ImageVision by AppsTek
Text Analytics and Its Challenges

Text Analytics and Its Challenges

In today’s dynamic world, most of us might be familiar about ‘Text Processing and Analytics’.

Needless to say, text processing and analytics is one of the fastest growing areas in the field of machine learning as data arrives at an incredible speed. The Internet generates text-heavy data from blogs, websites, comments, social media etc. Computer software generates log messages and audit trails. Additional sources of data include emails and audio/video content that gets transcribed into text.

With such a huge volume of textual data available for free, businesses can capitalize on this data through text analytics. Corporates can plan and strategize using insights. However, the majority of the data available today is unstructured. Analyzing text brings forth several unique challenges. Text data is as large as numeric data. Secondly, text does not have a fixed structure or schema, which makes it more challenging.

I feel a major challenge with analyzing textual data from different sources such as social media, emails, chats, WebPages etc, is extracting insights from the unstructured data and to speed up the analysis. There may be too much of data to deal with.

Hence, the first and foremost technique we use to handle this unstructured data is to programmatically bin this data to tone down its volume and understand the broader spectrum effortlessly, also known as text clustering.

Text clustering algorithm leverages machine learning and natural language processing (NLP) to derive insights and cluster unstructured text.

Interested to know how it works?

Typically, the free-flow text i.e., descriptors (sets of words that describe topic matter) is extracted first from the document using the following steps:

1. Removing punctuations;

2. Transforming to lower case;

3. Grammatically tagging sentences and removing pre-identified stop phrases (Chunking);

4. Removing generic words of English language;

5. Stripping any excess white spaces;

6. Stemming.

We can leverage NLTK and Regular expressions to perform the above mentioned text preprocessing steps I.e., Once we load the data ,we can import re package to remove punctuation and lower() method to transform data to lower case.To remove generic words of english language ,extra spaces and stemming we will be using modules from NLTK corpus such as ‘stopwords’ and ‘stem.SnowballStemmer’ or ‘PorterStemmer’.

Once done, this free-flow text is analyzed by a text clustering algorithm using the five steps mentioned below:

· Transformations on raw stream of free flow text;

· Creation of Term Document Matrix;

· TF-IDF (Term Frequency – Inverse Document Frequency) Normalization;

· K-Means Clustering using Euclidean Distances;

· Auto-Tagging based on Cluster Centers.

TfidfVectorizer , the feature extractor module from Sklearn package will pull out unique words using IDF score which can eventually help in clustering data.There are lots of options to explore here to get different results, including CountVectorizer() method .Alternately, we can also use TermDocumentMatrix() method from textmining package to create document matrix.

Clustering being an unsupervised operation, KMeans requires specification on number of clusters. To find an optimal cluster for the data, one simple approach is to plot the SSE for a range of cluster sizes using ‘cluster. kmeans’ module from sklearn and look for the “elbow” where the SSE begins to level off.

To observe the trend you can print the top keywords based on their TFIDF score for each cluster.

What next?

As we have seen how to use text clustering method to convert large amounts of unlabeled data into bins wherein each bin holds unique characteristics or properties, to tag or assign labels (classes).

Now, these labeled dataset is used to solve a classification problem for common applications such as extracting sentiment, topic and Intent analysis on the text using machine learning or deep learning algorithms.

Sometimes, the quality of data is more useful than the powerful algorithm. With better data, even a simple algorithm can give great results.

Text Analytics is a path towards a great future and we see many enterprises taking this path to harness the power of insights and translate into strategic business activities to deliver better customer experience and build a sustainable relationship with their clients

I hope this article was useful for you to get a basic idea about Text Analytics.