Text Analysis Using Topics Model: What do Newspapers Tweet About and How do Readers React

Ng Yong Kad

6/25/2020

0. Introduction

Nowadays, most newspapers, if not all, engage their readers on social media platforms. In most of the time, the way of engagement is via the sharing of news, reports or some noteworthy analysis.

Here, a quick exploratory exercise is conducted on the activities of 4 major Malaysia English newspapers on Twitter - The Star, Malay Mail, New Strait Times, and The Edge. The Star and New Strait Times are dailies, whereas The Edge is a weekly. Malay Mail ceased its print operations in 2018 and is now totally digital. The overall goal of this exercise to guide or generate hypothesis for further analysis in this area.

1. Data

The data were collected on 23th to 24th Jun 2020 via Twitter API using R’s rtweet package. Twitter’s limit on the extraction of timelines is 3,200 of the most recent tweets. Due to the fact that each newspaper has different tweeting pattern and frequency, the sample period has to be normalized across all 4 newspapers.

After reviewing the raw data, the sample period is normalized to 5th Jun 2020 to 22nd Jun 2020. Figure 1 below presents the volume over the sample period for the respective newspaper.

From Figure 1, it is apparent that New Strait Times engaged its readers less on Twitter as compared to the other 3 newspapers. This is confirmed by the average of number tweets per day in Figure 2. Of the 4 newspapers, the most active is Malay Mail. This could be due to the fact that Malay Mail has ceased its print circulation and, therefore, active engagement with its readers on social media platform like Twitter is crucial.

Another noteworthy observation from Figure 1 is that the numbers of tweets are significantly lower in the following dates: 13th, 14th, 20th and 21th for The Edge. Co-incidentally, these 4 dates are weekends.

2. Analysis of Readers’ Interest

The interest of readers with respect to tweets can be assess via:

  1. tweets tagged as favorite by readers;
  2. tweets that are re-tweeted by readers.

The tweets with most readers tagged as favorite are shown Figure 3. For tweets that were tagged as favorite by more than 400 times, The Star has the most (3), whereas The Edge has none. Even though New Strait Times did not tweet as heavy as its peers, it managed to garner high volumes of favorite for some of its tweets. However, the top 3 favorite tweets of New Strait Times are arguably identical. These 3 tweets are related to a lady accidentally captured on TV footage. For the rest of the newspapers, the topics covered by the top 5 favorites are nonetheless quite concentrated: Malay Mail in politics, The Edge is financial news, and The Star in Covid-19 news. The lack of significant overlapping in the topics (of the top 5 tweets) across the 4 newspapers are indeed interesting. This could suggest that the readers base of the 4 newspapers are very different.

ind text
1_malaymail Daim: In Malaysia, when you steal Milo, you go to jail; but when you steal billions of ringgit, you walk away smiling
2_malaymail Report: CEO Jalil Rasheed to exit PNB
3_malaymail Klang man puts tombstone in pothole leading to authorities taking immediate action
4_malaymail Anwar: Pakatan has decided on PM candidate but needs to confer with allies first
5_malaymail Anwar says rejected DPM route to become PM, hints at mistrust of Dr Mahathir
1_NST_Online Mystery solved! Find out why ‘kakak baju putih’ was running with an orange cat in @BuletinTV3’s news footage.
2_NST_Online Perhaps if this report can go viral and attract the attention of ‘kakak baju putih’, she can tell us why she was running while carrying an orange cat.
3_NST_Online Mystery solved!
4_NST_Online #Covid19: Zero local transmissions a good sign, says Dr Noor Hisham
5_NST_Online Dr Noor Hisham delivers his briefing perfectly in Bahasa Malaysia and English, definitely an example to more than a million civil servants in this country.
1_staronline PKR: Yes only to Anwar, no to third time for Dr M
2_staronline Malaysia will be declared free from Covid-19 if no new cases for 28 days, says Health DG
3_staronline Outgoing Petronas CEO tells colleagues to stand by their principles
4_staronline ‘Malaysia has 93.8% recovery rate’
5_staronline Health Ministry studying ways to reduce congestion in health facilities
1_theedgemalaysia Update: Malaysia April unemployment rate spikes to 5% — highest since 1990 -
2_theedgemalaysia Outgoing Petronas CEO Wan Zul calls on colleagues to always stick to their principles -
3_theedgemalaysia Flash: PM: Property tax exemption for Malaysians for disposal of up to three properties between June 1, 2020 and Dec 31, 2021
4_theedgemalaysia Exclusive: Even $3 billion not enough to settle 1MDB case with Goldman, says Zafrul - Reuters
5_theedgemalaysia Flash: EPF records RM12.16b gross investment income for 1Q20

After examining tweets’ favorite counts, the attention can be turned to re-tweeted count. For a particular tweet, the expectation is that the favorite count is less than the re-tweeted count, where re-tweet is akin to sharing. Figure 4 below shows this is not necessarily true across the board. In fact, from Figure 4, the re-tweet counts of the top favorite tweet of New Strait Time (1_NST_Online) and The Edge (1-theedgeonline) are more than 100% of their favorite counts. This suggests the readers were particularly keen about the mystery (New Strait Time) and very concerned about the state of the economy (The Edge), that is, the readers staright away re-tweeted after reading them.

ind text
1_malaymail Daim: In Malaysia, when you steal Milo, you go to jail; but when you steal billions of ringgit, you walk away smiling
2_malaymail Report: CEO Jalil Rasheed to exit PNB
3_malaymail Klang man puts tombstone in pothole leading to authorities taking immediate action
4_malaymail Anwar: Pakatan has decided on PM candidate but needs to confer with allies first
5_malaymail Anwar says rejected DPM route to become PM, hints at mistrust of Dr Mahathir
1_NST_Online Mystery solved! Find out why ‘kakak baju putih’ was running with an orange cat in @BuletinTV3’s news footage.
2_NST_Online Perhaps if this report can go viral and attract the attention of ‘kakak baju putih’, she can tell us why she was running while carrying an orange cat.
3_NST_Online Mystery solved!
4_NST_Online #Covid19: Zero local transmissions a good sign, says Dr Noor Hisham
5_NST_Online Dr Noor Hisham delivers his briefing perfectly in Bahasa Malaysia and English, definitely an example to more than a million civil servants in this country.
1_staronline PKR: Yes only to Anwar, no to third time for Dr M
2_staronline Malaysia will be declared free from Covid-19 if no new cases for 28 days, says Health DG
3_staronline Outgoing Petronas CEO tells colleagues to stand by their principles
4_staronline ‘Malaysia has 93.8% recovery rate’
5_staronline Health Ministry studying ways to reduce congestion in health facilities
1_theedgemalaysia Update: Malaysia April unemployment rate spikes to 5% — highest since 1990 -
2_theedgemalaysia Outgoing Petronas CEO Wan Zul calls on colleagues to always stick to their principles -
3_theedgemalaysia Flash: PM: Property tax exemption for Malaysians for disposal of up to three properties between June 1, 2020 and Dec 31, 2021
4_theedgemalaysia Exclusive: Even $3 billion not enough to settle 1MDB case with Goldman, says Zafrul - Reuters
5_theedgemalaysia Flash: EPF records RM12.16b gross investment income for 1Q20

When the data is examined across time, the intensities of re-tweet and favorite do not appear to be correlated among the 4 newspapers. This is supported by the low correlation coefficients presented. The hypothesis that can be put forward is that each newspaper tweeted about different topics on each date to engage their readers, and their readers reacted (favorite or re-tweet) differently. The classification of topics can be examined using statistical technique and will be shown in the section below.

For now, what is more interesting from Figure 5 is all 4 newspapers have surge and spike on different dates. Without resorting to complext techniques, One can just look into what are the top tweets on those dates.

Correlation between Favorite Count
malaymail NST_Online staronline theedgemalaysia
malaymail 1.0000000 -0.1728637 0.2387417 -0.0566475
NST_Online -0.1728637 1.0000000 0.1052614 0.0295277
staronline 0.2387417 0.1052614 1.0000000 0.1914764
theedgemalaysia -0.0566475 0.0295277 0.1914764 1.0000000
Correlation between Re-tweet Count
malaymail NST_Online staronline theedgemalaysia
malaymail 1.0000000 0.2867568 0.4385848 0.0697081
NST_Online 0.2867568 1.0000000 0.3017692 -0.0182701
staronline 0.4385848 0.3017692 1.0000000 0.2823018
theedgemalaysia 0.0697081 -0.0182701 0.2823018 1.0000000

2.1 Tweets at Particular Date

Malay Mail has a surge in favorite count on the 17th. From Figure 6, 2 of its top 5 favorite tweets (see Figure 3) fell on this date. Thus, this explains the surge. On this date, all of the top 5 tweets are politics.

ind text
1_malaymail Daim: In Malaysia, when you steal Milo, you go to jail; but when you steal billions of ringgit, you walk away smiling
2_malaymail Anwar: Pakatan has decided on PM candidate but needs to confer with allies first
3_malaymail Dr Wan Azizah defends Hannah, slams intimidation after probe into child marriage tweet
4_malaymail Scared of DAP? Don’t call yourselves Malays, Dr Mahathir tells ‘cowards’
5_malaymail Malaysia needs ‘fresh start’ and Anwar best PM candidate, PKR man says as DAP demands answer

New Strait Times has 2 spikes on the 14^th and 16th. The spikes of favorite and re-tweet counts on the 14th can be attributed attributable to the a lady accidentally captured on TV footage. The surge of re-tweet count on the 17th is related to a trial of a politician. For unknown reason, New Strait Times tweeted twice about the trial and both tweets are identical words by words.

ind text
1_NST_Online Mystery solved! Find out why ‘kakak baju putih’ was running with an orange cat in @BuletinTV3’s news footage.
2_NST_Online Perhaps if this report can go viral and attract the attention of ‘kakak baju putih’, she can tell us why she was running while carrying an orange cat.
3_NST_Online Mystery solved!
4_NST_Online All it took was 55 days of free movement for #Covid19 to hit #China back with a vengeance.
5_NST_Online #TV3 airs report on highway users, Netizens zoom in on ‘kakak’ carrying orange cat

ind text
1_NST_Online Mystery solved! Find out why ‘kakak baju putih’ was running with an orange cat in @BuletinTV3’s news footage.
2_NST_Online Perhaps if this report can go viral and attract the attention of ‘kakak baju putih’, she can tell us why she was running while carrying an orange cat.
3_NST_Online Mystery solved!
4_NST_Online All it took was 55 days of free movement for #Covid19 to hit #China back with a vengeance.
5_NST_Online “He was found unconscious at the centre and was sent via ambulance to Hospital Canselor Tuanku Mukhriz at Universiti Kebangsaan Malaysia.” #Covid19 #Coronavirus
1_NST_Online “It’s an interesting thing working in an unfamiliar environment. It’s like going on a new adventure everyday.”
2_NST_Online Datuk Seri Ahmad #ZahidHamidi’s younger brother brought two properties worth RM5.9 million using monies purportedly from #YayasanAkalbudi.
3_NST_Online Malaysia is seeing an upward trend in #dengue cases with an average increase of eight per cent each week.
4_NST_Online Datuk Seri Ahmad #ZahidHamidi’s younger brother brought two properties worth RM5.9 million using monies purportedly from #YayasanAkalbudi.
5_NST_Online The survey also revealed that younger #lawyers were more likely to be sexually propositioned by their colleagues, clients or bosses. #sexualharassment #harassment

How about the tweets on the 9th for The Star? The top 5 favorite and re-tweeted tweets are identical on this date (even though their ranking based on re-tweet counts are different). The top 2 favorite and re-tweeted tweets are both related to jobs and refugees.

ind text
1_staronline Ismail Sabri: Rohingya refugees will be deported back to Bangladesh
2_staronline Foreigners no longer allowed to apply for hawker licences in S’gor, says Rodziah
3_staronline MR DIY has 1,000 job vacancies
4_staronline This 21-year-old animator from Batu Pahat, Johor, is in the running to win one of the comic book industry’s highest honours this July, thanks to her autobiographical webcomic ‘Fried Rice’. .
5_staronline Former Sabah CM Musa Aman acquitted of 46 charges of corruption, money laundering

ind text
1_staronline MR DIY has 1,000 job vacancies
2_staronline Ismail Sabri: Rohingya refugees will be deported back to Bangladesh
3_staronline Former Sabah CM Musa Aman acquitted of 46 charges of corruption, money laundering
4_staronline This 21-year-old animator from Batu Pahat, Johor, is in the running to win one of the comic book industry’s highest honours this July, thanks to her autobiographical webcomic ‘Fried Rice’. .
5_staronline Foreigners no longer allowed to apply for hawker licences in S’gor, says Rodziah

For the Edge on the 15th, the record high unemployment rate attracted a lot of reactions from its readers.

ind text
1_theedgemalaysia Update: Malaysia April unemployment rate spikes to 5% — highest since 1990 -
2_theedgemalaysia Flash: PNB announces resignation of president and group chief executive Jalil Rasheed
3_theedgemalaysia Flash: FBM KLCI tumbles 3.25% or 50.25 points to 1,495.77
4_theedgemalaysia Aviation: Mavcom, AGC objecting to AirAsia leave application for judicial review of RM2m penalty -
5_theedgemalaysia MATRADE introduces MDG guideline to support exporters - Bernama

ind text
1_theedgemalaysia Update: Malaysia April unemployment rate spikes to 5% — highest since 1990 -
2_theedgemalaysia Flash: PNB announces resignation of president and group chief executive Jalil Rasheed
3_theedgemalaysia Flash: FBM KLCI tumbles 3.25% or 50.25 points to 1,495.77
4_theedgemalaysia MATRADE introduces MDG guideline to support exporters - Bernama
5_theedgemalaysia The departure of Petronas president and CEO Tan Sri Wan Zulkiflee Wan Ariffin before the end of his tenure has raised eyebrows. We look at the circumstances surrounding the leadership change, and the challenges the new chief Tengku Muhammad Taufik Tengku Aziz will face.

From the simple exercise conducted above, it is apparent that the behavior and the reactions of the readers toward tweets are different across the 4 newspapers. This could be due to each newspaper has its own specialty (as in topics/areas such as politics, financial, and etc.) and the readers were attracted to it due to its specialty.

3. Topics Analysis

Unsupervised clustering correlated topics model (CTM) approach is used to categorize the topics associated with the tweets. The tweets are first tokenized into word by using R’s tidytext package. For data cleaning purposes, numbers and ringgit amounts are first removed. Then, stop words are removed using the list embedded in tidytext package. Additional stop words such as “Tan Sri”, “Datuk”, “Seri”, “Malay Mail”, news agencies “Bloonberg”, “Bernama”, Reuters“,”June“, and”Malaysia" are also removed. These additional stop words appear in a large number of tweets and do not provide important information to the formation of topics. Therefore, their removal will make the data cleaner and the topics more distinctive.

CTM clustering can be conducted in R using the topicmodels package. Before running CTM clustering, the clean tokenized data needs to be converted into document-term matrix with cast_dm() function of the tidytext package.

The number of topics (i.e. clusters), k, needs to be specified before hand. After numerous experiments, k = 6 appears to work well. Setting k > 6 does not result in more distinctive clusters, whereas setting k < 6 will result in coarser clusters. The results on beta is shown in Figure 13. Beta is the probability that a word belongs to a particular topic. In Topic 1, the high probability words are “Covid,”Singapore“,”China" and “Malaysians”. It could be inferred that Topic 1 is related to the pandemic CoVid-19 and the situations in the region.

Topic 2 includes “oil”, “market”, “economy” and “stock” which are financial related. Topic 3 has “penjana”, “prihatin”, and “government” points to this topic is about government policy. Topic 4 contains “court”, “trial”, “tax” and “bank” and thus suggest this topic is about domestic news. Topic 5 can be attributed to police enforcement. Finally, Topic 6 could be conveniently labeled as “others” as there doesn’t seem to have a common topic for the words.

In short, the CTM clustering suggests 6 topics in the data:

  1. Covid-19 and Regional;
  2. Financial news;
  3. Govt policy;
  4. Domestic news;
  5. Police enforcement;
  6. Others.

While beta is a probability pertinent to a word, gamma is for an entire document which is a tweet for in this case. A tweet has a gamma for each of the 6 topics. Then, a topic can be assigned to a tweet based on the highest gamma.

The distribution of tweets based on topics can be examined as in Figure 14. In terms of topics, Mail Mail has quite equal distribution across the topics, albeit slightly higher in Covid-19 and Regional. New Strait Times and The Star also geared more towards Covid-19 and Regional, but the distributions in other topics are not as even as Malay Mail. As for The Edge, it has highest percentage in Financial News.

3.1 Accuracy

How accurate are the topic assignments? The table below shows the topic assignment of the top 10 favorite tweets. Overall, the assignments can be deemed quite accurate. Note, however, that some wrong assignments are to be expected as statistical techniques cannot be 100% accurate.

Top Favorite Tweet and Their Topic Assignment
screen_name post_date text favorite_count topic_desc
malaymail 06-17 Daim: In Malaysia, when you steal Milo, you go to jail; but when you steal billions of ringgit, you walk away smiling 450 Police Enforcement
malaymail 06-13 Report: CEO Jalil Rasheed to exit PNB 421 Govt Policy
malaymail 06-16 Klang man puts tombstone in pothole leading to authorities taking immediate action 376 Police Enforcement
malaymail 06-17 Anwar: Pakatan has decided on PM candidate but needs to confer with allies first 286 Others
malaymail 06-21 Anwar says rejected DPM route to become PM, hints at mistrust of Dr Mahathir 253 Covid 19 & Regional
malaymail 06-16 TNB to review mechanisms after complaints of sky-high bills during MCO, says chairman 252 Domestic News
malaymail 06-11 Fresh leak of February Bersatu meeting suggests PM’s resignation was preplanned 242 Others
malaymail 06-10 Malaysians demand removal of ‘dehumanising’ billboard portraying Lim Kok Wing as ‘King of Africa’ 184 Govt Policy
malaymail 06-06 Chin Tong: Dr Mahathir, Anwar should work together for sake of nation to oust Muhyiddin, Najib 175 Domestic News
malaymail 06-19 Dr Mahathir says will never work with Muhyiddin, confident voters will reject money politics in GE15 173 Others
NST_Online 06-14 Mystery solved! Find out why ‘kakak baju putih’ was running with an orange cat in @BuletinTV3’s news footage. 690 Domestic News
NST_Online 06-14 Perhaps if this report can go viral and attract the attention of ‘kakak baju putih’, she can tell us why she was running while carrying an orange cat. 407 Domestic News
NST_Online 06-14 Mystery solved! 307 Domestic News
NST_Online 06-09 #Covid19: Zero local transmissions a good sign, says Dr Noor Hisham 274 Covid 19 & Regional
NST_Online 06-18 Dr Noor Hisham delivers his briefing perfectly in Bahasa Malaysia and English, definitely an example to more than a million civil servants in this country. 223 Domestic News
NST_Online 06-11 Malaysia will declare itself free of #Covid19 when the country records zero new cases of infection for 28 consecutive days. 208 Covid 19 & Regional
NST_Online 06-08 #NewZealand has won praise for its handling of the pandemic, which involved a strict seven-week #lockdown that ended last month after the virus was contained. #coronavirus #covid19 139 Others
NST_Online 06-13 Before Tan Sri Abdul Aziz Abdul Rahman clocked out on his last day in MAS’s office in 1991, he left it with a RM5 billion reserve, which was a huge sum 29 years ago. 120 Police Enforcement
NST_Online 06-07 Malaysia can break the #Covid19 infection chain without the need for a vaccine. 101 Covid 19 & Regional
NST_Online 06-20 After 3-month closure, #Makkah mosques to reopen 101 Govt Policy
staronline 06-19 PKR: Yes only to Anwar, no to third time for Dr M 705 Others
staronline 06-11 Malaysia will be declared free from Covid-19 if no new cases for 28 days, says Health DG 491 Covid 19 & Regional
staronline 06-16 Outgoing Petronas CEO tells colleagues to stand by their principles 443 Financial News
staronline 06-18 ‘Malaysia has 93.8% recovery rate’ 334 Financial News
staronline 06-08 Health Ministry studying ways to reduce congestion in health facilities 265 Covid 19 & Regional
staronline 06-12 Covid-19 positive barber had been providing illegal house-to-house haircuts during MCO 256 Others
staronline 06-09 Ismail Sabri: Rohingya refugees will be deported back to Bangladesh 255 Covid 19 & Regional
staronline 06-10 Anwar: Pakatan Plus’ 107 MPs will move forward as a team 230 Others
staronline 06-14 Malaysia has been recognised as among the five nations in the world to have succeeded in controlling the spread of Covid-19, said Datuk Seri Ismail Sabri Yaakob. Well done, everyone! 224 Covid 19 & Regional
staronline 06-16 Abdullah Sani: PKR leaders ‘totally’ rejected Dr M as PM during Monday (June 15) meeting 221 Covid 19 & Regional
theedgemalaysia 06-15 Update: Malaysia April unemployment rate spikes to 5% — highest since 1990 - 315 Financial News
theedgemalaysia 06-14 Outgoing Petronas CEO Wan Zul calls on colleagues to always stick to their principles - 114 Financial News
theedgemalaysia 06-05 Flash: PM: Property tax exemption for Malaysians for disposal of up to three properties between June 1, 2020 and Dec 31, 2021 103 Domestic News
theedgemalaysia 06-06 Exclusive: Even $3 billion not enough to settle 1MDB case with Goldman, says Zafrul - Reuters 90 Govt Policy
theedgemalaysia 06-13 Flash: EPF records RM12.16b gross investment income for 1Q20 84 Domestic News
theedgemalaysia 06-14 Highlight: Khazanah’s Zulqarnain Onn a candidate to replace Jalil at PNB - 80 Govt Policy
theedgemalaysia 06-09 Highlight: Prosecution drops all charges against Musa Aman, judge grants acquittal - 77 Domestic News
theedgemalaysia 06-18 The majority of the 4.1 million applications for the i-Lestari Account 2 withdrawals from the EPF are those from the B40 or bottom 40% income group. Worryingly, some have exhausted their Account 2 after only 2 months. Read 71 Domestic News
theedgemalaysia 06-06 Flash: Petronas president and CEO Wan Zul made Malaysia Airlines chairman 66 Financial News
theedgemalaysia 06-21 This year’s Big Bad Wolf sale is going to be massive and entirely online. 57 Others

4. Readers’ Interest

What do the readers like to read about? For expository purposes, a tweet is defined as liked by the readers if the favorite count is greater than 50. The threshold is set rather high in order to make this exercise as insightful as it can be.

From Figure 15 below, as compared to its peers, The Star’s readers seem to like all of the topics except for Domestic News. It is surprising that The Star even outperformed The Edge in Financial News. For Malay Mail, its readers liked Domestic News and Others, where Others includes politics. For New Strait Times, its readers were attracted to its Domestic News and Government Policy. Lastly, for The Edge, its readers focused on Domestics News, Government Policy and Financial News. From this figure, The Star seems to be very successful in tweeting the news that the readers like to read, as compared to its peer.

Re-tweets can be viewed as something that elicit stronger reaction from the readers. A reader re-tweets only if he/she finds the tweet very important or useful, not only for him/herself, but also for others. Otherwise, the readers will not re-tweet to share with others. Therefore, re-tweet is an indication that the particular tweet is important.

Since re-tweet count is in general lower than favorite count, the threshold is set at 30. That is, only the distribution of topics of the tweets with re-tweet count greater than 30 is considered. The results from Figure 16 is interesting. The readers of The Edge is much willing to share Domestics News and Financial News. In summary, the readers find the followings newspapers provide important news in the area of:

  1. Covid-19 and Regional: New Strait Times, The Star.
  2. Domestic News: Malay Mail, New Strait Times, The Edge.
  3. Financial News: The Star, The Edge.
  4. Govt Policy: New Strait Times, The Star.
  5. Others: Malay Mail, The Star.
  6. Police Enforcement: The Star.

5. Conclusion

From a relative small sample, it is quite clear that not all newspapers uses Twitter in the same intensity. It is also interesting to observe that the top favorite tweets and the top re-tweeted tweets are very different across the newspapers.

Clustering technique allows for the examination of what the newspapers tweet about and what the readers like to read. The clustering technique used in this exercise results in 6 topics. It seems that The Star, based on this data, is rather successful in engaging its readers as compared to its peers. This is because the percentages of tweets that were re-tweeted are higher versus those of its peers, across almost all topics.