The impact of social media on business is not a new concept, but how best to use it, is a field that is only now gaining some maturity. Companies recognise that social media offers a great deal of insight into its market in a number of ways. The problem facing businesses is how to make use of that data: how to identify it, sort it, and quantify it in a meaningful way that can help drive intelligent business decisions. In order to make sense of the mass of data generated on social networks, a new field is emerging called “social media analytics”. The field is developing institutional knowledge of its own, and developers are creating sophisticated tools to accomplish these ends.
This article will look at social media analytics: its challenges and solutions. It is an important topic – one you cannot afford to overlook – whether or not your business has an active web presence. Because with or without a website of your own, if you offer a service, own a store, or run any kind of business at all, people are talking about you online. Moreover, to do the best for your business, you should know what they are saying and what it means.
Social media analytics is the methodology for doing exact that: gathering data from social media websites, blogs, forums and online reviews, and analysing that data to help make informed, predictive business decisions. The process of accumulating this information is also known as “social media listening” or “opinion mining”, a type of natural language processing that is used to monitor how the public feels about a given product.
People are talking
Exposure is the first statistic marketing groups want to identify: how many people have seen their brand? How many times has the average person seen the brand name? How many know the name, and how many know something more about the product? With the advent of social media, there is a lot more untracked communication going on than there used to be, in which people discuss products and brands, and weigh in on quality, and often their impressions of the company itself.
While word of mouth has always been a desirable way to disseminate a brand name, now the sheer volume of this kind of communication being done online, makes it a critical factor to track. Social media, by allowing people to voice their opinions to a wide audience, has given the public an unprecedented ability to influence other people’s opinions and thus, their buying habits. People will trust what their friends have to say about a product more than they will trust the advertising or promotional efforts of a product’s maker.
Adding to the importance of social media’s influence is the trend-driven nature of online communication. It is in your company’s best interest to know what is being said online about you, because if five people are saying it today, fifty will say it tomorrow, and five hundred the day after.
Because of the trendiness of communication across social networks, it is important to monitor customer sentiment over a period. You are better able to map the popularity of your product, and whether it is gaining or losing in popular sentiment, is a good predictor of future sales. You can also learn what factors might affect a change in popularity, and respond to market forces before they have a big negative impact on your bottom line. If people think your product should perform a certain way, you can tap into that sentiment and act on it, responding in nearly real-time to consumer expectations.
Another important piece of information marketers wants to learn from social media is whether their media strategy is working. For instance, if your company needs to improve its public image, and has developed a strategy to do so, social network monitoring is one of the best ways to track the strategy’s success or lack thereof.
What does it all mean?
While the benefits of social media analytics are obvious, how to turn all that unstructured data into meaningful marketing data is much more difficult. Interpreting data from social media faces three main challenges:
- The huge quantity of data
- The fact that incoming data is unstructured
- Context matters, but is hard to quantify
Although gathering data is relatively easy, figuring out what a large collection of data means takes skill, intuition, and superior analytical skills. In the case of data gathered from social media, the first challenge facing the analyst is the amount of data collected. If your business is of any size at all, the amount of discussion online is enormous. For a human being to sort through the data is an impossible task. There is just too much of it. In order to put this data to work, it is necessary to employ software tools, which are the only feasible way to sort through quantities of data.
However, the data from social media sites is largely unstructured, coming as it does in the form of comments, tweets, social network posts, and so on. While marketers do employ some structured methods in the social media sphere, such as multiple-choice surveys, the results of surveys are a tiny percentage of the data that is available to be analysed.
Turning unstructured data into structured data requires resources to classify these different types of inputs. Even if the quantity of data were of a scale manageable by a human, the fact is that people are terrible classifiers. There is some subjective judgment employed in the classification, leading to the likelihood that two different people might classify the same thing differently. Even the same person might classify the same data differently on two different days. In this area, computer intelligence, with its lack of imagination and mood, is far better at cataloguing based on a set of rules than the human mind, and software tools are essential for consistent and meaningful classification.
The last challenge is one of context. It is not only what is said, but also in what context something is said that matters. If you have a negative comment, is it a negative comment that stands by itself, or is it part of a discussion that includes positive comments? Was it one of many comments in which your company appears at a disadvantage? What is the point of view of the author? Another important aspect of context is influence – some social networkers exert a disproportionate influence. Celebrities such as actors or sports figures have enormous influence because of their fame, but there are people who are famous because of their contributions in the social media sphere. In evaluating data from social media, the source is part of the context, and in some way, you must be able to assign greater weight to the comments of contributors who are highly influential.
Ultimately, social media analytics strives to put hard numbers to soft interactions. To quantify something is to measure it, and measurement is the heart and soul of marketing. But how do you quantify qualitative data? Making nuanced distinctions is easy for humans to do, but we are very slow, taking an enormous amount of time for each piece of data – enormous in computer terms, which is the crux of the problem with massive amounts of unstructured data.
Fortunately, the field of social media analytics is becoming more mature, and market-savvy developers are creating tools whose main purpose is to analyse unstructured data in order to identify market trends and customer sentiment. Enter text analytics, which is the method used by these software tools to categorise and quantify unstructured data.
Putting text analytics to work
Text analytics strives to derive quantitative meaning from the written word. A system for opinion mining collects and categorises sentiment about a product. Analysing unstructured data is tantamount to asking a computer to do not only what computers do well, but also what humans do well. The best systems use artificial intelligence (AI) to sort through data to classify it, allowing the software to “learn” as it goes, and to become more accurate with time.
In order to be able to sort through data, ontology must be built to tell the machine how to view the data it needs to sort. Ontology, in this context, is a framework of relationships and definitions within a subject that allows a machine to categorise data intelligently. By giving the computer a way to assign both context and meaning to data, the computer can then provide semantic tagging and categorisation of the information. These hooks allow software to see the content along with its context. Making use of these hooks, data can be categorised in a meaningful way, drawing an accurate picture of opinion at a given time, and identifying trends as they develop.
A number of analytical functions are used to achieve this, and the following are some of the terms that are used when discussing text analytics: Topic identification, concept mining, information extraction, and computational linguistics. All of these subjects refer to translating linguistically driven human communications into data points for analysis.
The software starts by answering some basic questions about the content. It looks at the type of document, for instance. Is it a publication like a peer-reviewed journal, or a newspaper article? Alternatively, is it one comment in a series of comments, in which the original post and the conversation before and after it have to be taken into account? What was the public sentiment at the time of the post? Did it follow positive or negative press coverage, which is influencing the author’s point of view?
The software is also tabulating “reach” and exposure as it goes. A comment may not rate as very important, but it will be counted as an exposure – that is, a person mentioned or exposed to a brand name.
The computer must apply linguistic tools to sort through context and to classify what it finds. Referential disambiguation is used to evaluate pronouns and to parse a huge variety of slang, colloquialisms, and synonyms.
Using all of the information it has gathered, the software now attempts to tabulate the inputs: are they positive? Negative? Neutral? Does it reflect a person’s actual experience of a product, or is it hearsay? This process is called semantic orientation, in which the software also attempts to determine whether the content is fact, conjecture, or opinion; whether it is opinion stated as fact, fact stated as opinion, or whether it is farce, or satire.
This kind of analysis is very difficult for computers, but easy for humans, so to improve accuracy, the software will assign a rating that indicates its level of confidence in its determinations. After computer analysis, humans review the items with a low confidence rating and correct or affirm the computer’s decisions. Feedback from this process is used to improve the analysis process.
Using social media analytics
In order for all of this to have any meaning, however, you need to bring one more thing to the table: goals. If you want to make sense of this mass of data gathered from social media, you need to understand why you are collecting and sorting through all this data. What business goals are you hoping to achieve? Are you trying to market a new product? How do people feel about what you already have on the market? Are you trying to identify purchasing trends, or provide better customer service? Do you need to improve public perception of your company?
Within the vocabulary and framework of social media, work with your marketing team to identify your key performance indicators. Is it the quantity of positive comments? Do you want to see specific phrases, or references to a product? Are you counting likes and shares?
Define your goals clearly, and employ good analytical software, and you will find a wealth of information about your company’s performance, and public perceptions when you mine social media opinions. Armed with this information, you will be better able to predict the market for your product, influence public opinion, and act quickly, ahead of the market, to respond to market needs and perceptions.
Gunjan Tripathi, digital marketing executive at Cheap SSL ShopLeave a comment on this article