Magazine article Online Searcher

Still Feeling Our Way toward Sentiment Analysis

Magazine article Online Searcher

Still Feeling Our Way toward Sentiment Analysis

Article excerpt

What is sentiment analysis and how will it change our searching behavior and our data analysis? More than a year ago, on July 28, 2014, Cynthia Murrel wrote on Steve Arnold's Beyond Search blog ("From Search to Sentiment"; arnoldit.com/wordpress/2014/07/28/fromsearch-to-sentiment) about an entity-level sentiment analysis patent ("Signal Processing Approach to SentimentAnalysis for Entities in Documents"; US 8725494) granted May 10, 2014, to Attivio, Inc.

Six months later, Murrel noticed a December 2014 alliance between Attensity and Moreover, a media monitoring company now owned by LexisNexis ("Attensity Partners With Moreover at LexisNexis; arnoldit.com/word press/2015/02/06/attensity-partners-with-moreover-at-lexis nexis). This deal gives Attensity direct access to Moreover's 54,000 news sources via the Moreover API that it can use in its sentiment analysis products aimed at helping companies discover trends and perceptions that will affect their brands (prnewswire.com/news-releases/attensity-announces-stra tegic-alliance-with-moreover-extends-lead-over-competitorsfor-most-comprehensive-sentiment-analytics-on-tbe-market- 300006595.html). The April 2015 launch of LexisNexis Newsdesk, based on the Moreover acquisition, incorporates sentiment analysis.

SENTIMENT ANALYSIS, TEXT ANALYSIS, AND NLP

Accurate analysis of the sentiment behind words, keeping them in context, continues to improve. Sentiment analysis goes beyond text analysis and natural language processing, which have been around for a while. It strives to identify the tone behind the words, the emotions involved, and the affective nature of the text. Technologies such as latent semantic indexing and sentic computing also contribute to advances in sentiment analysis.

Extracting meaning from unstructured text for further analysis is the holy grail for forward-thinking product/brand marketers, social scientists involved in the construction of surveys and interpretation of responses, and pollsters (who want to understand not only which political candidate will win your vote, but what the deciding factor was).

Text analysis can be performed at the sentence, document, or entity level. The analytic tools look at term frequency, position, and presence and often apply statistical methods such as Bayesian inference. Depending upon the purpose for the analysis, content-, concept-, or context-based approaches are chosen to extract and analyze the meaning behind the words.

Text analysis should not to be confused with text mining, although both use elements of computational linguistics. The latter automatically extracts information from a variety of sources, combining the extracted information to create new knowledge and understanding. Text mining results reveal previously unrecognized connections among the thoughts expressed in documents from multiple disciplines.

Natural language processing also differs from sentiment and text analysis. One example: Cerebrant looks at what's being contributed to large repositories such as PubMed and alerts users when something new has been made available. Cerebrant, a product of Content Analyst (contentanalyst. com), does not depend on Boolean search strings, taxonomies, or term lists. Instead, it identifies and ranks "the most conceptually related documents, articles, and terms" from whatever piece of text users give it. That text could range from a short phrase or sentence to an entire document. Content sets analyzed by Cerebrant could, according to the company, scale up to millions of text items.

How does Cerebrant do this? It leverages the power of Content Analyst's learning engine. Content Analyst employs latent semantic indexing (LSI), developed in the 1980s at Bell Laboratories, which is "a mathematical approach to text analytics designed to extract every contextual relation among every term in every text object within a collection."

SenticNet (sentic. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.