticed on semifinal day (July 10th), when the team that eliminated Brazil in quarterfinals (Belgium) lost to France 1-0.
Figure 3. Data collected on Twitter between 15 June and 31 July 2018. Message classification into tweets and retweets (a). Histogram with the messages’ classification into tweets and retweets collected per day (b). with 35,709 retweets. A difference (29,105 retweets-4647 tweets) also can be noticed on semifinal day (July 10th), when the team that eliminated Brazil in quarterfinals (Belgium) lost to France 1-0.
Figure 4. Cloud of words with the identification from the most referenced hashtags in tweets and retweets
Figure 4. Cloud of words with the identification from the most referenced hashtags in tweets and retweets
Table 9. The five most referenced hashtags between 15 June and 31 July 2018.
Table 9. The five most referenced hashtags between 15 June and 31 July 2018.
Figure 5. Cloud with the most referenced hashtags. (a) indicates tweets and (b) retweets. Through the interaction provided by the tool, it was possible to apply filters to create and separate hashtags’ clouds by tweets and retweets. It was observed that the most referenced hashtags also depended on analyst interest in tweets (Figure 5a) or retweets (Figure 5b). It may be seen from the first two lines of the data represented in Table 10 that the hashtags #Copa2018 and #BRA were the most commented on for both tweets and retweets. In the third line, it is observed that the hashtags change according to that interest.
Figure 5. Cloud with the most referenced hashtags. (a) indicates tweets and (b) retweets. Through the interaction provided by the tool, it was possible to apply filters to create and separate hashtags’ clouds by tweets and retweets. It was observed that the most referenced hashtags also depended on analyst interest in tweets (Figure 5a) or retweets (Figure 5b). It may be seen from the first two lines of the data represented in Table 10 that the hashtags #Copa2018 and #BRA were the most commented on for both tweets and retweets. In the third line, it is observed that the hashtags change according to that interest.
Table 10. Hashtags most referenced by tweets or retweets. 6.5. User Analysis
Table 10. Hashtags most referenced by tweets or retweets. 6.5. User Analysis
Figure 6. Users who posted the most tweets and retweets.
Figure 6. Users who posted the most tweets and retweets.
Table 11. Rating of tweets and retweets by user. The environment also identifies mentions, likes, and hashtags quoted in each tweet and retweet. Four retweets that were posted by user InfosFutebol can be observed in Figure 7. It is interesting to note how many likes (defined by the column favorite_count—14,975, 10,462, 12,679, and 9999) each retweet received. The environment also identifies mentions, likes, and hashtags quoted in each tweet and retweet.
Table 11. Rating of tweets and retweets by user. The environment also identifies mentions, likes, and hashtags quoted in each tweet and retweet. Four retweets that were posted by user InfosFutebol can be observed in Figure 7. It is interesting to note how many likes (defined by the column favorite_count—14,975, 10,462, 12,679, and 9999) each retweet received. The environment also identifies mentions, likes, and hashtags quoted in each tweet and retweet.
Figure 7. Retweets published by user InfosFutebol. 6.6. Sentiment Analysis
Figure 7. Retweets published by user InfosFutebol. 6.6. Sentiment Analysis
Figure 8. Classification (positive, negative, and neutral) of tweets and retweets per day (a). Classification (positive, negative, and neutral) of tweets and retweets per hour (b). Figures 8-10 indicate user sentiment about the topic “Brazilian National Soccer Team” on eets and retweets between 15 June and 31 July 2018. According to Figure 8a, the highest peak positive-rated tweets and retweets (21,146) came on July 7th, the day after Belgium beat Brazil 2-1 quarterfinals. It can be noted that even losing the match, the users were favorable to the Brazilian m. It is also noted in Figure 8a that the highest peak with negative polarity occurred on June id (13,007) during Round 2 of three of Group E: Brazil 2 x 0 Costa Rica. Probably, this negative ercussion is due to the fact that Brazil tied the match on June 17th during Round 1 of three of oup E: Brazil 1 x 1 Switzerland.
Figure 8. Classification (positive, negative, and neutral) of tweets and retweets per day (a). Classification (positive, negative, and neutral) of tweets and retweets per hour (b). Figures 8-10 indicate user sentiment about the topic “Brazilian National Soccer Team” on eets and retweets between 15 June and 31 July 2018. According to Figure 8a, the highest peak positive-rated tweets and retweets (21,146) came on July 7th, the day after Belgium beat Brazil 2-1 quarterfinals. It can be noted that even losing the match, the users were favorable to the Brazilian m. It is also noted in Figure 8a that the highest peak with negative polarity occurred on June id (13,007) during Round 2 of three of Group E: Brazil 2 x 0 Costa Rica. Probably, this negative ercussion is due to the fact that Brazil tied the match on June 17th during Round 1 of three of oup E: Brazil 1 x 1 Switzerland.
Figure 9. Ranking of tweets and retweets per day with neutral (a), positive (b), and negative (c) polarities.
Figure 9. Ranking of tweets and retweets per day with neutral (a), positive (b), and negative (c) polarities.
Figure 10. Overall ranking of tweets and retweets by the algorithm Pattern Analyzer.
Figure 10. Overall ranking of tweets and retweets by the algorithm Pattern Analyzer.
Table 12. Peak polarities of tweets and retweets per day.
Table 12. Peak polarities of tweets and retweets per day.
Figure 11. Relationship among the entities’ polarity (positive, negative, and neutral) and the user (a). Users who published the most tweets and retweets with positive polarity (b). Link analysis aims to integrate information from various Twitter entities (users, hashtags, tweets, retweets, mentions, sentiment polarity, and images) to debug, organize, and interpret raw data, allowing the analyst to detect existing patterns and connections. Figure 11a presents a graph of the connection between the polarities (positive, negative, and neutral) and the user. For this analysis, we randomly selected 3000 indexed users in the environment. It can be seen from the graph in Figure 11a that more users were related to the neutral and positive polarities. It was also found that several users were related to two or three different polarities (positive, negative, and neutral). In this case, some tweets and retweets of those users were rated as positive and others as negative and neutral.
Figure 11. Relationship among the entities’ polarity (positive, negative, and neutral) and the user (a). Users who published the most tweets and retweets with positive polarity (b). Link analysis aims to integrate information from various Twitter entities (users, hashtags, tweets, retweets, mentions, sentiment polarity, and images) to debug, organize, and interpret raw data, allowing the analyst to detect existing patterns and connections. Figure 11a presents a graph of the connection between the polarities (positive, negative, and neutral) and the user. For this analysis, we randomly selected 3000 indexed users in the environment. It can be seen from the graph in Figure 11a that more users were related to the neutral and positive polarities. It was also found that several users were related to two or three different polarities (positive, negative, and neutral). In this case, some tweets and retweets of those users were rated as positive and others as negative and neutral.
Table 13. Polarity of tweets and retweets. 6.7. Link Analysis
Table 13. Polarity of tweets and retweets. 6.7. Link Analysis
Figure 12. Relationship between hashtag and user entities (a). Hashtag most referenced in tweets and retweets (b).
Figure 12. Relationship between hashtag and user entities (a). Hashtag most referenced in tweets and retweets (b).
Figure 13. Number of users and hashtags (a), tweets and retweets (b). Most commented on hashtag (c). Rating of posts that have been included with that hashtag in tweets and retweets (d). Users who sent the most messages with that hashtag (e). Figure 13. Number of users and hashtags (a), tweets and retweets (b). Most commented on hashtag (c). i On July 6th, during Brazil’s fifth match, in the quarterfinals, the hashtag #Copa2018 stood out on he Dashboard. For more detailed information, a filter was applied to that hashtag. The filter result and s resulting information can be observed as follows: Figure 13a,b shows that 1022 users posted 4098 t i tweets and retweets with that hashtag; another 368 hashtags (Figure 13a) were related to #Copa2018; i can be noted in Figure 13d that 2869 posts with that hashtag were classified as retweets and 1229 as tweets; user torcidasfotos (Figure 13e) was the one who sent the most messages with that hashtag; may be noticed in Figure 14a that 31.92% of users were in favor, 13.88% against, and 54.2% neutral; the hashtag #Copa2018 (Figure 14b) was classified into three different polarities (positive, negative, and neutral); in Figure 14c, it may be spotted that the highest peak of tweets and retweets with that hashtag happened on June 27th (773).
Figure 13. Number of users and hashtags (a), tweets and retweets (b). Most commented on hashtag (c). Rating of posts that have been included with that hashtag in tweets and retweets (d). Users who sent the most messages with that hashtag (e). Figure 13. Number of users and hashtags (a), tweets and retweets (b). Most commented on hashtag (c). i On July 6th, during Brazil’s fifth match, in the quarterfinals, the hashtag #Copa2018 stood out on he Dashboard. For more detailed information, a filter was applied to that hashtag. The filter result and s resulting information can be observed as follows: Figure 13a,b shows that 1022 users posted 4098 t i tweets and retweets with that hashtag; another 368 hashtags (Figure 13a) were related to #Copa2018; i can be noted in Figure 13d that 2869 posts with that hashtag were classified as retweets and 1229 as tweets; user torcidasfotos (Figure 13e) was the one who sent the most messages with that hashtag; may be noticed in Figure 14a that 31.92% of users were in favor, 13.88% against, and 54.2% neutral; the hashtag #Copa2018 (Figure 14b) was classified into three different polarities (positive, negative, and neutral); in Figure 14c, it may be spotted that the highest peak of tweets and retweets with that hashtag happened on June 27th (773).
Figure 14. Polarity of tweets and retweets (a). Graph with the relationship between hashtag and polarity entities (b). Histogram with the amount of tweets and retweets collected per day with that hashtag (c). 6.9. Outliers Analysis
Figure 14. Polarity of tweets and retweets (a). Graph with the relationship between hashtag and polarity entities (b). Histogram with the amount of tweets and retweets collected per day with that hashtag (c). 6.9. Outliers Analysis
Figure 15. Data collected from Twitter between 15 June and 21 June 2018. Discrepant users. For this work, we consider outliers elements those that do not follow a pattern of the set of users to which they were grouped according to the interest criteria analysis. They are users with discrepant activities for a period of time that require special attention as they usually produce values with unreliable effects. On June 21st, before the second match of the Brazilian National Soccer Team in the 2018 FIFA World Cup, it was found that the account dobresdelena presented a great distance from other users, being considered by the analysis an outlier (Figure 15). According to Table 14, users dobresdelena, lorenzopaag, whindersson, cleytu, adrianowilkson, and lacaxarruda) did not post tweets. It was also noted that the user dobresdelena had posted 6723 retweets. A detailed analysis was conducted to try to identify the characteristics of that user.
Figure 15. Data collected from Twitter between 15 June and 21 June 2018. Discrepant users. For this work, we consider outliers elements those that do not follow a pattern of the set of users to which they were grouped according to the interest criteria analysis. They are users with discrepant activities for a period of time that require special attention as they usually produce values with unreliable effects. On June 21st, before the second match of the Brazilian National Soccer Team in the 2018 FIFA World Cup, it was found that the account dobresdelena presented a great distance from other users, being considered by the analysis an outlier (Figure 15). According to Table 14, users dobresdelena, lorenzopaag, whindersson, cleytu, adrianowilkson, and lacaxarruda) did not post tweets. It was also noted that the user dobresdelena had posted 6723 retweets. A detailed analysis was conducted to try to identify the characteristics of that user.
Table 14. Numbers of tweets and retweets by user.
Table 14. Numbers of tweets and retweets by user.
Figure 16. Dashboard with information about user dobresdelena. Number of hashtags (a), tweets and retweets (b). Discrepant user (c). Number of messages classified as retweets (d). Histogram with the amount of retweets collected per day (e) . Interestingly, all retweets were rated by the Patterns Analyzer algorithm as positive, influencing the analysis of user sentiment regarding “Brazilian National Soccer Team”. Information details may be observed as follows: in Figure 16a, it can be observed that user dobresdelena did not post any hashtag, Figure 16d shows that all posted messages were classified as retweets; it can be noticed that the highest peak of retweets (Figure 16e) happened on June 18th (3917); 100% of the retweets were classified as positives (Figure 17a,b); the highest peak of retweets (374) rated as positive happened on June 17th at 10 p.m. (Figure 17c).
Figure 16. Dashboard with information about user dobresdelena. Number of hashtags (a), tweets and retweets (b). Discrepant user (c). Number of messages classified as retweets (d). Histogram with the amount of retweets collected per day (e) . Interestingly, all retweets were rated by the Patterns Analyzer algorithm as positive, influencing the analysis of user sentiment regarding “Brazilian National Soccer Team”. Information details may be observed as follows: in Figure 16a, it can be observed that user dobresdelena did not post any hashtag, Figure 16d shows that all posted messages were classified as retweets; it can be noticed that the highest peak of retweets (Figure 16e) happened on June 18th (3917); 100% of the retweets were classified as positives (Figure 17a,b); the highest peak of retweets (374) rated as positive happened on June 17th at 10 p.m. (Figure 17c).
Figure 17. Information about dobresdelena’s retweet polarity (a). Relationship among the entities’ nolarity (nocitive) and the 11cer (h) Claccification (nocitive) of retweete ner hour (c)_
Figure 17. Information about dobresdelena’s retweet polarity (a). Relationship among the entities’ nolarity (nocitive) and the 11cer (h) Claccification (nocitive) of retweete ner hour (c)_
In this scenario, we analyzed dobresdelena’s account on Twitter for relevant details to the interest analysis. Figure 18 shows that at the time of this analysis, the last retweet posted by that user had 36,334 likes. It can be noted that those retweets are always posted at the same time with a variation of seconds (June 17th, 23:56:36, June 18th, 23:56:36, and June 19th, 23:56:25). It can also be identified that in retweets, there is a link (https://t.co/I[YLzqYHXF) featuring three photos of the Brazilian national team goalkeeper, which were used as an account strategy to earn likes.
In this scenario, we analyzed dobresdelena’s account on Twitter for relevant details to the interest analysis. Figure 18 shows that at the time of this analysis, the last retweet posted by that user had 36,334 likes. It can be noted that those retweets are always posted at the same time with a variation of seconds (June 17th, 23:56:36, June 18th, 23:56:36, and June 19th, 23:56:25). It can also be identified that in retweets, there is a link (https://t.co/I[YLzqYHXF) featuring three photos of the Brazilian national team goalkeeper, which were used as an account strategy to earn likes.
Figure 19. Reverse image analysis with the tools TinEye (a) and Google Images (b).
Figure 19. Reverse image analysis with the tools TinEye (a) and Google Images (b).
Figure 20. Analysis through the TweetBotOrNot tool. The tool TweetBotOrNot [52] was also used to analyze the account dobresdelena. This tool uses machine learning to analyze metadata and classify account behavior on Twitter, indicating whether the user is a bot [52] or not. Figure 20 shows that TweetBotOrNot rated the account dobresdelena as a possible bot (0.813).
Figure 20. Analysis through the TweetBotOrNot tool. The tool TweetBotOrNot [52] was also used to analyze the account dobresdelena. This tool uses machine learning to analyze metadata and classify account behavior on Twitter, indicating whether the user is a bot [52] or not. Figure 20 shows that TweetBotOrNot rated the account dobresdelena as a possible bot (0.813).
Related topics:
Analytical ChemistryElectrical and Electronic Engineering
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved