MoodViews — FAQ

Some frequently (and some not so frequently) asked questions about MoodViews.

• Did you only consider blogs in English in your evalution of the blogosphere? If not, wouldn't it be interesting to compare the reactions of bloggers from different countries to the same event?
• Have you gotten any reactions from bloggers to your study? What were they like?
• How many bloggers have downloaded the "moodsticker" so far?
• Aren't your findings rather trivial? How could they be of use to marketing or other companies?
• You say that Moodstickers "track the most popular mood". What do you mean by "most popular" mood?
• Would you consider doing something similiar to MoodViews with the "music" tag found in LiveJournaal posts instead of the "mood" tag?
• Do you intend to open-source the MoodViews applications?
• What are those "flat" areas that you occasionally see in Moodgrapher's plots? Or those "empty intervals" in Moodstickers or Moodteller?
• Are you surprised by some of the results you obtained?
• Who is using your tools today (besides you, in your institute)?
• What are the emotions you can recognize?
• In blogs, do you search for specific words you have associated with emotions?
• Do you think your MoodViews tools could be integrated in man-machine-interaction?

• Currently, we are only analyzing blogs in English, for technical reasons. Our technology is language-independent and could work, in principle, for any language; in fact, we are currently collaborating with a large Dutch blog platform on a number of fronts, one of which is aiming at a "MoodViews-like" application for Dutch blogs. In that case, it will be interesting to compare reactions from different countries to the same event, as well as compare the events that trigger reactions in different countries (e.g., EU constitution vs. Halloween).

• There are hundreds of blog posts talking about MoodViews, and we received many direct reactions. The vast majority of reactions are very positive, appreciating the project and its outcomes. A small number of discussions on the blogspace expressed concern about a "big-brother" scenario where emotions of people are detected and aggregated for wrongful purposes—this is, of course, not our aim. Some of the reactions—from bloggers, but also from journalists and from various other professionals—have expressed wishes to share the data or collaborate on some level, both for research and commercial purposes. Overall, MoodViews appears as interesting and fun for many people.

• Moodstickers are used on more than 300 blogs (and other sites) currently. The MoodViews site itself has been visited by over 150000 unique visitors since it was launched; on busy days, MoodViews is viewed more than 50000 times a day.

• First, you need to distinguish between the three components we currently use: Moodgrapher, Moodteller, and Moodsignals. Moodteller predicts the moods of bloggers by reading the text they write; using sophisticated text-analysis methods it achieves very high precision, and is in no way trivial (a lot of previous work in this field of detecting emotions from text—Sentiment Analysis—shows that this is a difficult task).
  Moodsignals uses other text analysis techniques to match mood changes with ongoing events—again, a non-trivial task; you as a human may know that an increase in global sympathy in September 2005 is related to Hurricane Katrina, but for a computer it is not obvious—it has no understanding of what sympathy is, or what a hurricane causes.
  Regarding Moodgrapher, I agree that the idea itself is straightforward: just count, and graph, the different mood reports per hour; the innovation here is simply coming up with the idea and implementing it effectively. Having said this, there are still interesting findings: maybe the 'loved' reaction to Valentine's Day is trivial, but is it trivial that U.S. bloggers are more sympathetic towards the London Bombing victims than towards the U.S. victims of Katrina?
  About commercial potential: it has been shown that positive references to products are indicative of good financial results for these products. Successful companies such as BuzzMetrics, Umbria Communications and Cymfony base their entire business model on this. We are not marketeers, but certainly like to see what emotions people express towards a "topic"—MoodViews—in their blog posts, and MoodViews itself allows us to do that.

• We mean, "the mood that is reported by the largest amount of LiveJournal users at a given time".

• Regarding "MusicViews"—we actually did some initial investigations in that area (or rather, students of ours did); however, it proved much more chaotic than the mood-related work. One definite conclusion reached was that LiveJournal users rock: they listen to rock music much more than rock music is ranked by weekly music charts (genre identification was done through FreeDB).

• Currently we don't have plans to release MoodViews externally. A reasonably thorough description of the internals is available from the papers we wrote about it. Alternatively, if you're interested in consulting, we do offer that to non-academic organizations. Mail us at moodviews@science.uva.nl.

• Moodgrapher and the other MoodViews tools use data supplied by LiveJournal. Sometimes we are not able to get to LiveJournal's data feed, either because of downtime at their end, network issues, or downtime at our end. Periods for which we are unable to obtain data result in "flat" or nearly "flat" areas in the plots produced by Moodgrapher, and in "empty intervals" in the curves displayed by Moodstickers and Moodteller.

• First, we were surprised by the high accuracy of the system: large scale events such as disasters, terror attacks and so on are discovered immediately; within a few minutes users can view the response of bloggers worldwide (see our global events page). But even after months of operation, we continue to be surprised by small discoveries MoodViews makes: for example, that people are more excited about a new Harry Potter book than about Christmas; or that, while every weekend the level of drunkness increases, in the few weekends after New Year's Day the drinking habits were somewhat restrained (maybe due to new year resolutions, or to excessive hangovers from the new year weekend).

• MoodViews has a steady user base of a few hundred people every day; since it was launched in the summer of 2005 it was accessed by more than 100,000 unique visitors (typically, there are big bursts after an important discovery MoodViews makes). During its operation, we have been contacted by economy specialists, journalists, bankers, psychologists and, of course, affective computing and other computer scientists; all of these found interesting angles in which this data can be used in their expertise area.

• Theoretically, we work on all moods supported by LiveJournal (a detailed hierarchy is here). In practice, we sometimes limit ourselves to the most common 30-40 moods to increase system accuracy.

• This is different in the various applications within MoodViews. Moodgrapher does not look at text at all; it relies on the fact the some bloggers explicitly say what their mood is, and uses these labels for its graphs. Moodteller looks at a small set of words, and uses them to estimate the moods (details are in the publication above); the words have been discovered by Moodteller itself, rather than fed to it by a human. Moodsignals first identifies an abnormality, then finds out all words associated with it; there is no hard limit, but usually a small set of words is sufficient.

• MoodViews is not an affective computing application in the traditional definition, i.e. a system that responds to the user's emotions. However, it works the other way around: it enables the user to see a computational entity—the blogosphere—as a more human one, with emotions and moods. In that sense, it does increase affective interaction between the user and computers. In the future, we intend to implement additional tools which will increase the amount of interaction MoodViews offers to users, such as abilities to search for moods associated with certain words, people, and places (Moodspotter) and natural language explanations of substantial changes in mood patterns (Moodsignals). Finally, affective computing is usually described as an interface between a single person and a computer—but what happens when the computer wants to respond to the mood of not one person, but a large population? MoodViews enables this by letting a computer "know" what emotions are strong and weak at every point in a large population.

• MoodViews > 
• FAQ >