Sentiment analysis techniques and applications across diverse fields

Sentiment

There are as many sentiment analysis techniques as there are reasons for conducting sentiment analysis. Analysis techniques are employed to discern sentence, phrase, word and text meanings, and predictive, machine-related, emotional and psychological aspects are measured by sentiment analysis as well. This literature review will attempt to navigate the various avenues presented by such diverse usage of sentiment analysis and provide information that categorizes and differentiates between the various techniques employed.

The review will focus on a number of different analytical techniques, including sentiment analysis, predictive analysis, text, phrase, sentence and word analytics, and why and how the different arenas approach the use of their specific style of analysis.

Sentiment analysis has been described as using a lexicon "with information about which words and phrases are positive and which are negative" (Wilson, Wiebe, Hoffman, 2009, p. 400) and as an analytical tool that "focuses on identifying positive and negative opinions, emotions and evaluations expressed in natural language" (Wilson, et al., 2009. P. 399). Keeping that description in mind, sentiment analysis can take place using a variety of methods to evaluate diverse areas of interest, including business, medical and education fields. Most analysis takes place using a step-by-step process no matter in which area of study the evaluation is taking place.

In the Wilson study the next step was to tag the lexicon clues into one of three polarities; positive, negative or neutral. Wilson categorized the words in order to determine at what level positive, negative or both were used in the various texts, phrases and sentences.

A word in the lexicon is deemed positive if (out of context) the word evokes a positive feeling, deemed negative if it evokes negativity, and is tagged as both if it is both positive and negative, and neutral if it deems no positivity or negativity at all. However, as Wiebe states in a 1994 report, sentiment analysis is a type of subjectivity analysis (Wiebe, 1994).

A certain amount of subjectivity is going to be found in almost every study, but it would seem that analyzing sentiment would be especially difficult in a country where the ways that the English language is spoken is as diverse as the people who speak it. One of the problems of sentiment analysis concerns the manner in which people speak, as well as verbal and non-verbal communications. One expert states "Humans communicate with each other in ways such as languages, facial expressions or gestures that possibly cover their sentiments" (Yang, Lin, Chen, 2008, p. 331). Sentiment analysis is subjective then because of the words used, but also the context in which they are used, as well as how the participant uses them. The question could be asked, if it is so subjective, then why do it? The answer might come from Yang et al. study that discovered "the analysis of sentiments helps us identify the latent meanings associated with communication and thus enhances opinion analysis, as well as mining individuals sentiments towards specific subjects or events" (Yang et al., p. 332).

If that answer rings true, then sentiment analysis can work well for a variety of purposes. Some of those purposes include such things as gauging customer (or potential customer) responses to products or services, Current literature provides a number of examples of such uses, including 2009 study by Nan et al.

Their study determined that "sentiment analysis helps understand and predict the behavior of the customer through analysis of the texts in blogs, forums and survey responses" (Nan, Xun, Xinli, Chao, Desheng, 2009, p. 227) and more specifically the study found that sentiment analysis is employed to probe into the correlations between information sentiment and asset price volatility (.p228). In other words, some researchers even use sentiment analysis to make money on the stock market. Since the purchase or sale of stocks and bonds is often based on emotion rather than business acumen, it actually makes quite good sense to employ sentiment analysis in such a manner.

Most experts, however, use sentiment analysis for much more banal endeavors. A number of textual studies called sentiment analysis the "text mining technique employed by a number of researchers to determine the attitude of a speaker or a writer with respect to a specific topic" (Ahmad & Almas 2005; Chaovalit & Zhou, 2005; Turney 2001, Turney 2002, Turney and Littman, 2003). Using sentiment analysis in this manner is what is considered the 'normal' approach of sentiment analysis, although it is true that studies are used every day to determine whether business customers are satisfied with services and products specifically regarding the language they use when talking to customer service representatives, and perhaps more importantly the language the customers hear from those same reps (Kho, 2010, p.11).

One other interesting avenue taken by a recent study implementing sentiment analytics was a study that looked at two prominent characterizations of morality; what is good and what is right is in direct correlation with happiness.

The claim was made by the study that it is good and right to do what our moral sentiments tell us to do, and the claim that it is good and right to do what would maximize the amount of happiness in the universe (Fehige, Frank, 2010, p. 141). Here was a study that sought to determine morality with the use of sentiment analysis. What the study determined was that there was one common way of drawing distinctions of a moral sentiment as a motivational force, or in other words; "a person feels good or bad about something in a certain kind of way, and this comprises or causes a tendency to bring about or continue that thing in the positive case, or to prevent it or stop it in the negative case" (p. 141).

Sentiment analytics are also used in predictive analysis. According to one expert some of the questions that predictive analytics attempts to answer include "which of my customers are likely to churn? What kinds of offers will persuade my customers to stay or new customers to buy? Price? Service options? Which potential customers are likely to be highly profitable? Which are likely to cost me money? What should I show this surfer when I serve the next page? (Monash, 2005, p. 36). Specifically, businesses are looking to analyze what word, phrase, sentence or text is going to be successful in drawing in customers, keeping them buying the goods or service, what pitch will work best with which client or customer and the language that will be most profitable for company use.

Siddharth Taparia, a senior director for software developer SAP, recently stated to Industry Week that "predictive analytics is becoming an advanced business intelligrence" (Alpern, 2010, p. 48).

In fact, Best Buy recently incorporated predictive analytics regarding the behavior of its customers regarding purchasing patterns and why and what customers bought as compared to what was advertised. According to Alpern, Best Buy's first step was to "study customers buying patterns to discern predictive patterns of purchasing (p. 48).

As shown by the literature, sentiment analysis can be used in a variety of areas, from purchasing, to morality to deciding exactly what an author or writer meant by the use of a certain phrase, text, sentence, or word.

Text Analytics

When referring to sentiment analysis, many experts are referring to analyzing the meaning of words in certain texts, sentences, phrases or contexts. There are a number of levels where sentiment analysis regularly takes place. One of the most common areas is in text (referring to passages, articles, books, journals and even text messages). One of the most comprehensive descriptions of text analytics comes from Grimes when he states that text analytics applies "statistical and/or linguistic and/or structural techniques to identify, tag, and extract entities, concepts, relationships, and events (features) within document sets" ( Grimes, 2008, p. 11). Furthermore, he defines text analytics even further by describing the process as a way to apply statistical pattern-matching and similarity techniques in order to classify certain documents or text along organized features according to specific categorized taxonomy" (Grimes, 2008). In fact, Grimes was also part of a study in 2009 that provided further classification and description by providing a process for using text analytics. He defined text analytics as the use of software to simply the process of annotating the information and then analyzing those annotations and other extracted information from a textual source or sources. The reason behind this process was to extract from the test, "entities, concepts, topics, facts and attitudes" (Grimes, 2009).

Some of the reasons behind the analyzing of text is not only to determine what the author meant upon writing the particular text, but other insight can be gained as well. The author is not the only person of interest, the person affected by the text is also affected. The question could be asked as to why such interest would be shown regarding what individuals think upon reading certain text, but examining the question from a business person's viewpoint; if a company could determine how a consumer was going to respond (negatively or positively) from a certain ad, article or promotion via the text that was being used, the information would likely be considered pretty valuable. Lange wrote that "studies show that consumers are much more trusting of other user's opinions than of the marketing produced by the manufacturer" (Lange, 2010, ) and the vast majority of text that those customers see, according to Lange is unstructured text that has keywords that evoke positive images. Promoting those positive images would be a good thing for company's who wish to enhance their bottom line(s). According to Lange, from the company perspective listening and analyzing what people are saying, reading or thinking about your products and services is an important step in engaging with your audience via dialogue (Lange, 2010).

Another study said that analyzing the text plays a crucial role in many of the important decisions that a company has to make on a regular basis. The study determined that "information hidden or stored in unstructured data can play a critical role in making decisions, understanding and conducting other business functions" (Prasad & Ramakrishna, 2010, p. 2201).

According to the current literature, it would seem that many of the uses for analytic techniques would have to do with profitability, improving products, or (at the very least) enhancing the marketing that companies undertake in order to promote their products or services. Text analysis can also take the form of providing historical context and enlightenment into historical documents, including what some of the great philosophers meant when they wrote the words they wrote. Accordingly, text mining is used as the technique to determine those details, facts and figures.

Prasad and Ramakrishna state that data mining from textual sources allows for knowledge discoveries from text(s), but that what makes text analytics different is that it looks to determine or provide knowledge of a structure that is inherent in the text itself. To them 'text analytics is the answer to overcome unstructured data' (Prasad & Ramakrishna, 2010). Again according to Prasad et al. The textual sources should help to overcome that lack of structure, but it should also allow for a statistical analysis of the extracted data to determine concepts and patterns that can then be categorized or classified. Text analysis does not always mean the written word itself, it can also be applied to video, audio, images, websites, and (of course) documents. An example of categorization might be when a media such as websites are data mined, the researcher is often looking to analyze not only the text but the identifying terms, numerical expressions and even concepts. Text analysis can also include text wrapping and annotations that are analyzed as freeform, however that type of analysis is normally corresponsive with domain-specific entities be identified as text wrapping (Surehka & Sreekumar, 2005).

With the different methods for analyzing texts and the varied arenas in which it is employed, recalling that text analytics started out in the educational field grounds the theory in a manner that it might not otherwise be grounded. Theory is fine, but having the theory battered by the educational process certainly lends a more certain air to the entire process. As one expert succinctly stated "a carefully chosen, word, phrase, or sentence could unlock the door to a constellation of meanings" (Welsh, 2010, p. 28).

Another expert in the educational field Nel Noddings described the process of how an analytic philosopher might, "for example, analyze the concept of teaching or of education…and considerable attention would be given to the various linguistic contexts in which the concept appears" (Noddings, 2007, p. 43). Text analysis in the education environment makes just as much sense as using text analytics in business or government.

Both business and education benefit from the textual analytical process. School and business administrators can make decisions with the organized data gleaned from text analytics. One author wrote that the implementation of text analytics if it is executed in the right manner assists executives and administrators by offering an effective approach to make current and ongoing disposition decisions confidently (Santangelo, 2009).

With the advent of the internet and other technological advances using analytics makes sense regarding all kinds of text, mainly because such technology allows for the solving of a variety of problems regarding the administration of data gleaned through that technology. Santangelo provided his take concerning data management when he wrote "it is stated that in solving the problem of large, unmanaged data repositories that increase an organization's cost and legal exposure, the need for the development of text analytics search technology along with the information management strategy is critical" (Santangelo, p. 23).

As the current literature shows, text analytics can be quite helpful in a variety of ways and in diverse areas of study such as business, technology and education. Further breakdown of text analytics can occur with phrase, sentence and word analytics, which are discussed further in this paper.

Sentence Analytics

The current literature does not contain a mountain of information or studies based on analyzing sentences. That could be due to the fact that many times sentence analytics are grouped under the same heading with phrase analytics. However there are a couple of items of interest to be discovered when researching the literature for sentence analysis. One is that most analytics that take place at the sentence level breaks down each and every word of the sentence into various categories for further study.

Those categories include noun phrases/clauses, adjectival phrases/clauses, verb phrases/clauses, determiners, premodifiers and the head. Knowing how sentences are composed and how they can be broken down allows for a much closer scrutiny than the 'normal' English class might offer. Such analysis is designed to provide a much deeper insight into the use of words and phrases. Analyzing sentences or phrases on this level necessarily provides for a better comprehension of what the author meant and how his or her concept or idea is developed within the context, and perhaps more importantly, outside of the context.

Word Analytics

Specific words have specific meanings, yet can be ambiguous at the same time. Specific authors use specific words to portray their thought(s) with the exact level of clarity required for the specific scenario. Understanding the words chosen provides a deeper insight into the mind of the author. Word analysis can also help in other ways, including how to teach or correct speech in young children.

A recent study determined that analyzing the English words used by children "showed that words that fell within any frequently occurring frame consistently belonged to the same grammatical category (eg noun, verb, adjective, etc.) (Chemla, Mintz, Bernal, Christophe, 2009, p. 396) and that categorizing these words using a cross-linguistic analysis revealed distinct distributional acquisition strategies. In other words, by studying the words that children used to communicate, the teaching of language acquisition skills could be improved upon, at least according to the study conducted by Chemla et al.

Another area of study using word analysis is how individuals communicate electronically and the correlations that can be discovered regarding information sentiment online. Gauging how people communicate through electronic means is an interesting field of endeavor. One recent study of online access using keywords and what the sentiments the keywords represented provided interesting correlations between information sentiment and asset price volatility. The study observed that the overall sentiment for the who article is acquired by combining all the sentiment values of the keywords used. Whenever a piece of news concerning the particular asset was released online the study observed word reactions from online sources and how those keywords played a role in how volatile the asset price was. The study used a process where "each piece of news is decomposed and converted into a keyword array, with the same sequence as in the article, each of which is assigned a specific sentiment value" (Li., Liang, Li, Deshang, Wang, 2009).

Having the capability to make money off the words used by others is an interesting use of word analytics, however, such word analysis of keywords used online could lead to a troubling concern; that of government intervention. With the capability to measure online word use, it is conceivable that other forces in the world can watch for keywords as well. With social networks and blogs becoming the medium of choice, typing in keywords or phrases online could bring much more trouble than what might be suspected. One study determined that social networking services (SNS's) and blogs are becoming more and more popular with large SNS's having more than 100 million users and that the DCNL automatically checks texts uploaded to social networking services or blog pages, detects the words that are used that might reveal private information and warns the user about them (Katakoka, Watanabe, Mizutanki, Yoshiura, 2010, p.1). A situation where the analysis of words (positive or negative does not matter) could lead to additional restraints or punishments based on what words or phrases are used. Freedom of speech would not then be as free.

For those individuals interested in more than just making a profit off the words of others, the moral equivalent of word analysis often takes place using scriptures as their basis. In the Bible it states In the Beginning was the Word, and that could be said to be true in analytics as well. If one wishes to discern the deepest thoughts and deeds of mankind, one must certainly look in books, text, writings and ultimately at the written word. As one piece of current literature espoused, "the human spirit would always be mysterious and that we can never really 'know' another person…perhaps analytic work may simply alleviate a degree of human suffering" (Proner, 2006, p. 435).

Though this paper does not espouse an alleviation of human suffering based on sentiment analysis, perhaps it does provide additional fodder for mankind to digest regarding what words, sentences, phrases and text is used as a form of communication. Knowing what is positive, what is negative and the uses of both in the various forms of communication now available to society could lead to improved ideas, thoughts and the process that is used to communicate with one another. At the very least the current literature provides evidence that such an ideal is not a fallacy.