What is Sentiment?
“Opinions” are key influencers of our behaviours. Our beliefs and perceptions of the reality are conditioned by how others see the world. Whenever we need to make decisions, we often seek “opinions” of others. In the past, individuals sought opinions from friends and family and organizations use surveys, focus groups, opinion polls, etc. With more than 700 million people using online social media, such as facebook, twitter etc., to communicate with each other across the globe, companies are looking this as an opportunity to reach people and do business. Since businesses rely largely on word of mouth marketing, the social media has now become e-wom (electronic word of mouth) marketing tool. Even the customers have become smart by constantly looking at product ratings, reviews, blogs, micro-blogging, before making purchase decisions. All of these social media technologies are changing the customer experience and are increasingly being used to connect with customers to build strong relationships and converting a regular customer into a brand advocate. Online social media and e-wom, have rapidly changed the e-commerce to a new face called social media business or social commerce or social media marketing.
The internet and the web have changed the way people communicate. People can now post their feelings or opinions on the web freely. They can write about product reviews, express their views about the services in any forums, company websites or e-mails, blogs or social sites like facebook. If one wants to shop for a new product, he or she can go to any forums or website to check the particular product reviews and make decision to buy or not. For a company, this is a new marketing challenge and also it may not need to conduct any surveys or feedback to find out customer satisfaction about its products and services, and how competitors are doing, as it can now be available for the companies instantly.
According to the recent survey conducted Local Consumer Review Survey (2012), “approximately 72% of consumers surveyed said that they trust online reviews as much as personal recommendations, while 52% said that positive online reviews make them more likely to use a local business.”
According to Nielsen, a global leader in measurement and information, “Thirty-six percent of global online consumers report trust in online video ads and 40 percent say they believe ads viewed in search engine results. Sponsored ads on social networking sites are deemed credible only by 36 percent of global respondents. However, in India, the numbers are higher with 48 percent online consumers trusting online video ads and 52 percent believing ads viewed in search engine results. Sponsored ads on social networking sites fare better with 54 percent of respondents trusting this form of advertising.”
What is Sentiment Analysis?
Opinion mining or sentiment analysis is the process of determining the sentiment or opinion of a given topic a document. Political parties may be interested to know whether people are supporting their program or not. Social organizations may want to find out people’s opinion on current debates. Cell Phone Company may be interested to know:
a. What users are saying when a product is launched
b. Which features are liked most
c. What features they do not like
d. Are they talking positive or negative
Finding opinion from the web sources can be a formidable task because of huge volume of data (text). It is difficult for a human reader to go over the 1000s of reviews to form an opinion. In many cases, opinions are hidden in a long discussion posts or blogs. Thus, it is essential to have an opinion discovery and summarization system which can do these formidable tasks automatically using Natural Language Processing and machine learning techniques.
In general, opinions can be expressed on products, services, individuals, organizations, an event or a subject. An opinion expressed consists of a target entity (an object) that has been commented on and its attributes (or properties). Each object can have a set of components and a set of attributes. Thus, an object can be hierarchically decomposed based on the relationship.
To understand better, let us look at the following review comment of Nikon 300s camera from bestbuy.com:
“I love this camera (1)!! No regrets whatsoever......highly recommended(2). I specially love the quality of the pictures(3)! I will be using this primarily for my photography business and I am truly satisfied with it (4). I also love the easy access to some of the most used features (mode, shutter, HD video and aperture) (5). I went from a Canon Rebel to a D300s(6)!! What an upgrade(7)! For ladies, I recommend Nikon D7000, it's lighter and similar to D300s, but with Half-plastic body(8). But I got a little cons for it, high ISO noise control, above 800 ISO, you can clearly distinguish the noise points on your shooting (9). Anti-dust system is effective but a little bit noisy(10). I continue to learn/discover something new about the features on this camera....Very happy with it(11)!! I would recommend this to a friend!(12)”
There are several opinions in this review. There are positive opinions (1,2,12) and negative opinions (9,10). There are opinion comparisons (6, 7), opinions regarding specific features (5,8). However, the overall review opinion is positive (12). The first two sentences (1, 2) expresses opinion on the camera as a whole. Then we notice that the opinions expressed in next sentences have some targets or objects on which the opinions are expressed. For example, in sentence 3 the opinion is on the quality of the picture. In sentence 4, the opinion is on the use of camera. Similarly, in sentence 5, the opinion is about specific feature – mode, shutter, HD video. In sentence 6, the opinion expressed is comparing Nikon 300s with Canon Rebel camera and Nikon 7000. The sentence 9 and 10 expresses negative aspects of the camera – noise and anti-dust system. In the last sentence, opinion is recommending to a friend. With this example in mind, we now formally define the sentiment analysis or opinion mining problem.