This week, we do something a little bit different and provide our perspective on sentiment analysis. This includes, what sentiment analysis is, why it’s used, how it’s conducted and factors influencing the accuracy in sentiment analysis. Let’s begin!
What is sentiment analysis?
Sentiment analysis – also referred to as opinion mining – involves determining the opinion and attitude inherent in a post, comment, or piece of text, and then extracting the sentiment and emotion expressed. Whilst this could be done by humans reading the post and making a judgement, this is not practical and feasible. Instead automated tools are used which are proficient at discovering higher level trends and patterns, as well as extremities in opinion. Using a tool to do so also assists to keep costs low. This form of analysis started gaining traction in 2010, and has continued to grow in use and comprehensiveness.
Why it is used?
Sentiment analysis can be used to gain an insight into overall public opinion, with a better ability to assess their ‘virtual popularity’, as well as determine the opinion of the brand in comparison to competitors. This form of analysis simply creates another way for organisations to ‘listen’ to their customers and gain feedback about their products and offerings. For example, conducting this analysis on customer reviews will allow organisations to quickly detect and respond to negative reviews, improving customer trust and loyalty, as replying to the customer in a timely fashion means they are are likely to have a more positive perception of the brand. An organisation is also able to use it to assess / measure the success of a marketing campaign by analysing what people are saying about it on social media. An organisation can also then manage the campaign better once implemented – for example, Expedia, upon their sentiment analysis finding that consumers found the music in their advertisement irritating, responded in a playful and humorous manner, by broadcasting a version of the advertisement in which the offending violin was destroyed.
How is sentiment analysis conducted?
In the most fundamental sense, sentiment analysis examines the number of positive and negative keywords occur in a piece of text. If the number of positive words (e.g. great, fantastic) outweighs the number of negative keywords (e.g. terrible, poor), the text is considered positive content, and similarly, if there are more negative words, the text is deemed to be of a negative sentiment.
However, in many cases, sentiment analysis is more comprehensive and takes into consideration more factors than a count of keywords. Natural language processing (NLP) is the next step up, which attempts to process the language and text in terms of its human meaning – which involves understanding the various components of speech and how they tie in together (i.e. that several words constitute a phrase, a number of phrases compose a sentence, and sentences convey ideas).
More thorough sentiment analysis involves factoring in the location and demographics of the author, such as gender, age, and salary, as well as the context in which the post was written.
Factors impacting the accuracy of sentiment analysis
Language is complex – even when humans communicate with one another, miscommunication and misunderstandings happen, so imagine a machine trying to understand that, with all the sarcasm, jargon, and figures of speech we use! The accuracy and reliability of results depends on the accuracy of the underlying algorithm. The algorithms of most social media analysis vendors likely have a 50-60% accuracy rate. As Christine Day, a digital marketing specialist states, this level of accuracy is often not sufficient to base decisions off, which means the results of sentiment analysis should feed into, though not be the sole basis of, decision making. Even NLP, which is considered superior to simple keyword processing, still has several limitations, including difficulties detecting sarcasm, social media jargon, hyperbole, cultural variations, and misspellings. Subtle nuances which a human may be able to determine the difference of, are also not likely to be accurately interpreted in NLP – for example, the difference between “Thank you.” and “Thank you!!!!” in social media communication would not be detected by NLP.
So why is this? Take a look below at the following summary table, which details the 5 main factors behind the difficulties in gaining accurate sentiment analysis results.
However, despite the shortfalls of sentiment analysis, the insight that can be gleaned from large volumes of data (e.g. millions of Facebook comments) outweighs the reliability and accuracy of a sentiment rating at the granular level (e.g. one Facebook comment). The focus for companies, therefore, is on interpreting and actioning these insights.
What does the future hold for sentiment analysis?
Sentiment analysis is still relatively immature, although progress is being made, and algorithms will require continual refinement and testing. In the future, we can expect sentiment analysis to have a greater level of accuracy, and to better address gaps in meaning interpretation. Furthermore, human expression and emotion doesn’t only fit into 3 categories of positive, negative, or neutral, so we could potentially see the introduction of multi-dimensional scales measuring aspects, such as skepticism and excitement, or other human emotions, such as fear, surprise, and disgust.
Sentiment analysis is clearly a burgeoning area, allowing organisations to better understand their brand image, and their customers. As algorithms improve, we can see this trend continue, as more organisations tap into the possibilities that sentiment analysis enable.
As always, we’d love to hear your thoughts, so please drop a comment below.
Bannister, K., 2015, ‘Understanding Sentiment Analysis: What It Is & Why It’s Used’, Brandwatch, weblog, accessed 13 May 2016, <https://www.brandwatch.com/2015/01/understanding-sentiment-analysis/>
Day, C., 2015, ‘The Importance of Sentiment Analysis in Social Media Analysis’, Linkedin Pulse, accessed 13 May 2016, <https://www.linkedin.com/pulse/importance-sentiment-analysis-social-media-christine-day>
Donkor, B., 2014, ‘Sentiment Analysis: Why It’s Never 100% Accurate’, BRNRD.ME, weblog, accessed 13 May 2016, <http://brnrd.me/sentiment-analysis-never-accurate/>
Sela, R., 2014, ‘What You Need to Know About Social Media Sentiment Analysis’, Curatti, accessed 13 May 2016, <http://curatti.com/social-media-sentiment-analysis/>