Anthill Inside 2019

Sentiment Analysis from text is a well known problem in machine learning where a given text document can be either positive, negative or neutral. In the last few years, Sentiment Analysis has become a hot-trend topic of scientific and market research in the field of Natural Language Processing (NLP) and Machine Learning.

Existing approaches to sentiment analysis can be grouped into three main categories: knowledge-based techniques, statistical methods, and hybrid approaches. Knowledge-based techniques classify text by affect categories based on the presence of unambiguous affect words such as happy, sad, afraid, and bored. Statistical methods leverage elements from machine learning such as latent semantic analysis, bag of wordd, Word2Vec, Doc2Vec, Pointwise Mutual Information for Semantic Orientation, and deep learning. To mine the opinion in context and get the feature about which the speaker has opined, the grammatical relationships of words are used. Grammatical dependency relations are obtained by deep parsing of the text. Hybrid approaches leverage both machine learning and elements from knowledge representation such as ontologies and semantic networks in order to detect semantics that are expressed in a subtle manner.

Sentiment analasis on emails is a bit complex task and only few data email data sets are available publically: https://www.kaggle.com/wcukierski/enron-email-dataset

We show one industrial use case of sentiment analysis on emails data which we are implementing in Freshworks and it will be used by Freshworks clients. We are analysing email conversations between sales agents and their customers to predict the underlying deal outcomes. We are also using this data to identify high potential customers/targets by extracting sentiments from emails. The key challenge in this approach is that corporate emails are generally written in politically correct (and therefore logically complex) manners. Extraction of correct sentiment from these emails is therefore not easy. We use contextualised word embedding to capture different contexts. We further employ a method to upscale important words and downscale less important words present in the corpus. Eventually, we train different classifiers to extract the sentiment from these emails. We compare these classifiers on the basis of well-established accuracy measures such as precision, recall, accuracy etc.

Outline

In the conference, we will talk about:

1. Overview of the Problem Statements and challenges
2. Value Proposition from this problem
3. Overview of contextualised word embedding (BERT, ELMo etc). We will also learn about how they can be applied to our data.
4. We also talk about a method which upscales important words and downscales less important words in the corpus
5. We show an implementation of our framework
6. we will finally talk about hyper-parameter tuning and accuracy measures, followed by a comparison of different algorithms utilized in our work.

Requirements

NA

Speaker bio

Rahul Sharma is a Data Scientist with 7 years of industry experience in applying data science and advanced machine learning techniques to diverse sectors including telecom, healthcare and equity research for quantitative hedge funds. Rahul has completed more than 20 large-scale machine learning projects using structured and unstructured data sets, including data sets with 100s of TB size. Rahul holds a masters’ degree in computer science from Indian Insititute of Technology (IIT) Kharagpur, India.

PLease note that this session will have 2 speakers - Rahul and Swaminathan.

Links

• LinkedIn profile - www.linkedin.com/in/rahul-s-b1863661
• Freshworks - https://www.freshworks.com/

Slides

https://www.slideshare.net/secret/1QhibJF61s4kBD