Next to listening, understanding is the most important next step when it comes to knowing how your audience is talking about you.
With thousands of tweets, blogs and forums talking about a brand, reading every article can be a chore few would attempt to do. Hence, the need for a concise and summarized view of the social landscape.
While text analysis has been around for many years, it becomes a lot more challenging with social media. Online conversations are informally written, there are too many grammatical and spelling errors, and there is far too much data.
Over the last few years, our team has developed several key algorithms for machines to make sense of the data: sentiment analysis, language translation, short document summaries, keyword word clouds, visual buzzgraphs, popular phrases, semantic analysis with entities, and key conversations. In this post, we will cover the last five features, while leaving the rest for the future.
We’re excited about the latest addition, a new version of Word Cloud. It displays the most commonly used words in the search results in a user-friendly manner. Associated CSV export capability includes more details and occurrence frequency for each of the word. Five colour themes and three font themes are available to make the resulting graphic look beautiful and appear in your presentations and report naturally.
The word cloud below shows most frequent keywords surrounding the new BlackBerry PlayBook.
A staple within MAP and Heartbeat has been the BuzzGraph, which displays the leading keywords and, as important, the relationships between them. Different words are connected by dashed, thin or bold lines to show weak or strong correlation between them. Obviously, you can drill down in to any keyword to see a synopsis. And if you love data and numbers, CSV exports will make you happy.
Like the Word Cloud, the BuzzGraph delivers a user-friendly snapshot of lots of data. In this case, we’re displaying the leading keywords from more than 100,000 mentions of the recently launched Blackberry Playbook. It shows many related topics but RIM, tablet and blackberry form the heart of the conversation with the bold connections between them.
Moving beyond just words and relationship between them, MAP also displays a list of popular phrases found in the results. Again, you can drill down by clicking any one to refine your searches further.
Our Entities feature goes beyond words and phrases to analyze the text semantically. By using natural language processing, conceptual entities are extracted and categorized as being a person, name, city, company, etc. You are no longer limited to the entire universe of words and phrases, but more meaningful entities categorized automatically by the system.
Below are the extracted entities from about the newest Playbook tablet and a zoomed in version of just Industry Terms. Here, the text is analyzed semantically to categorize “New York Times” as one entity meaning a publication, whereas “New York” is another under cities.
Last but not least, Heartbeat and MAP include Key Conversations, which shows complete meaningful sentences that summarize all the results based on analysis of the key discussion themes and topics. This makes it a very powerful but concise tool to quickly see exactly what is being talked about by reading just a few sentences.
Below is a graph showing the key conversations. Notice, how we have moved from simple word lists to full meaningful sentences summarizing the content – and how good these sentences are in showing a very succinct snapshot of all that is being talked about the new tablet.
Text analytics is one area Sysomos has focused on from the beginning, never separating it from just monitoring. Most of what you see here is a result of our in-house research and development team, consisting of PhDs and experts in data mining, specifically tuning the algorithms for social media content.
Having in-house technology not only means our continued ability to innovate and adapt to our customer needs, but also no extra fee to our users for using any of these features.
As always, we are always interested in hearing your feedback and suggestions on making these useful features even more awesome.