When it comes to extracting content from social media feeds, the majority of social media monitoring or listening systems in the market are doing the same thing - using a set of keywords the user believes will lead the system to the content they want. The problem with this approach is two-fold. Not only does a single search lead to a lot of data noise, but by definition, a system using keywords the user provides will not find content using language the user is not familiar with.
That’s why at Eyesover we take a dual approach to our content gathering. We use two levels of search terms and we use artificial intelligence (AI). Here's why they work well together and what they do for you.
To filter the good stuff from the noise is a big job for one layer of keywords. We use an outer layer of search terms that give us a wide selection of content, and a second, more specific layer of search terms to find the content from the discussions about your areas of interest
Within that framework is where our AI operates. From just a few keywords for the topics you are interested in, the Eyesover system gets the idea of what you are looking for and then automatically finds and starts to use new keywords in order to find new content discussing the topics you want to know about. Once this process is started, the system constantly learns from what it previously learned.
How do we know this works? As one would expect, at the beginning, our system adds many new search terms every hour, the content they find filling in any gaps the initial search terms left. Many times this can more than double the amount of content found. As the system quickly matures, the keyword discovery tends to focus on uncovering the changing or emerging issues of interest to the user indicating the system has filled in the gaps in the ongoing conversations.
What does this mean for our customers? Ease of use and accuracy. You don't have to spend any time fine tuning as the AI takes over that task and objectively finds the more specialized or local issues you may be interested in.
It all leads to more and better data that can help your decision-making.
The strong lead Doug Jones held in the Alabama Senate race last week is all but gone as Roy Moore's support has rebounded driving Jones' lead down to 4 points.
As a result of Jones' lead last week, it appears that Moore supporters have become more engaged as there has been an increase in the number of individuals talking negatively about Jones online, while discussions about Moore have become more positive.
While 90% of last week's discussions pertaining to the race were about Moore, we now see a 60-40 split between Moore and Jones. The issue for Jones is his sentiment scores have fallen from an overall positive to an overall negative in a relatively short period of time.
Today's Eyesover Support Index indicates that Alabama Republican Senate candidate Roy Moore has been unable to stop the erosion of his support over the past week and now trails Democratic candidate Doug Jones by 12 points.
Over the past week there has been 10x more online discussion pertaining to Moore than Jones, but the overwhelming majority of those discussions have been marked with very negative sentiment towards Moore, the cause of his deteriorating support.
Jones has been in the background while Moore has been the focus of the media over the past week, but he will clearly be attracting more attention from all sides now that the race is competitive. With his online mentions increasing over the past two days, we'll be watching to see how it will affect his sentiment and support scores.
In our last post, we discussed a few of the challenges the polling industry is facing. Now let’s imagine a polling technique that never asked you anything. Nobody called or texted you. It didn't even ask you to fill out an online questionnaire. Instead, it just read the content you posted publicly online. Using your own opinions expressed on Twitter, Facebook, or whatever social platform you use, this 'pollster' deciphers what your leanings might be, using your own words, without asking you one question. This is exactly what Eyesover does.
Of course, there are challenges with this approach just as there are with traditional polling. For example, in the same way landlines can cause age range distribution issues, the issue can arise with social media data. The major difference is where the phone problem is only getting worse, social media usage is rapidly increasing across all age ranges and according to the Pew Research Center, there is now only a 15-16% drop in usage between the 18-29 age group and the 50-64 age group on Twitter and Facebook.
But there is still no question that there are many reasons why polling from online sources can and should be incorporated into public opinion tracking.
The most obvious reason is the massive amounts of data is available online. While a traditional poll might be produced from a sample of 500-2000 respondents, a poll derived from online sources can be based on hundreds of thousands, if not millions of respondents - a concept that would have seemed ludicrous to statisticians (and the firms that would have to pay for a poll that size!) not that long ago. This combined with the rapidly improving distribution of online users will tend to reduce the potential for sampling errors.
Another concerns that we often hear about online data is the potential for spam, bots, and fake accounts to skew the data. The Eyesover system eliminates the potential for such skewing by not only blocking bots, but we aggregate mentions by social media username. In other words, our data reflects individuals not simply mentions.
Does it work? You be the judge.
Eyesover’s forecast in the 2015 Canadian Federal Election was on a par with traditional polling and we also fared quite well in the 2016 US Presidential Election as we saw an under-reported base of Trump support throughout the campaign and reported the Electoral College was up for grabs heading into election night.
Our results show that the power of online, real-time polling has too much potential to ignore!
As we often post the results from our political Real-Time Polling on our site, we thought a couple of posts discussing the current challenges traditional pollsters face and how Eyesover can help solve some of those problems would be in order.
Traditional pollsters face challenges on multiple levels these days. It has always been no easy task to deliver accurate, up-to-date information on the opinions of the public, but in today’s changing society, it has become even more difficult.
Although there have been technological advancements in the industry with automated phone calling, which greatly reduced the cost of calling voters across the country, the reality is that 2016 marked the first year where more than 50% of households in the US did not own a landline. Of course, cell phones are increasingly being targeted by the industry, but reaching the owners of these devices poses a significant challenge as well due to the wide range of rapidly improving call-blocking apps available.
Even worse for polling companies, the households that still have landlines are increasingly skewing towards those in older age groups. More than seven in ten households under the age of 35 are now cellphone-only. These issues create significant and costly problems in creating a sample that accurately reflects the population, a necessity for accurate polling.
Considering these challenges, credit must be given to polling companies that, in-spite-of these adverse conditions, still manage to produce accurate and insightful polls!
But as the landscape of telecommunications changes, we believe polling derived from other sources of information will play a growing role in the industry and we’ll explain why in our next post.
Canada's governing Liberal Party continues to hold a wide lead over its opponents based on Eyesover's Real-Time Polling. The Liberals led by Prime Minister Justin Trudeau, are currently capturing 44% of voter support expressed online, compared to 29% for the Conservatives and their leader Andrew Scheer.
The New Democratic Party have been steadily gaining ground under new leader Jagmeet Singh over the past month and have reached the 20% level.
The Liberals and NDP have been trending in opposite directions with the government seeing a slow erosion of their support over the past month while the Conservative's support has been stable.
Eyesover's Trend Discovery Software allows enterprise and political organizations to use online data for issues management, polling and ad targeting. Our analysis of political parties consists of measuring party mentions and their related sentiment from Twitter, Facebook, YouTube and Reddit, and extrapolating that information into our Real-Time Polling.
Mention count, number of views, expected reach; these are just a few of the terms used by the social media marketers, but what does this information really tell us?
A major problem with online metrics is the potential for mention count to be manipulated by multiple mentions coming from both legitimate accounts and bots. Hundreds of comments, likes, or retweets from the same account will obviously inflate metrics, sometimes to the point where wrong decisions are made based on the flawed data.
Similarly, this inaccuracy can apply to the ‘number of views’ or expected reach. Often, neither metric will give you accurate information regarding the number of unique individuals who viewed your article or video if the numbers are inflated by individuals using different devices or web browsers.
The same holds true for inbound content. Knowing exactly how many individuals are talking about a particular subject can be far more valuable than just knowing how many mentions there are. Too many times organizations will be put on full alert and devote resources to an issue that is rapidly growing in mentions online, yet the reality is the mentions are coming from a handful of accounts.
One of the unique features of the Eyesover system is that we analyze our data on an individual by individual basis. This gives us the ability to report not only on the number of mentions, but also, the far more important metric of the number of individuals that are actually talking about the subject.
This method cuts through the spam and counts high volume accounts as exactly one individual, regardless of how many tweets they posted that day. By analyzing online content in this manner, we not only identify and discover real trends as they develop, we can use the individual opinions for key features such as our real-time polling and ad targeting while ensuring users are not distracted by noise.