Over the last few years, more and more digital research and analytics methodologies have emerged that leverage vast new data sets from online user-generated content, such as such as consumer reviews, social media, blogs, and forums. At the forefront of this growth exploiting User-generated Content (UGC) is an analytics tool used for understanding enormous data sets: Natural Language Processing.
Natural Language Processing (NLP) is a scientific discipline at the intersection of humans and computers. Simply put, it’s a way for computers to analyze and derive meaning from human language, helping us better understand ourselves and the world around us.
You can think of Natural Language Processing as a toolkit based on linguistics, computer science and AI, focused on understanding and generating human language. It helps computers understand, interpret and even mimic human language.
In market research, we use a form of NLP analytics called text analytics. Text Analytics is deriving high-quality information from text. An individual review, tweet or Facebook comment might not be terribly meaningful to brands, but when analyzed as a robust data set, it enables insights extraction for companies and brands. The sheer size of data of the Internet (100 trillion words) is impossibly large for a human audience to digest. Tools like NLP help us absorb and manage this data more efficiently.
NLP enable computers to understand natural language as humans do – no matter if the language is spoken or written. It uses artificial intelligence to absorb or take in language, process it and make sense of it. In the same way that our eyes and ears take in language and our brains process it, only here, the computer does the work – and can process so much more than we can.
There are two general phases to NLP: data preprocessing and algorithm development. The first, data preprocessing, involves preparing and cleaning the data. For market research that utilizes UGC, this requires the ability to scrape or extract large amounts of unstructured data in a timely fashion.
Data can be either structured or unstructured. Structured data, or data that usually resides in relational databases, are typically length-delineated data like phone numbers or zip codes, that are easily searchable. Most data sets that live online as UGC is unstructured and not as readily searchable. This may be audio, video, or social comments. Part of the data preprocessing phase is organizing this data so the algorithm can do the heavy-lifting of analyzing these vast data sets. In many cases, this means giving structure to the data and cleaning it.
A couple ways that this happens:
Tokenization is when text is broken down into smaller units to work with. Some UGC, like review data, is focused around specific products and has specific attributes that can be broken out into unique data points, such as: Star Rating; Date & Time; Platform, Brand, & Product; and Review Text and Title. Other data, like social media and blog data, is organized into group conversations around topics and include data points such as: User, Posts, Comments, Likes, and Date. A necessary task of applying NLP to the UGC is breaking up each post or review into these discrete parts, typically organized in a spreadsheet.
Stop word removal is when common words are removed from the text so unique words that offer most information about the text remain. In the case of review and social data, excluding conjunctions, prepositions, etc. – words such as “and,” “the,” and “of – helps to ensure only the important ideas float to the top of the analysis.
Lemmatization and stemming are when words are reduced to or grouped by their root forms so they can be analyzed as a single term. Sometimes this can be grouping all forms of the same verb to one term, such as “are,” “am” and “is” to be. This may also refer to grouping singular and plural forms of the same word together, such as “cars” and “car”. For reviews and social analysis, it can also refer to reducing words to their root form to understand prevalence of a word or idea, such as package, packaging, packaged.
After data preprocessing comes the algorithm development. There are many different NLP algorithms but the two main types include 1) Rules-based system, which uses carefully designed linguistic rules and 2) Machine-learning based system. Which uses statistical methods. Here the algorithms learn to perform tasks based on the training data they are fed, and adjust their methods as more data is processed.
Natural Language Processing is now employed in some way by nearly every person and every sector of modern life. If you’ve ever plugged an unfamiliar language into Babble or Google Translate, asked Siri or Alexa a question, used a search engine or relied on spell check – you’ve benefited from NLP. For brands and companies, the sheer amount of online UGC on their own products and their competitor brands has made NLP analysis an important part of their insights strategy and business planning.
Our next post in the series will delve into the evolution of NLP – where we were, where we are now, and exciting future applications of the NLP and Machine Learning.
Comments