NOTE: If you find this blog post useful please bookmark it. I am planning to include here the latest info on Twitter tools.
For some time now, medicine I am doing my best to participate in a Twitter chat dedicated to social media measurement, generic #smmeasure, pharmacist as a means of exploring what practitioners do and of discovering new tools. Last Thursday (November 4) the discussion who brought together marketing/communications practitioners, bloggers and representatives of social media measurement companies have focused on influence and influencers – identifying them, measuring their influence/reach, engaging them. So besides the need of defining these notion (which Andreana Drencheva, or @addy_dren attempted to do in her post) a necessity to reveal and discuss tools used for research emerged. In this post I am going to list the Twitter tools I used so far and group them into measurement objectives/categories.
TwentyFeet is a rather new platform dedicated to aggregating statistics related to Twitter and Facebook accounts. Once authorized, the platform collects and produces visualizations based on activity metrics: followers, lists, mentions, RTs, favorites, links and more. What’s interesting is the platform’s interpretation of data into reputation (followers and lists), influence (mentions, RTs), conversation indicators (Tweets, RTs, @replies). While I believe it is true that the number of followers could indicate an increase in reputation as well as an increase in the number of mentions could point out to an increase in an account’s influence, without any context to the data such assumptions are very difficult to prove. Take for instance the existence of automated following programs or the involvement of people in weekly chats. In these cases the number of followers of one account can be artificially increased. Similarly, with the participation in a weekly chat there is a higher potential for an increase in mentions, however this does not necessarily equal with an increase in influence.
TwitterAnalyzer provides a variety of activity evolution metrics including user activity (number of tweets, chats, subjects, hashtags, links) and follower metrics (online followers, growth rate, density map, RTing accounts) and user interaction (mentions). This last category is a little bit unclear as it provides statistics for both mention updates (as made in public tweets) and social mentions (as made in chats). If you consider that most tweets and chats are public, than this differentiation doesn’t mean much unless it makes reference to automates posts (as in generated for instance by paper.li and other automated updates) versus @ mentions. One advantage of TwitterAnalyzer is that integrates qualitative data (as in tweet excerpts) in the same window giving thus context to the visualized quantitative data. The platform also provides metrics for reach (daily exposure rate) and popularity (could be understood as influence; also shown on a daily basis) but is is unclear what algorithm is used to calculate them.
TweetEffect visualizes the fluctuations of followers for designated account. It uses data from the last two weeks of a given account and visualizes it in two ways – first as a simple graph, then a table that combines qualitative and quantitative data. The table shows an account’s activity tweet by tweet and highlights its presumed influence on gaining or losing followers. However, the platform doesn’t yet take into account that the loss of followers could be influenced by the account holder’s decision to block access of other users.
TweetStats visualized Twitter usage patterns for designated accounts including graphs for tweeting timeline, daily tweeting density, aggregate daily and hourly tweets, platforms used, mentions made by the account and RTs made by the account. This is a good tool for behavioral and usage patterns observations and can, if used consistently, highlight technology adoption and working patterns. Used in combination with TwitterAnalyzer that shows the subject about which an account tweets, TweetStats could show when people tweet and help thus communicators find out both the what and the when that makes an influencer tick.
Xefer is an alternative to TweetStats. Their graphs confirm each other meaning that they most probably use similar algorithms and similar data.
TwitterCounter track the activity of an account showing the fluctuations in numbers of followers, followed accounts and tweets. Premium services provide metrics for mentions and RTs, but with the existence of platforms like TwitterAnalyzer and Twitter’s own metrics one could wonder whether they are indeed providing a unique service.
TwitterGrader is measuring the power, reach and authority of a twitter account. Its algorithm includes a number of factors such as number of followers, power of followers, updates, update recency, follower/followers ratio and engagement. According to this post, the score is calculated “based on the factors mentioned and then used to compare a user against all other users that also have a score. The grade is calculated as the approximate percentage of other users that have an equal or lower score. So, a Twitter Grade of 80 means that about 80% of the other users got a lower score. At the time this article is being written, over 2.1 million users have been graded”. Using a ranking application can be useful when aiming to find out where accounts that one follows or is followed by exist within the Twitter sphere. However, without context – and by context here I mean observing the tweeting behavior and topics covered by one account – rankings are just numbers.
TweetGrade grades an account’s activity based on a quantitative assessment of the account’s “reach and influence in the Twitter community. Based on the account’s interactions with others, the frequency and content of updates, and the account’s overall contributions to Twitter” the platform assigns a simple letter grade that ranges from an ‘F’ to an ‘A+. This could be a good platform for beginners in search for a bit of Twitter fun.
RetweetRank provides a score for the number of times a user has been retweeted by others recently (though it doesn’t say how recent) and it doesn’t say to whom the percentile is reported to. For my score (98.67% on November 6) I assume that the interpretation of the result would indicate than I am RTed more than 98% of the other assessed Twitter users… Promising but hard to believe. Such metrics depend on a wide range of variables including size of network, relevance of message and even time of tweeting.
Twitaholic displays a number that is supposed to be a user’s rank within Twitter. According to the platform, they scan twitter a few times a day to determine who’s the biggest twit but they do not say how do they calculate this: whether is a ratio of followers/followed or a ratio between followers/followed/tweets. Moreover, while the number given might be encouraging or quite the opposite, the rank is calculated in relation to entire Twitter user base, making the results quite useless if looking to rank as a measure of popularity/visibility. This could be much better if geo-location information would be added.
Tweerank presents itself as an “engagement score” which is assigned to any Twitter user, which has to do more with conversation and reach in my view. According to the platform the score takes into account a multitude of parameters includingthe number of followers, the number of followed accounts, the number of tweets, RTs and mentions. Since there no information to what dataset the rank is reported, I assume it calculates the percentile looking at the entire Twitter-sphere.
Ranks are also calculated by CrowdEye but there is little detail on how that “cool math” really works.
TweetReach provides a comperehensive set of metrics about a search term or user impact on Twitter and complements the quantitative data with excerpts of qualitative inputs. An explanation of how the scores are calculated is available here. Calculating reach is useful when aiming to assess the number of exposures a message/account could gain as facilitated by its network.
As mentioned earlier, TwitterAnalyzer provides reach metrics as well but offers less details about how it obtains its data for the visualizations it produces.
TwitAlyzer provides a score for impact and an assessment for the type of influencer the scored account would be. The impact score is the result of a number of factors (number of followers, unique references and citations, unique RTs, tweeting frequency) but the formula is not disclosed. The score displayed as percentile is reported to the total of Twitter accounts the platform is tracking. This posits me in the superior 77+%, meaning that my impact score is higher than 77% percent of the 0 active Twitter accounts they are tracking. The platform also provides a set of definitions for the terms they use, a welcomed feature when attempting to understand the terminology. The promising feature of TwitAlyzer is it’s geo-location. I am still unsure how they integrate their impact score with the city/area of the Twitter account but should this be possible, those statistics would help researchers and communicators find relevant tweeple.
Twinfluence “is a simple tool for measuring the combined influence of twitterers and their followers, with a few social network statistics thrown in as bonus”. I have to applaud the platform for having the most detailed information about how and where the data is retrieved, the algorithm used and the definitions employed. Twinfluence goes beyond calculating a rank of a given account, by providing metrics about how tight its network is (centrality), the accumulated influence (velocity) and the social capital of an account (the average first-order network). These metrics are highly important as one’s network could be tightly or loosely knit making for instance the information shared to spread at different speeds. Twinfluence could be paired up with FlowingData’s MentionMap that visualizes the network of a given twitter account.
Klout scores range from 0-100 and measure the overall online influence of a Twitter account. By overall I assume they mean twitter overall rather than internet overall. According to the platform, the score is “a factor of over 35 variables broken into three categories: True Reach, Amplification Score and Network Score” where “True Reach is the size of the account’s engaged audience and is based on the followers and friends who actively listen and react to your messages. Amplification Score is the likelihood that an account’s messages will generate actions (retweets, @messages, likes and comments) and is on a scale of 1 to 100. Network score indicates how influential an account’s engaged audience is, also on a scale of 1 to 100. The Klout score is highly correlated to clicks, comments and retweets.” Perhaps the most useful of Klout’s features is its influence matrix interpretation of the score that makes it appealing even for social media beginners.
Trendistic enables comparative searches of terms within Twitter. It is great for contextual analysis since the visual data (that can span from 24 hours to 30 days) is supported by the inclusion of qualitative data (the tweets that were used to generate the visuals). It is also a great tool for analyzing words in context as well as observing emergence of trends and patterns of communication. Correlated with data from outside of Twitter, such as media reports or personal data Trendistic could help summarize the communication ups and downs of an event as translated by the attention/mentions received.
TweetVolume enables comparisons only between 3 search terms. Since last time I visited the application, considerable improvements were made including clarifying where the information is gathered and how the graphs are generated (calculated): “the TweetVolume results are an estimated count of the number of times a word or phrase has appeared in a Twitter post. We generate the estimates by utilizing resources from Google’s search engine and Twitter’s open API to capture and compare data points.” The platform now also displays tweets included in the graph generation.
The alternative to TweetVolume is Monitter that displays tweets related to 3 search terms. While the platform doesn’t visualize the data nor calculates volumes, it does however filter it based on geo-location data.
Twopular not only tracks emerging trends but enables one to compare up to 10 of them. This is useful when comparing alternative hashtags as well as tracking the variations in volume of mentions of words, hashtags, brands in time.
Twitscoop that also generates comparative graphs based on volume of mentions of notions and hashtags. Like Trendistic it includes the tweets as well.
Twittratr‘s screen includes a sentiment meter and 3 columns dedicated to positive, negative and neutral sentiment related to a searched term. The sentiment is calculated using an existing list of negative and positive words. This could be good but is not ideal as it lacks context. Some words for instance are negative or positive by default (take sadness and happiness for instance) but they can change meaning depending on the words surrounding them (as in abolish racism or promote racism). For this automated list to work it would not only need constant updating but also an addition of combinations of words that can be positive or negative. While the number of total tweets is reported making it easy to understand the sample, it is still unclear what the time span of that sample is. The displayed tweets do include links to Twitter but this complicates rather than eases the work. A similar modus operandi is displayed by TweetFeel and TweetSentiments with the latter being more complex. To its display of negative, positive and neutral sentiment percentages and color coding the tweets, TweetSentiments also includes an assessment of the education, flamboyance, slang, gender and age which could be particularly relevant when assessing accounts rather than events.
CrowdEye mentioned earlier for its ranking metrics, is a promising sentiment analysis platform. It provides volume of searches for as much as 14 days, sentiments split into positive and negative boost and geo-location filtering options. To these the tweets displayed include the CrowdEye rank of the user enabling one to potentially assess the impact and reach of the tweet. Reading and interpreting the data would be facilitated if platform adopted the color coding system of Twitrattr and TwitterSentiments.
Other Twitter sentiment analysis platforms worth noting are Twendz and Chatterscope, both developed by Public Relations agencies, Waggener Edstrom and Lewis PR respectively. Both platforms (the free versions) provide similar data but visualize it differently. Moreover, Twendz continues to update its content unlike Chatterscope that displays its metrics based on the last 1000 tweets found. Additionally, Twendz also includes subtopics and a wordcloud enabling further exploration of related content.
TwapperKeeper allows one to create an archive for a hashtag, keyword or person making access to Twitter data even after the search API expired easy. If set up in advance the archive can be an invaluable source of information that will help track evolution, sentiment, volume, behavior, network connections after the event is over. Archives will also permit comparisons, something desirable when attempting to look at research precedents or to investigate previous practices. An alternative for TwapperKeeper is the Twitter Tag Downloader, but unlike TwapperKeeper it does have a limit 5000 tweets.
The Archivist also archives twitter data but it has the added value of providing some visual interpretation by default. This makes the analysis process even easier than with TwapperKeeper without taking out the flexibility of investigating the data outside the given patterns since it has options for viewing and downloading the data. Moreover, the Archivist has also a desktop application which would enable working with other datasets as well.
Other ways of archiving Twitter include back-up options reviewed by ReadWriteWeb. Of some relevance might be Twistory and TwInbox, both downloading data to calendars and Outlook Express respectively. Personally, I see these two options as good for simple back-up but difficult for more in-depth data analysis.
A final word
All the tools I have mentioned here are free or operate on a fremium model. Before using them it is highly recommended that you have some objectives that you seek to evaluate. Then, in assessing those objective do use more than one tool in order to verify results.
What tools do you know? What tools have you used?