Sentiment analysis provides the tools which enable you to learn what your customers are saying about your product. This will help you build a better understanding of your customers.
“Sentiment analysis or opinion mining refers to the application of natural language processing, computational linguistics, and text analytics to identify and extract subjective information in source materials.” http:/
/ en.wikipedia.org/ wiki/ Sentiment_analysis
Eh?
Basically it’s good to know what people are saying about your company or your product.
So how do you gather this information?
In the old days, companies spent lots of cash doing consumer research by surveying people on the street, in focus groups and via the telephone. Today, companies can ask people to complete online surveys when the visit their website or contact them via email to follow up an online purchase. Despite the advancement in technology, the questions remain the same.
What do you think of the service? How easy was the transaction? Would you recommend the product? And so on.
However, people lie in surveys. Think about it. How truthful are you when you complete a survey? Do you always tick the right age box? What about your financials, do you pretend to have a £10k or a £100k a year job. I mean what harm does it actually do?
The point I’m making is that the research is only so good. It’s like stats.
63% of people who visited this blog ‘liked’ the page on facebook.
Over what time period?
The previous stat doesn’t reveal that! Stats can be used to hide a lot of relevant information.
So what are the options?
Customer surveys are only part of tracking feedback. We live in a culture which publishes blogs, reviews, status updates and emotions online and in public. This is good as it allows companies to mine that data for references that are relevant to them. Sentiment analysis allows you to use that information to find out what people are saying about your company / brand / products / staff.
How to start.
Define your overarching objective and be realistic.
Why do you want to know what your customers are saying?
Is it going to affect your price point, your marketing strategy, what products / services you provide?
Be honest.
What’s your budget?
Do you have the resources to carry out customer surveys?
Ask your customers a direct question. If you are on social networking sites, ask your followers for feedback. You can use the front page of your website to promote a customer feedback survey or just ask people to complete a comments section.
Let’s talk about conducting online research
You could go to google and enter a search term and collate the results, but how useful would that be?
Sentiment analysis is all about getting real data together that you can then use to shape your future strategy, policy, or product line.
Before discussing some of the tools that can help you conduct your research I want to explain how you will score content so that it is actually useful.
Let’s use the example that you’re trying to find out what people think of the “iPhone”.
Example content from a review site:
“I bought an iPhone 3GS. Well I say bought, but it was free with a contract. The contract is expensive but the phone will be worth it. I hope! After only a few hours use I can see that the battery wont last all day, but I don’t mind as I’ve already downloaded lots of cool apps. I love Apple products.”
Example content from twitter:
“One thing I can’t live without? My iPhone 4.”
Example content from facebook:
“Woohoo!! Just got a white iPhone. I’m now one of the cool kids 😉 Anyone know any good apps or have tips on how I can really make use of it? I can’t figure out how to get my music onto it either, any help would be great. Thanks.”
Example review from Apple.com:
“The iPhone 4 is the best phone ever, not just by design but it also has the best apps. I use it everyday.”
Example review from Play.com
“The HTC Sensation is way better than the iPhone. While people claim the iPhone has better apps, that is no longer the case. The HTC has a far better battery than the iPhone and while the iPhone has a great screen, it’s smaller than the sensation as it therefore not as good”.
From the five examples above the iPhone would seem to get favourable reviews. However, let’s take a closer look.
To quantify the data you need to set some parameters.
Who is writing these reviews? Let’s assume that they are written by ordinary people.
Do they seem genuine? I believe that the reviews above are genuine.
What are they talking about? The iPhone of course!
Really?
Which model?
This is the first obstacle you will encounter. From the outset the search term was too broad and produced results that weren’t specific enough. Each of the examples could be talking about a different iPhone. Not one mention the size of the included memory.
Keep thinking about your overall goal. If your objective is just to research the brand “iPhone” then the examples above can still be used. If it’s to assess the iPhone 4 then the first review can be discounted.
Once you are sure that the data you have collated is valid i.e. refers specifically to the task at hand, you should then try and score each statement. You can do this by asking the following questions:
Is the statement positive?
Is the statement negative?
Is the statement neutral?
You will need to develop a scale which you can then score the statement against e.g. a positive statement receives a score of +2, neutral 0 and negative -2.
It’s not always easy to judge whether a statement is positive or not. There are additional factors that will need to be considered.
- What are the emotional components of the sentence and how do these influence the classification e.g. anger, sadness or happiness?
- We would need to how much influence that statement could have e.g. is it a tweet to 20 people or an article on Amazon?
- Is the facebook page private?
- Is the statement opinion or fact?
- Is the statement provided by the owner or is the statement a quote by another?
In the examples above I have added the bold font and red colour, but what if the original author used different fonts to create emphasis on their words?
The intended message could have a different meaning with certain words written in bold. It is important that you factor this into you calculation. Consider why a person has taken the time to highlight a positive (or negative) feature.
Where does the author mention the pros and cons of the product within the review? These positional features indicate the strength of the piece. A review that starts on a negative tone will most likely be negative overall. People tend to lead with their strongest emotion.
Here is a quick guide on what to do:
1. Determine objective – “I bought an iPhone”.
2. Determine document subjectivity – is it a factual statement or opinion?
3. Determine document orientation – is the statement positive, negative or neutral?
4. Determine the strength of the orientation – i.e. weakly positive, mildly positive or strongly positive.
5. Determine the sentiment – what emotional components are in the statement i.e. it’s a nice phone.
6. When was the statement written? This can help deduce what product model the review refers to.
If you follow these six steps you will have a good understanding of what the statement says about your company or product.
What if an article contains both positive and negative phrases? How can that be evaluated? Is there a weighting formula?
Break up the statement into scoring chunks.
Weight the statement by keyword, emphasis (e.g. bold type), where it’s published, small following, how influential, private / public etc.
Things to remember: how many times is the keyword mentioned? Is there a lot of emphasis? Does the person have a small social media following? Is it a popular website? Is the post public or private?
1 = keyword
2 = positive orientation
3 = negative orientation
4 = level of influence
5 = negative orientation
This is just an example to give you an idea of how useful plotting your data on a chart or a graph will be in determining the overall sentiment associated with just one phrase.
You should look at each resource individually i.e. score updates on twitter, then score updates on facebook, then product review sites, then blogs etc. Once this is completed you will have total scores for each network that you can then plot on another graph which will give you an overall snapshot of opinion.You may decide to weight each network differently e.g. if you sell on Amazon, an Amazon review is going to be more influential than a blog post.
A simple formula that could help you with this process is:
Ci = {C1, C-3, C4, C0}
and
D = {Ci, Cii, Ciii, Civ, Cv}
C can be used to represent a classification e.g. a keyword. So if your company was trying to assess sentiment against a range of products each product would be identified by a different keyword and hence would be represented by Ci, Cii, Ciii etc.
Ci is the sum of the figures within the series {C1, C-3, C4, C0}.
C can also represent different places you have researched e.g. Ci = twitter, Cii = facebook, Ciii = Amazon etc.
You just need to make sure that you understand what you need C to represent and then run with it.
D will represent your companies / products overall score and will provide a representative sentiment analysis.
D is the sum of the figures within the series {Ci, Cii, Ciii, Civ, Cv}
If you find that you are swamped by data you can try using an automated service such as Twitter Sentiment(screenshot above) which is “a Twitter sentiment analysis tool. Research the positive and negative opinions about a product or brand.”
Tools
There are many tools that you can use to track sentiment online, but you can start with google alerts, twitter and facebook searches.
With google alerts you can establish a search query specifically relevant to you and have google email that to you each day. Google will search websites, blogs, news sites etc and email you the results, thus saving you from having to repeat the search on a daily basis.
Twitter allows you to save searches and to track #hashtags, but I’d recommend using a tool like Seesmic or tweetdeck to view these searches. These will allow you to see every time you have been @ mentioned or how many times someone has tweeted about your brand name or product. You can have many ongoing searches making it easy to monitor on an ongoing basis.
Seesmic and tweetdeck will also monitor your facebook pages and notify you when someone leaves a comment against one of your posts.
Bing has agreement with facebook that gives them access to facebook profiles, so head to http:/
Blekko is another site that searches facebook e.g. http:/
so don’t limit your tools to just a few applications.
One trap that you definitely do not want to fall into is spending all day everyday searching social media sites. If you set these tools up correctly, you should only be checking in a couple of times per day. Respond where necessary and record sentiment when it comes up. Only analyse the data when you can set aside the appropriate amount of time. So your work flow could be that you check your data once per day for 10 minutes and you spend another 10 minutes capturing feedback. Compile that information in whichever way suits you e.g. copy and paste into Word or onto a spreadsheet. At the end of the month you can then spend a few hours going through the data with the aim of producing a sentiment analysis which you will then use to review your existing products or services.
Conclusion
With sentiment analysis it is easy to get carried away and spend too much time focusing on finding and rating content that describes your product, which can take your focus away from developing great products or services. You definitely need to find the correct balance between conducting the research and actually carrying out your business activity.
However, the importance of sentiment analysis cannot be stressed enough. Even a little research into what people think about your products can help your business overall. If you are deaf to customer complaints your business will start to get into trouble.
Feedback! I’d like your thoughts on this article.
Do you think it is wrong, factually incorrect, glosses over important topics?
What has been your experience with sentiment analysis?
Is your company doing it?
Do you have any tips that you’d like to share?
Thanks for reading.
#datamining #digitalmarketing #opinionmining #sentimentanalysis #social
