What is Sentiment140?
Sentiment140 – which can be accessed via http://www.sentiment140.com/ – is a sentiment analysis tool for Twitter. Sentiment140 aggregates and examines all tweets and attempts to decipher whether they are positive or negative in relation to the subject of the comment. This tool may be deemed useful to various media professionals who would be interested in the general social media (public) feeling about a specified subject.
Below are six links which additionally describe Sentiment140 and/or discuss the use of sentiment analysis in detail.
“Sentiment140 – General Information” (Sentiment 140, n.d.): An overview of Sentiment140 provided by the service’s creators.
“How Sentiment Analysis works” (Jenkins, 2009): A detailed overview of how sentiment analysis is used and the issues associated with it.
“Live Twitter Sentiment Analysis” (Annenberg Innovation Lab, 2011): An example of how sentiment analysis can be used to predict an election.
“Is Automated Sentiment Analysis Reliable?” (Beal, 2009): An article which questions the reliability and usefulness of sentiment analysis.
“Predicting Elections with Twitter: What 140 Characters Reveal about Political Sentiment” (Tumasjan, Sprenger, Sandner & Welpe, 2010): A research paper which examines whether twitter sentiment analysis can be used to accurately predict an election.
“What’s Great and What’s Not: Learning to Classify the Scope of Negation for Improved Sentiment Analysis” (Councill, McDonald & Velikovich, 2010): A research paper which discusses some issues with sentiment analysis and develops a project which aims to improve sentiment analysis accuracy.
How to use Sentiment140
Sentiment140 is quite a basic tool to use. To get started, type in the URL http://www.sentiment140.com/. This will take you to the screen seen in Figure 1 (see below). You have the option to change the language from English to Spanish and vice versa if need be. Next, type in a subject that you want to examine for recent sentiment analysis.
After clicking the ‘Search’ button, you will be taken to a page like that seen in Figure 2. You will be presented with two graphs (a pie graph and a bar graph) which break down the number of positive and negative tweets about your subject. The bar graph also reveals the number tweets that have been taken into account, which allows you to quantify the strength of the data field. Both graphs can be saved as images for later use. In addition, the tweets that have contributed to the graphs are made available for the user to see – positive comments are highlighted green, negative comments are highlighted red and neutral comments remain white – so you can look for interesting comments/tweets and also verify whether you think the graph is accurate.
Once you have acquired what you want from this search, you can simply type another subject in and start the process again!
Analysis of Sentiment140 – just how useful is it?
All tools have affordances and constraints; Hutchby (2001) proposes that affordances are properties that enable action and constraints are properties that limit action. These terms can be broken down further and categorized as either technical or purposive. Technical affordances and constraints relate to how the user interface physically enables or limits action. On the other hand, purposive affordances and constraints relate to how the user interface encourages or discourages action based on its intended purpose (Rintel, 2012).
Sentiment140 has several technical affordances. First of all, the site aggregates the most recent tweets about a specified subject and attempts to decipher whether they are positive or negative (or neutral) in relation to the subject. This has the advantage of providing users with a virtually live account of what is being said on Twitter about the subject that is being researched, which can help media professionals develop story ideas and angles.
Sentiment140 presents its data in a simple format, with the percentage of positive versus negative tweets being put into a pie graph, alongside a bar graph which shows the numerical value of sentiments being expressed. The site also places the actual tweets below the graphs. This all adds to a final product which is easy to read and simple to make conclusions from.
Other affordances provided in the layout are that the bar graphs actually reveal the number of positive and negative tweets used in the sample size, which can help to show how reliable the data is. Additionally the tweets that are shown include the date and author of the comment, which results in sources that can be cited easily.
Sentiment140 encourages the option for users to be notified of negative tweets about keywords for a monthly fee. This would be classified as a purposive affordance of the service, as it is intended mainly for businesses to limit social media damage to their products. By getting alerts of negative tweets, a business may benefit by knowing sooner if their website is down for example, or if their brand is being defamed (Sentiment140, n.d.). A negative of this affordance is that it costs money to be provided with this extra function of the site.
According to some scholars, another purposive affordance of Sentiment140 and other sentiment analysis tools is that they can be used to even predict election results. A research article by Tumasjan, Sprenger, Sandner and Welpe (2010) demonstrated this by analyzing over 100,000 relevant Twitter messages in the lead up to the 2009 German federal election. They found that the volume of tweets about each political party strongly reflected preferences and nearly mirrored traditional election polls. More importantly, it was observed that the “sentiment of Twitter messages closely corresponds to political programs, candidate profiles, and evidence from the media coverage of the campaign trail” (Tumasjan, Sprenger, Sandner & Welpe, 2010, p. 183). There are many purposive affordances of sentiment analysis similar to this kind of project.
There are several technical constraints associated with the service provided by Sentiment140. To start with, the website is quite basic in design. This has the advantage of allowing easy access around the site, but it does mean that the options with how you can use the site are limited. Furthermore, searches are limited to just Twitter sources, and only the most recent tweets can be analyzed (i.e. you are not able to choose an earlier date range for sentiment analysis to be completed). This might contribute to the simplicity of using the site, but it does provide a significant disadvantage as your sources of data is limited (Facebook for example might have a more diverse range of sources) and you are not able to compare the current sentiment analysis of your subject to a previous time.
Finally, the most crucial constraint of Sentiment140 is the difficulties it has with deciphering whether a comment is actually positive or negative. The comments shown in Figure 3 provide a perfect example of this: the tweet at the top of the page expresses the belief that “Julia Gillard is ruining our country,” yet it is classified as a neutral comment about Gillard. As can be seen in the research paper written by Councill, McDonald and Velikovich (2010), sentiment analysis is always improving in this regard. Nonetheless, this kind of misclassification affects the results shown in the graphs and thus undermines the validity of the data. This is why many (e.g. refer to Beale’s article linked above) are skeptical about sentiment analysis, and rightly so.
Often incorrect sentiment classification is a result of unambiguous negation patterns in text – it is not always as easy as ‘I like this’ versus ‘I don’t like this’. Councill, McDonald and Velikovich (2010) put forward five categories which are known to trouble analysis tools: denials (e.g. “there is no doubt it is excellent”); rejections (“e.g. I expected to be let down, this was not the case”); imperatives (e.g. “don’t refuse to try their brilliant food”); questions; and repetitions. Another common issue found with sentiment analysis tools is that they have problems dealing with sarcasm, as these kind of negative comments are usually categorized as positive (Rintel, 2012).
An example of how Sentiment140 could potentially add to a story
Aside from issues of data validity, sentiment analysis can be used to improve a news story. An example of an article which could be improved by the use of sentiment analysis is Craig Christopher’s (2012) article “Formula 1: Grosjean Ridiculously Suspended, Maldonado Slapped on the Wrist Again,” written in the Bleacher Report, an online sports news website. The article is highly critical of controversial F1 driver Pastor Maldonado (following the Venezuelan’s incident-filled race the day before) and Christopher makes reference to figures from a recent poll to support his claims against Maldonado. However, by using Sentiment140, the article could have included graphs indicating the most recent sentiments expressed on Twitter about Maldonado – this would not only improve the presentation of the article, but also ensure that the most recent data was used.
Table 1: Summary matrix table for Sentiment140.
|Selecting a topic|
|Media research (Ricketson 2004, pp.95-186; Spencer 2006, pp.25-122)||Can search for trends on a pre-arranged idea/subject.||Does not give you ideas for trends – have to be already thought of.|
|Newsworthiness (Ricketson 2004, pp95-186; Spencer 2006, pp25-122)||Not applicable.|
|Angle (Ricketson 2004, pp95-186; Spencer 2006, pp25-122)||Results of sentiment graphs and actual tweets can lead to an angle.||Results can be misleading (sentiments picked up incorrectly as positive or negative), thus affecting the credibility of an angle.|
|Defining topic (Ricketson 2004, pp95-186; Spencer 2006, pp25-122)||See above (for angle).|
|Choosing sources (Ricketson 2004, pp95-186; Spencer 2006, pp25-122)||Aside from Twitter sources, it does not provide any information for sources.|
|Facts and figures (Ricketson 2004, pp95-186; Spencer 2006, pp25-122)||Provides graphs which can be used as part of a story.|
|Interviews (Ricketson 2004, pp95-186; Spencer 2006, pp25-122)||Not applicable.|
|Anecdotes (Ricketson 2004, pp95-186; Spencer 2006, pp25-122)||Tweets could possibly be used for anecdotal purposes.|
|Documents (Ricketson 2004, pp95-186; Spencer 2006, pp25-122)||Not applicable.|
|Photographs (Ricketson 2004, pp95-186; Spencer 2006, pp25-122)||Not applicable.|
|Checking credibility (Ricketson 2004, pp95-186; Spencer 2006, pp25-122)||User can verify whether they think comments are actually positive or negative.||All sources are merely Twitter users and are hence not necessarily credible.|
|Selecting most important data (Ricketson 2004, pp95-186; Spencer 2006, pp25-122)||Data used in graphs cannot be determined by user.|
|Writing the article|
|Pyramid structure (Ricketson 2004, pp95-186; Spencer 2006, pp25-122)||Not applicable.|
|Flow/clearness (Ricketson 2004, pp95-186; Spencer 2006, pp25-122)||Not applicable.|
|Designing layout (Ricketson 2004, pp95-186; Spencer 2006, pp25-122)||Useful as the graphs provided can add an extra element to the story.|
|Audience reach||Not applicable.|
1) Annenberg Innovation Lab. (2011). Live Twitter Sentiment Analysis. Retrieved August 29, 2012, from http://www.youtube.com/watch?v=nWvpY27C3Ws
2) Beal, A. (2009). Is Automated Sentiment Analysis Reliable? Retrieved September 3, 2012, from http://www.marketingpilgrim.com/2009/08/why-sentiment-analysis-is-about-as-reliable-as-a-canary-in-a-coal-mine.html
3) Christopher, C. (2012). Formula 1: Grosjean Ridiculously Suspended, Maldonado Slapped on the Wrist Again. Retrieved September 3, 2012, from http://bleacherreport.com/articles/1320684-formula-1-grosjean-ridiculously-suspended-maldonado-slapped-on-the-wrist-again
4) Councill, I. G., McDonald, R. & Velikovich, L. (2010). What’s Great and What’s Not: Learning to Classify the Scope of Negation for Improved Sentiment Analysis. Proceedings of the Workshop on Negation and Speculation in Natural Language Processing.
5) Hutchby, I. (2001). Technologies, Texts and Affordances. Sociology, 35(2), 441-456.
6) Jenkins, C. (2009). How Sentiment Analysis works. Retrieved September 3, 2012, from http://www.slideshare.net/mcjenkins/how-sentiment-analysis-works
7) Rintel, S. (2012). Lecture 04: Affordances and Constraints 2, A Framework for Resource Reports [Powerpoint slides]. Retrieved from JOUR2722, University of Queensland Blackboard Online http://www.elearning.uq.edu.au
8) Ricketson, M. (2004). Writing Feature Stories: How to Research and Write Newspaper and Magazine Articles. Crows Nest: Allen &Unwin.
9) Sentiment140. (n.d.). Sentiment 140 – General Information. Retrieved August 29, 2012, from http://help.sentiment140.com/home
10) Spencer, L. M. (2006). News Writing: The Gathering, Handling and Writing of News Stories. Boston: D.C. Heath & Co.
11) Tumasjan, A., Sprenger, T. O., Sandner, P. G. & Welpe, I. M. (2010). Predicting Elections with Twitter: What 140 Characters Reveal about Political Sentiment. Proceedings of the Fourth International AAAI (Association for the Advancement of Artificial Intelligence) Conference on Weblogs and Social Media.
ALP hit back in the polls – but can Gillard overcome approval crisis?
Recently, the noise in Australian politics has been that Julia Gillard and the Federal Labor government is back on the way up.
After dropping to a near record low in the polls in July this year, the latest Newspoll figures show that Labor have bounced back at the LNP, jumping five percentage points to 33 per cent on primary vote. The Gillard government has also narrowed the gap on the two-party preferred vote, although they remain behind 46 to 54 per cent.
Ms Gillard has faced an uphill battle for popularity and approval since deposing former Prime Minister Kevin Rudd in June 2010. The polls indicate she is slightly behind Opposition Leader Tony Abbott in the eyes of the public as the preferred Prime Minister – 36 to 38 per cent.
For the members of the public who are passionate about politics or simply concerned by a topic raised by Ms Gillard or Mr Abbott, social media sites often provide an appropriate space to unleash their opinion. It is therefore useful to consider how these high profile politicians are faring in terms of online popularity.
Using a Twitter sentiment tool to pick out the positive and negative tweets about a subject, it could be found that Mr Abbott fared respectably, with more than half of the tweets about him being positive.
Meanwhile, doubts could be raised about Ms Gillard’s comeback in the polls after referring to the sentiments being expressed on Twitter.
Intriguingly, the man who it seems can never be fully erased from discussion about the top job – Mr Rudd – had a considerably higher percentage of positive tweets than the leaders of both the ALP and LNP.
This form of analysis provides an interesting insight into how members of the public view the nation’s politicians. In addition, the sentiment analysis device allowed the opportunity to see what positive and negative comments had been made.
By reading the comments, it could be seen that the most severe criticisms were dished out to Ms Gillard, confirming the graph results. Among the comments made, one person wrote “Julia Gillard is ruining our country” while another cleverly made their point: “Ah Julia Gillard. A bit of a Vegemite figure. Half the country hates her, & the other half think she belongs on the end of a knife.”
It seems Ms Gillard has plenty more catching up to do before the ALP can be considered a genuine chance of winning the nation’s next election.