Oliver Conner
The implications of social media based research are set to be huge, and we are still very much in the early days. Still, it is important to cut through the hype and understand exactly what social media listening is and how it can be of benefit to researchers.
The social media listener is a new breed of researcher. They are as nimble with a large dataset as a quantitative analyst but possess a qualitative researcher’s keen eye for a good verbatim. Accompanying this amalgamation of quant and qual, is an addiction to social media and a firm knowledge of the trends that shape our emerging digital landscape.
Below, there are ten questions to ask yourself before embarking on a social media listening project:
What is social media listening?
It is important to have a working definition in order to best understand how you can use it. It is also important to understand what it is not so that you can avoid applying social media listening to the wrong problems.
Social media listening is the process of finding relevant conversations from blogs, forums and social networking sites (such as Twitter and Facebook) using search queries and keywords. These conversations are analysed to provide social intelligence and insight into the researched audience.
It is not (yet) a replacement for survey based research or focus groups. Instead it is a new method of research that removes interviewer bias, gives you access to millions of people that might never have been a part of other research projects and gives you datasets that run into the millions.
Should you be investing in social media listening?
The answer to this question will normally be ‘yes’. However, by asking yourself this question you are able to assess the business value of starting a social media listening program and determine to what extent you should invest.
You should ask:
- Is there enough of a conversation going on about my brand/industry/category to warrant social media listening?
- Will there be a measurable return on my investment?
- What sites/networks are the conversations I’m interested in, taking place?
- What is the general tone of the conversation that is being held?
These questions should help you decide how much time and money you invest in a social media program.
What are your goals?
Although a certain degree of flexibility should be factored into your social media listening method, creating a solid objective with clear goals is essential. Some of the common uses for social media listening are as follows:
- Brand monitoring: Tracking mentions of your brand is probably the first and most basic application of this form of research.
- Competitor tracking: Listing and searching for the conversations around your competition can help you determine what it is they lack and what it is that other people like.
- Searching for sales leads: Including a ‘?’ in a search query will pull up questions, this will help you understand the types of queries that your customers are asking and provide you with an opportunity to engage.
- Category research: Social media listening is powerful when exploring categories. By starting with a broad search term you can begin to drill down as themes and topics surrounding that term emerge.
- Influencer research: Tools like Klout can analyse network activity and indicate who is influential and what they are influential about.
Have you learnt the lingo?
The digital vocabulary seems to increase every day and can be very daunting. However, knowing your Klout from your Quora is an essential part of being an effective social media listener.
For example, when one single tweet is analysed there are dozens of data points that are made available. ‘Followers’ are the number of people that follow this person, ‘lists’ refer to the Twitter lists they are on, ‘retweets’ are how many times that tweet was shared etc.
The best way to familiarise yourself with the terminology is to join the different platforms and become immersed in the cultures. There are also an abundance of glossaries available on the net.
How do you choose your tools?
There are hundreds of social media listening tools available, ranging from the ludicrously expensive to the completely free. Choosing the right platform involves listing all of the social networks that you will need to extract data from, the amount of data you expect you will need and the types of analysis you want the platform to perform automatically.
For many research projects it is not necessary to invest in a platform and it makes more sense to explore the possibilities in the free tools that are available. Using a feature rich dashboard like Hootsuite is a great way to begin exploring social media data and it offers a wide range of both paid for and free ’apps’.
When choosing a paid for service it is important to keep your objectives in mind. Many social media tools will offer a suite of engagement tools – but if engagement isn’t part of your objective then paying for this will be a waste of money.
Make sure that you consider what kind of data enrichment services you will need. Do you require text analysis? Do you want your dataset to tag influential people? Do you want to visualise the major trends? Do you want automated reporting?
Personally, I like working with the raw data – so make sure that your provider offers you the data in a format that you are able to work with.
What is going in your search query
If you are a researcher then it is highly likely that you have written a few questions. When it comes to building a search query, you should apply the same level of attention and care as when you craft a question for a survey or focus group.
A search query can lead to biased results in much the same way that a leading question can in a survey. For example, if we were researching a computer game, a search for ‘Call of Duty’ would return results that were very different from a search for ‘COD’ (which is how fans often refer to the game).
Building your search query will involve covering all possible abbreviations, slang or common spelling mistakes that are applicable as well as any name variations. By the time you’ve considered all of these angles your search query will probably be pretty large.
One of the major problems you will then face is the vast amount of unnecessary data you will pick up. For example, if you were conducting research on the company Apple, your search will return lots of fruit related mentions.
This can be combatted by excluding particular terms from your search. Alternatively, some data providers offer a disambiguation service that will analyse the text in the data and determine which topic each case relates to.
Finally, make sure that you are familiar with all the search operators that your data provider caters for. For example, can you filter by platform, location, demographics etc.
How will you perform content analysis?
So you have a spread sheet with a sample of a few thousand mentions of your search term – the next step is to analyse it. The starting point is usually to create a word-cloud to quickly visualise the most commonly mentioned terms. There are plenty of free word cloud generators on the web, the most popular being Wordle (or, check out Tagxedo if you want to give your clouds a bit more style).
Other methods of visualising text include word trees, tag clouds and phrase nets – all of which are available on IBM’s Many Eyes website.
Word visualisations can only really go so far, and to really get the most from your data you will want to begin using a text analytics tool. Text analytics is a massive field of study and is the engine of social media research. It uses complex algorithms based on natural language processing to categorise content.
A good text analysis tool will categorise the content and provide you with a powerful pivot to navigate your way around your data.
What is sentiment analysis?
Sentiment analysis is determining how positive/neutral/negative a statement is. Most providers of social media data offer an automated sentiment analysis tool. However, due to the slippery nature of human language, automated sentiment analysis is incredibly unreliable.
If two humans were coding sentiment then it is unlikely that they would agree 100% of the time and will more likely agree on a sentiment around 85% of the time. Automated solutions vary but it is unlikely that you will get an analysis that will help you with your research.
There is a place for automated sentiment analysis and that is where you are monitoring a search query with millions of mentions every week. The reliability of automated sentiment tools will suffice to highlight dips or peaks in sentiment that would indicate something worth investigating.
If accurate sentiment coding is essential, then a great alternative is to use a crowdsourcing tool like Amazons Mechanical Turk. This allows you to pay a small amount to get a verified human to code each verbatim.
Will you need sampling and weighting?
When conducting a survey, it is usually impossible to field it to 100% of the people in your population. Similarly in social media research – it is unlikely that you will be able to acquire and analyse every mention of your search term.
When sampling with social media research, it is important to ensure that you are not pulling data from a limited range of social networks and websites. You must ensure that you are extracting data from YouTube comments, question and answer sites, forums, product review sites and all other platforms to ensure that you are creating a fair and representative sample.
After sampling, it is important to weight your data. Weighting is the process of ensuring reliability and validity of data by given more or less prominence to subgroups of people. This applies to social media in a similar way. For example, if 80% of your data is from Twitter, but only 20% of your customers are using Twitter then you should weight accordingly.
Do you understand the ethics?
Just like you wouldn’t publish all the names and addresses of people that answered a survey in your report, it is equally important to safeguard people’s anonymity when you are conducting social media listening.
On a basic level, this could mean keeping people’s personally identifiable information private. Furthermore, if you are quoting somebody then it might be a good idea to alter some of the wording a little bit so that the verbatim cannot be placed into Google and the identity of the author found.
Of course, a lot of the problems associated with privacy can be solved by simply asking the person if they can give their consent for their comment to be used. However, it is important to remember that a lot of people using social networks don’t really understand that their data is viewable by anyone and could be shocked or embarrassed by such a request.
Privacy and social media listening is a difficult issue and it is best to tread with caution, especially is you are tackling a sensitive area. For best practice, be sure to check out the ESOMAR Guideline on Social Media Research.
Oliver Conner is Head of Innovation at OnePoll
